Understanding Sponsored Search Core Elements of Keyword Advertising

1. 1Name the asymmetry first. Write one sentence per product: 'The advertiser knows X; the searcher cannot check X before clicking.' Searchers search to even out this imbalance, so every ad element must answer one of these gaps.
2. 2Rate how hard the product is to assess before purchase. The harder it is, the more the searcher relies on costly signals to form expectations. High difficulty means you need stronger evidence in the ad. Low difficulty means a plain statement can be enough.
3. 3Locate the searcher in the process. Information asymmetry drives searching especially in the early stages, so queries at that point are attempts to close an information gap. Match the ad to the gap the query exposes, not to your preferred pitch.
4. 4List every signal in the ad (title, description, URL, and the landing page it points to). Score each by cost to produce. Signals that are easy and cheap to fake are generally deemed less trustworthy, so don't rely on them alone.
5. 5Replace weak signals with informing content: facts, empirical evidence, and concrete information. The communication axiom you are addressing says the relationship is complementary (unequal), and the searcher is naturally wary. Your ad has to address that wariness directly.
6. 6Treat ad position as a signal in its own right. Where the ad appears on the SERP tells the searcher something. Check the rank-level click and conversion data before you assume that a lower position only costs you volume.
7. 7Design for seconds. The searcher must notice the ad, let its choices rise above the noise, and use information scent to select, all within response time measured in seconds. Put the strongest asymmetry-reducing signal where it is seen first.
8. 8Keep the signal chain unbroken to the landing page. The ad is a guide post to the landing page, so the page must deliver what the signal promised. Roughly 20 percent of users take under a minute to judge it, so confirm the promised information is visible at once.

Take Table 4.3 from the source, which shows how ad rank changes conversion potential. Rank 1 is the baseline at 100 percent click potential, 100 percent conversion rate, and 100 percent conversion potential. At rank 2, click potential falls to 59.8 percent and conversion rate to 91.1 percent, which gives 54.5 percent conversion potential, a 46 point drop. At rank 3 it is 47.5 percent and 75.1 percent, which gives 35.7 percent, a 34 point drop. Reading this through asymmetry: position is itself a signal, and searchers who see a lower-ranked ad convert at lower rates in most positions. Rank 10 is the exception. Its conversion rate rises to 87.7 percent, lifting conversion potential 10 points to 12.2 percent, though click potential is only 13.9 percent. The practical reading is that a searcher with an information gap does not choose by position alone. Your ad has to carry signals credible enough to hold up at the position you can afford. Consider a searcher buying an airline ticket (the source's own scenario), who may also be thinking of a hotel. An ad that supplies concrete facts about the offer closes the gap for the ticket, and a related ad can pick up the cross-sell. Rewrite the ad for the lowest position you can afford so it carries costlier-to-fake signals, then compare conversion rate, not just clicks.

• If the product is hard to evaluate before purchase, then put costly, hard-to-fake signals (facts, empirical evidence) in the ad. Cheap claims will not move a wary searcher.
• If a signal is easy to fake, then treat it as weak. Support it with another signal, or cut it.
• If the query reads as early-stage, information-gap searching, then lead with informing content. Defer persuasion and the call to action.
• If you are tempted to cut position to save cost, then check conversion rate by rank first. Position carries signal value, and conversion potential can drop by double digits per rank.
• If the ad's signal is not matched on the landing page, then fix the page before raising the bid. The ad is only a guide post.

• Asserting rather than informing: the ad is full of superlatives and no facts. The tell is a decent click-through rate with a weak conversion rate, because wary searchers click but do not trust the page.
• Ignoring position as a signal. Teams read rank only as volume. The tell is bid cuts justified by cost per click while conversion potential falls sharply at the lower position.
• Using cheap-to-fake signals for hard-to-assess products. The tell is the same generic claims that competitors can copy, with nothing that is costly to produce.
• Assuming searchers have one goal. The source says people rarely do, so an airline-ticket searcher may also be thinking about hotels. The tell is a single-purpose ad with no related ads that catch the adjacent need.
• Overlooking response-time limits: the signal is buried and the searcher decides in seconds. The tell is strong landing-page content that never gets visits because the ad's key signal is not visible.

• ▸Product/service: name the thing being advertised.
• ▸Searcher's information gap: what the advertiser knows that the searcher cannot check before clicking.
• ▸Pre-purchase assessment difficulty (low/medium/high): the higher it is, the more costly your signals must be.
• ▸Search stage and query: the query and whether it reads as early-stage gap-closing.
• ▸Signals in the ad: list each element (title, description, URL, claim) that tells the searcher something.
• ▸Cost-to-fake rating per signal: mark each easy or hard to fake. Easy ones are weak.
• ▸Facts/evidence to add: concrete, verifiable information that replaces weak signals.
• ▸Rank and conversion data: current position, click potential, conversion rate, conversion potential.
• ▸Landing page match: confirm the page shows the promised information within the first view.

1. 1Separate the three judgments the construct governs: (1) which engine the searcher picks, (2) how the searcher reads the whole results page, (3) how the searcher reads each individual link or ad. Brand image is a condition on all three, so audit each one separately. The source passages name the construct but give no scoring model. Treat the steps below as a working procedure built from the definition, not a published method.
2. 2Define your target searcher in one line, for example 'high-end framing buyer outside our local area'. Write down which engines that searcher plausibly uses. Mark these as assumptions to test, because the passages supply no engine-share data.
3. 3For each engine on the list, write what a searcher would plausibly believe about its trustworthiness and about how much to trust its paid placements versus its unpaid results. Label each belief 'hypothesis', not fact.
4. 4Walk a real SERP for your core keyphrase on each engine. Record where your ad sits, which competitors surround it, and whether the page looks like a trusted source or a sales page. That layout is the evidence for the SERP-level judgment.
5. 5Judge your ad as a link inside that engine's context. Ask whether the copy and the landing page would hold up if the searcher distrusted the engine, and whether they would hold up if the searcher trusted it. A few lines of text carry the whole claim, so keep the claims verifiable.
6. 6Align your own brand image with the engine's. Your business must read as credible within the environment where the searcher meets it. Write the image you want in the online marketplace (the passages pose this as the Branding Component question), then check that each ad and landing page matches it.
7. 7Split reporting by engine so you can see whether results differ. Treat a difference as a sign that engine brand image may be a factor. Confirm it with a test before you reallocate budget.
8. 8Re-audit on a schedule. The passages show that context shifts: sponsored search failed in 1996 and succeeded in 1998 as the Web context changed. Searcher trust in an engine is likewise a condition that can change, not a constant.

Take the book's framing shop. It has a Web site, competes against national chains, local franchises and mom-and-pop stores, and wants the national high-end market. Step 1: you define the target as a buyer of high-end framing outside the local area. Step 2: you list two candidate engines, Engine A and Engine B, and mark your beliefs about each as hypotheses. Step 3: you pull the SERP for the keyphrase on both and note that on Engine B your ad sits beside two national competitors, which signals a crowded, comparison-minded page. Step 4: you rewrite the ad's few lines so each claim can be checked on the landing page (the framing service and the shipping and receiving service), because a searcher who is wary of paid results will test claims. Step 5: you track clicks and purchases by engine. The history in the passages supports the logic of this check. OpenText's paid-placement attempt drew outcries, while GoTo.com's 1998 model, with more than 1,000 advertisers paying one cent to one dollar a click, drew few. The same mechanism succeeded once the surrounding context and searcher perception had shifted. The framing shop's results are illustrative, so no numbers are given.

• If the same ad and keyphrase perform differently across engines, then suspect engine-level trust or perception before rewriting the ad, and test it.
• If your claims in the ad cannot be verified in one click on the landing page, then tighten the copy, because distrust of the engine or of paid results is paid for at the link level.
• If your brand image is strong with the potential customer base, then lean on brand cues in the ad, since the passages note that a good brand image makes marketing easier.
• If you cannot support a belief about an engine's brand image with evidence, then label it a hypothesis and test it before shifting budget.
• If the competitive context changes (new entrants, a new revenue or advertising model, a changed frame of reference), then redo the audit instead of reusing old conclusions.

• Treating engine brand image as fixed. The tell: the audit has not been redone since launch, although the passages show context changes outcomes.
• Reading engine differences as ad-copy problems. The tell: endless copy rewrites while the gap between engines persists.
• Stating beliefs about an engine's reputation as facts. The tell: no test or data in the file, only assertions.
• Optimizing the link and ignoring the SERP. The tell: strong copy placed next to many similar competitors, with no look at the page-level impression.
• Mismatching your brand image and the engine environment. The tell: a premium positioning delivered through claims the searcher cannot verify quickly.

• ▸Target searcher: one line describing who searches and what they want.
• ▸Engine under review: name the engine being assessed.
• ▸Engine-choice hypothesis: why this searcher might choose it. Mark as hypothesis.
• ▸Trust belief: what the searcher likely believes about the engine and its paid placements. Mark as hypothesis.
• ▸SERP observation: your ad position, neighboring competitors, and overall page impression.
• ▸Link/ad-level check: which claims in your ad are verifiable on the landing page.
• ▸Brand image match: desired online image versus what the ad and page convey.
• ▸Observed performance: clicks and purchases for this engine, kept separate from other engines.
• ▸Next test and review date: the one test to run and when to audit again.

1. 1Write the chain for each keyphrase in the order the source models it: searcher need → query (the terms the searcher believes express the need) → matching criterion → advertiser keyphrase → ad. Perceived intent match is judged at the last link, so any break upstream shows up there. Query and keyphrase are 'cognitively linked associates', so write the query a real person would type, not the keyphrase you want.
2. 2Split the need by motivation before you write anything. The framing-shop example asks whether the buyer is framing a picture for themselves, a mother, a spouse or a friend. Each motivation changes behavior, mood and expectation, and the mood determines which ad copy works. Give each motivation its own row.
3. 3Tag each keyphrase with one intent class: informational, navigational or transactional. Add its SEC class: search good (all attributes knowable pre-purchase), experience good (attributes known only after use) or credence good (attributes may stay unknown even after). The SEC class tells you what uncertainty the searcher is trying to reduce, which the ad must address.
4. 4Set the matching criterion to your confidence in the intent. Matching runs from exact to very loose, including spelling and term variants. Use exact or tight matches where the query's intent is clear. Allow loose matches only where any plausible query carries the same need.
5. 5Draft the ad so it passes the three perception stages: detection (the ad is seen), recognition (it feels familiar) and identification (the searcher can label it as the answer to the need). Put the need-specific words from the query, and the attribute the SEC class says they are checking (price for a search good), in the title or description. Check that the URL and brand are read as trustworthy, because branding and opinions shift perception.
6. 6Compare your ad against the neighboring ads in the choice set for just-noticeable difference. If your wording, capitalization or semantics differ from the rest by less than a detectable margin, the searcher will not perceive a distinct match. Change the element the searcher scans first.
7. 7Test against satisficing. The searcher stops at a reasonable solution, not the optimal one, and may already have done ongoing prepurchase search. The ad must read as a reasonable solution at a glance, without further reading. Cut any claim that needs effort to decode.
8. 8Prune: for query-keyphrase pairs where the ad would be read as a mismatch, tighten the match or remove the keyphrase. The source states that leaving an ad slot empty can be more profitable than showing a nonrelevant ad.

Framing-shop scenario from the source: the shop's site is live and sponsored search is starting. The source asks whether the searcher is framing a picture for themselves, a mother, a spouse or a friend, and says each motivation changes mood, expectation and which copy works. Step 1: build four rows, one per motivation. Step 2: for each row, write the query that person would type (these queries are illustrative, not from the source) and tag the intent. A searcher looking for general framing options is informational. A searcher typing the shop's name is navigational. A searcher ready to order a custom frame is transactional. Step 3: a framing job is a service whose quality may not be known until after delivery, so lean toward experience-good treatment and write copy that reduces that uncertainty. Step 4: write one ad variant per motivation, for example a gift-oriented angle for the 'mother' and 'friend' rows and a personal-use angle for the 'self' row, and tie each to the matching query set. Step 5: use tight matching for the transactional row. Step 6: review the copy against neighboring ads for detectable difference. The judgment you are optimizing is whether each searcher thinks 'this ad addresses my need'.

• If the query's intent is ambiguous across informational, navigational and transactional readings, then tighten the matching criterion or drop the keyphrase, because a loose match invites a perceived mismatch.
• If the good is inexpensive, then expect some searchers to just buy rather than search, and write the ad to the transactional intent instead of an explanatory one.
• If the good is an experience or credence good, then the searcher is reducing uncertainty after the click as well as before it, so the ad must signal how that uncertainty is handled. For a search good, state the verifiable attributes, including price.
• If your ad differs from adjacent ads by less than a detectable margin, then change the wording, capitalization or phrasing so the difference passes the just-noticeable threshold.
• If the only available ad for a query would be read as irrelevant, then show nothing; an empty slot can be more profitable than a nonrelevant ad.

• Writing to the keyphrase instead of the query: the ad echoes the advertiser's term and not the searcher's wording. The tell is that the copy would not read as an answer to the typed query.
• One ad for every motivation: the framing shop sends the same copy to self, mother, spouse and friend searchers. The tell is that the copy fits one mood and ignores the others.
• Loose matching by default: the ad is served to queries whose intent differs from the ad's. The tell is an ad whose offer does not match the query it appears under.
• Assuming the searcher's evaluation starts when they see the ad: ongoing prepurchase search means impressions land on people who have already formed views of the brand or URL. The tell is that a URL trusted by one searcher draws suspicion from another and no copy change fixes it.
• Copy that is indistinguishable from neighbors: the ad is seen but not perceived as a distinct match. The tell is that it is detected but never identified as the answer.

• ▸Searcher motivation: who is the need for (self, mother, spouse, friend, other) and what mood or expectation follows.
• ▸Likely query: the exact terms this searcher would type, in their words.
• ▸Intent class: informational, navigational or transactional.
• ▸SEC class: search, experience or credence good, and the uncertainty the searcher is reducing.
• ▸Keyphrase and matching criterion: your term, and the exact-to-loose setting chosen with a one-line reason.
• ▸Ad title and description: the copy shown, with the need-specific attribute stated.
• ▸Perception check: can the searcher detect, recognize and identify this ad as the answer to the need? Note which stage fails.
• ▸Distinctness vs. neighbors: what differs from adjacent ads, and is it large enough to notice?
• ▸Trust and brand check: does the URL or brand read as trustworthy to this searcher?
• ▸Keep, tighten or drop: decision, including whether to leave the slot empty.

1. 1Map the real viewing frame first. Run your target query on the device your customer uses and count how many results sit above the fold. The source says ten is an exaggeration and that most users do not scroll. Your ad competes for attention inside that small visible region, so that is where you design.
2. 2Treat the ad as a signal, not the product. The ad does not solve the customer's problem and cannot deliver the product. Its only job is to convey that the answer the searcher wants is on the landing page. Write one sentence naming the need and one claim that the page resolves it, and cut everything else.
3. 3Write for fast perception. The ad is a few lines of text, perhaps with a small image, and people now process information in smaller chunks. Give each line a single job, such as the need, the substance, and the next step. Test each line by asking whether a searcher scanning it would register it without reading the others.
4. 4Minimize effort and choice. Searchers cut their option set quickly and take the path of least effort. Remove anything that makes the searcher compare, decode, or guess. Lead with the term that matches their query and state one clear offer instead of several.
5. 5Make the scent continuous from query to ad to page. The keyword you select is the direct link to the searcher, and the ad is the other. Echo the query's language in the ad, and make the landing page an extension of the ad copy, using the same promise in the same words.
6. 6Put the primary conversion action above the fold of the landing page. The same limited-screen logic that governs the SERP governs the page. A searcher who clicked on the strength of your signal should see that signal confirmed without scrolling.
7. 7Match ad treatment to keyword type. Query terms follow a power-law distribution: a few head terms are used a lot and many tail terms are used rarely. Find the sweet spot of terms that yields convertible traffic, and write ads that fit each term's specificity.
8. 8Audit and retest against live competition. Review your ad next to the other results returned for the same query. Ask why a searcher would read and click yours out of everything on the page. Keep the human-attention principles fixed and let the copy variants change, since the source argues that theory outlasts interface and algorithm changes.

The source's running case is a virtual framing shop that has just gotten its site indexed. Indexing is the easy part. The real challenge starts when a potential customer types a framing query and the engine shows ten results at most, with fewer than ten above the fold. Most users do not scroll, so your ad has to earn attention in that small space. Illustratively, you choose a query closely tied to the shop and write a short ad: the first line repeats the searcher's framing need, the second states the substance of the offer, and the third signals that the answer is on the page. You then check the landing page. The primary action, such as starting a framing order, must be above the fold, and the page must carry the same promise as the ad so it reads as an extension of it. You also compare your ad with the neighboring results for the same keyword. If a rival's ad matches the query's wording more directly and asks the searcher to do less, the diagnosis is a perception and effort problem, not a bidding problem. You then rewrite to close that gap.

• If your ad sits below the fold on the target device, then treat visibility as the first problem and fix it before refining copy, because most users do not scroll.
• If the ad promises something the landing page does not show immediately, then rewrite the page or the ad until the page reads as an extension of the ad copy.
• If the primary conversion action is below the fold of the landing page, then move it above the fold before running more traffic.
• If two ads appear for the same keyword and one gets more clicks, then compare them on attention and effort (match to the query, clarity of the signal, number of things the searcher must process) before touching bids.
• If a keyword is a head term or a tail term, then write the ad for that term's breadth. Keep head-term ads broad and quickly scannable, and make tail-term ads mirror the specific query.

• Writing ads from checklists or past results with no model of why they work. The tell is that performance collapses when the interface or context changes and no one can explain the drop.
• Treating the ad as the product pitch and cramming in multiple offers. The tell is a cluttered few lines with low click-through, because the searcher must process too many chunks.
• Breaking the ad-to-landing-page message. The tell is clicks that do not convert, because the searcher does not find the promised answer.
• Ignoring screen real estate and the fold. The tell is an ad that is well written but rarely noticed, or a landing page where the conversion action needs scrolling.
• Optimizing mechanics (bids, platform settings) while skipping the human side. The tell is endless interface tuning with no improvement in click or conversion, even though sponsored search is a people venture.

• ▸Target query: write the exact query a real customer would type.
• ▸Device and fold count: how many results (ads and organic) are visible above the fold on that device?
• ▸Ad position in the visible set: where does your ad fall among those results?
• ▸Need echoed in line 1: quote the words that mirror the searcher's query.
• ▸Substance claim: state the one specific thing the ad says the searcher will get.
• ▸Landing-page promise match: does the page repeat the ad's promise in the same words (yes/no, with quote)?
• ▸Primary action above the fold: name the action and note whether it is visible without scrolling.
• ▸Competitor comparison: for the same keyword, what does the strongest neighboring ad do with less effort for the searcher?
• ▸Keyword type: head, tail, or mixed, and how the ad's specificity fits it.
• ▸Next test: one change to the ad or page, and the attention or effort reason for it.

1. 1State the searcher's need as a product-or-service need, including the information they want about it. The source assumes every sponsored-search searcher is looking for a product or service, so write the need in those terms before you pick any keyphrase.
2. 2Classify the offering on the Search-Experience-Credence (SEC) continuum. Search goods have attributes you can inspect before purchase (books, clothes, music downloads). Experience goods reveal attributes only through consumption (brokerage service, vacation packages, healthcare). Credence goods leave the buyer unsure of long-term quality even after use (tax services, counseling, medical treatment). Treat the three as overlapping regions, not hard boundaries.
3. 3Read prepurchase quality uncertainty off the classification. It rises from search to experience to credence goods as quality gets harder to judge. Place your offering on that gradient and record the placement as the baseline uncertainty level.
4. 4Apply the cost moderator. If the product or service is inexpensive, the searcher may just buy it instead of spending time and effort searching. Mark low-cost items as low search-effort and expect little uncertainty-reduction behavior.
5. 5Identify what the searcher lacks: knowledge of the product, of the service, or of the expected outcome of consumption. Where that knowledge is missing, the searcher is more likely to run uncertainty-reduction processes, so list those gaps explicitly.
6. 6Map the searcher's likely query terms to your advertiser keyphrases. Follow the chain in Figure 3.4: need, then query terms, then matching criteria, then keyphrase, then ad. Choose terms that fit how the uncertainty shows up for your product type.
7. 7Account for the short-query constraint. Empirical research shows queries are typically very short, and people process only about seven chunks of similar information at a time. Keep ad words in chunks the searcher can recognize or remember.
8. 8Check for competing motivations such as entertainment or risk reduction. Uncertainty reduction is a major component of e-commerce searching but not the only one, so address the whole range of motivations the campaign meets.

The source describes a friend who owns a high-end calendar business with exquisite photos for niche markets such as kittens, airplanes, and fishing. High-end calendars are primarily an experience product: customers want to see the pictures and feel the paper quality. That is why most of his customers are repeat buyers who first bought a calendar at a big-box reseller. Searchers cannot resolve the uncertainty before purchase, so the in-store experience does that work. His sponsored-search campaigns therefore bid mainly on branded terms (name, URL), jumping ahead of the big-box sellers because he makes more money selling directly. He also cooperates with the in-store resellers so they provide the experience to his future direct customers. The SEC classification told him which terms fit the product and where uncertainty actually gets reduced.

• If the product is a search good (attributes inspectable before purchase), then expect uncertainty to be lowest and build keyphrases around the inspectable attributes.
• If the product is an experience good, then expect higher uncertainty and plan for the searcher to need an experience path (as with the calendar seller, whose resellers supply it) before they buy direct.
• If the product is a credence good, then assume uncertainty is highest and persists even after use, so expect heavy information searching.
• If the product or service is inexpensive, then expect the searcher to buy rather than search, and do not over-invest in information-heavy keyphrases.
• If your customers already know your brand from an earlier experience, then bid on branded terms (name, URL) to capture them directly.

• Treating the SEC categories as rigid boxes. The tell: you assign one label to an offering that sits between regions, when the source says the categories overlap on a continuum.
• Assuming uncertainty reduction is the only search motive. The tell: your campaign ignores entertainment or risk-reduction searchers, whom the source says sponsored search must also address.
• Ignoring cost. The tell: you build long information-seeking funnels for cheap products, where the searcher would simply buy.
• Writing long, multi-attribute queries and ad copy. The tell: the text exceeds what a searcher can chunk, while queries in practice are short.
• Skipping the need-to-keyphrase linkage. The tell: keyphrases are chosen because they historically worked, not because they match how the searcher expresses the underlying need.

• ▸Product/service: name the offering being advertised.
• ▸Searcher need: write the need, including the information wanted about the product or service.
• ▸SEC placement: mark search, experience, or credence, and note any overlap between regions.
• ▸Prepurchase quality uncertainty: rate low, medium, or high based on the SEC placement.
• ▸Cost level: note whether the item is cheap enough that the searcher may just buy instead of searching.
• ▸Knowledge gaps: list what the searcher may not know about the product, service, or expected outcome.
• ▸Likely query terms: write short queries the searcher would use to express the need.
• ▸Advertiser keyphrase and match: record the keyphrase and the matching criteria linking it to the query terms.
• ▸Other motivations: note any entertainment or risk-reduction motives to address.

1. 1Define the click before you measure it. Use the SEMPO definition from the source: a click is the act that directs a searcher from an ad to a specified landing page on your site. Count only that act. Impressions, views, and intentions to click are different events.
2. 2Treat the click as a behavior, not a feeling. Behaviorism holds that there is no knowable difference between two states unless a demonstrable difference in behavior marks them. So a click or no-click is your evidence. Do not claim the searcher 'liked' or 'trusted' the ad unless a behavior shows it.
3. 3Log each click with its context. The source says behavior has internal sources (cognitive, affective, expertise) and external ones (environmental, situational). Record the keyphrase, the ad, the rank, and the position slot, so you can separate what the ad did from what the setting did.
4. 4Keep rank and position as separate fields. Rank is the numerical label for an ad. Position is the slot it occupies. A change in click behavior may follow either one, so do not merge them in your records.
5. 5Read the ad as a signal. Signaling theory says searchers use ad cues to judge the Web site when they cannot assess it beforehand. Cues that are costly to fake earn more trust. Write ads whose claims are specific and verifiable on the landing page, then compare clicks across variants.
6. 6Group clicks by campaign. A campaign is a set of keyphrases, bids, and ads cognitively linked to one topic. Compare click behavior inside a campaign first, so differences reflect the ad or keyphrase and not a change of topic.
7. 7Log cost with the click. Under pay-per-click you pay only for clicks. Record the maximum cost-per-click, which is equal to or below the keyphrase bid, next to the clicks each keyphrase produces.
8. 8Treat the click as a transition, not a result. The click turns a searcher into a potential customer. Hand the click to your next measure on the landing page, and do not report it as a sale.

The source's running case is a framing shop that does not appear in the top ten organic results for the query 'frame shops' in its area. It therefore turns to sponsored search, where it pays the search engine only when someone clicks an ad that sends them to its landing page. Say the shop runs two ads on that keyphrase in one campaign. One gives a generic claim. The other states something the landing page can confirm, which is a costlier signal to fake. Following the behaviorist rule, the shop compares only observed clicks on each ad, at logged rank and position. If the second ad draws more clicks, that is a demonstrable behavioral difference, and the shop can say the stronger signal likely contributed. It cannot say it knows why the searcher clicked. Each click is also logged against the maximum cost-per-click, since the shop pays per click. The shop reads the click as a searcher becoming a potential customer, and it judges the business result on the landing page.

• If two ads show different click counts at the same rank and position, then treat the difference as behavioral evidence for the better signal. If the rank or position differs, do not attribute the gap to the ad alone.
• If you cannot point to an observed behavior that distinguishes two states, such as 'interested' versus 'not interested', then do not report the states as different.
• If an ad's claim is easy to fake, then expect lower trust and rewrite it with a claim the landing page can verify.
• If clicks rise while the cost you pay per click rises too, then check your maximum cost-per-click against the keyphrase bid before judging ad quality.
• If you are comparing click behavior across keyphrases, then compare inside one campaign (one topic) first, and only then across campaigns.

• Reading a click as proof of intent or satisfaction. The tell: reports say 'searchers want X' when the only data is click counts. The source allows cognitive and affective states to disconnect from behavior in some contexts.
• Mixing up rank and position. The tell: click comparisons contain no record of slot, and one 'position' column holds both labels.
• Treating the bid as what you pay. The tell: cost projections use the keyphrase bid, although the maximum cost-per-click is equal to or less than that bid.
• Ending analysis at the click. The tell: the dashboard stops at click-through and says nothing about what happens on the landing page. The click only makes someone a potential customer.
• Comparing ads across unrelated campaigns. The tell: a click gap is credited to ad wording when the keyphrases serve different topics.

• ▸Campaign (topic): name the set of keyphrases, bids, and ads tied to one topic.
• ▸Keyphrase: the query term the ad responded to, e.g., 'frame shops'.
• ▸Ad variant and signal claim: paste the ad text and note the verifiable claim it makes.
• ▸Rank: the numerical label the ad held when clicked.
• ▸Position: the slot the ad occupied. Do not merge it with rank.
• ▸Observed clicks: the count of clicks that led to the specified landing page.
• ▸Keyphrase bid and maximum cost-per-click: record both, noting that the max is equal to or less than the bid.
• ▸Context notes: any internal or external factor that may have shaped the clicks.
• ▸Next landing-page behavior: what the clicking searcher did next, so the click is not read as the outcome.

1. 1Fix the formula and the window before you read anything: CTR = clicks ÷ impressions over one stated time period. Compare only periods of the same length, and record the start and end dates beside every figure.
2. 2Define the unit you are measuring. A sponsored-search effort is one or more campaigns. A campaign is a set of keyphrases, bids, and ads linked to one topic. Compute CTR per campaign first, then roll up to the effort. Treat the effort-level number as a health reading, not a diagnosis.
3. 3Break each campaign down by keyphrase and by ad. CTR is an aggregate outcome, so the average hides which keyphrase or ad is carrying or dragging it. The source frames the questions as 'Why do certain keywords work whereas others do not?' and 'Why does that ad work well when others that are very similar do not?' Use those two questions to organize the breakdown.
4. 4Log rank and position next to every CTR reading. Rank is the numerical label for an ad. Position is the slot it occupies. Never compare the CTR of two keyphrases without noting where their ads were shown, because the first page and the top of the list is where most consumer traffic is.
5. 5Record the keyphrase bid and the maximum cost-per-click for each row. The maximum cost-per-click is equal to or less than the bid. A CTR change that follows a bid change is a pricing and position effect. A CTR change with no bid change points to keyphrase or ad selection.
6. 6Check keyphrase-to-ad fit. A keyword links a searcher's query to an ad, and a keyphrase is two or more keywords. For each low-CTR row, read the query the keyphrase would match, then read the ad as that searcher would, and note the mismatch.
7. 7Run one-variable tests on similar ads. Change one element, hold keyphrase and position constant, and compare CTR over equal windows. The source treats keyword advertising as an empirical field where you take nothing at face value and always test.
8. 8Pair CTR with a downstream result before you act. Under pay-per-click you pay for clicks, but a click only makes a searcher a potential consumer. The goal is a consumer who becomes a potential customer, someone who converts by purchasing or signing up. Do not call a high CTR a success until you have checked what those clicks did.

Faster Frames, the framing shop used throughout the source, does not appear in the top ten for the query 'frame shops' in its area. The owner builds a sponsored-search effort with one campaign on the topic of framing, containing a set of keyphrases, bids, and ads. For one period, say the first week, the owner records impressions and clicks for each keyphrase and divides clicks by impressions to get its CTR. Each row also notes the ad's rank and position and the keyphrase bid against the maximum cost-per-click. Two keyphrases have similar ads, but one has a clearly lower CTR. The owner does not conclude that the keyphrase is bad, because the source says to ask why similar ads perform differently. The owner first checks whether the low row sat in a lower position, then whether the ad copy matches the query. A single-variable ad test follows over an equal-length week. The source gives no benchmark CTR, so Faster Frames is judged against its own earlier periods and against its own keyphrases, not against an outside number. Finally, the owner checks whether the clicks led to sign-ups or purchases on the Web site.

• If CTR drops while rank or position also dropped, then treat it as a position and bid problem. Check the bid and the maximum cost-per-click before touching the ad.
• If CTR drops while position and bids are unchanged, then look at keyphrase and ad selection. Re-read the ad against the likely query for that keyphrase.
• If two similar ads show different CTRs in the same position and period, then test one element at a time to find the cause. Do not rewrite both ads at once.
• If the effort-level CTR looks healthy but conversions are weak, then audit keyphrases for clicks from searchers who never become potential customers. Do not report CTR as the success measure.
• If a period is shorter, longer, or differently timed than the one you compare it to, then discard the comparison. CTR is defined only within a stated time period.

• Reading the aggregate and stopping. The tell is that you report one effort-level CTR and cannot name which keyphrase or ad moved it.
• Ignoring position. The tell is that a keyphrase is judged poor when its ads simply ran lower on the page, which you can see only if rank and position were never logged.
• Confusing bid with maximum cost-per-click, or using loose terms interchangeably. The tell is that two people in a review mean different things by 'bid' and 'keyword' versus 'keyphrase', so the diagnosis goes wrong.
• Treating a click as a customer. The tell is a rising CTR beside flat sign-ups or sales. Clicks bring searchers, not consumers who convert.
• Changing several things at once. The tell is that CTR moves after an edit to ad copy, keyphrases, and bids together, and nobody can say which change caused it.

• ▸Campaign topic: name the topic the keyphrases, bids, and ads are linked to.
• ▸Keyphrase: write the exact keyphrase, noting whether it has one keyword or two or more.
• ▸Ad identifier: label the specific ad shown for this keyphrase.
• ▸Period: state the start and end dates, and use the same length for every compared row.
• ▸Impressions and clicks: record both counts for the period.
• ▸CTR: calculate clicks ÷ impressions for the row.
• ▸Rank and position: record the ad's numerical rank and the slot it occupied.
• ▸Keyphrase bid and maximum cost-per-click: record both, and confirm the maximum is equal to or less than the bid.
• ▸Downstream result and next test: note whether clicks led to a sign-up or purchase, then name the one variable you will change next.

1. 1Name one conversion per campaign in a single sentence: the exact action on the landing page (purchase, newsletter signup, paper download) that fulfills the advertiser's goal. If you cannot name it, no KPI can be measured against it.
2. 2Decide whether the goal is a sale now or a sale later. The source notes advertising can aim at a future sale through branding. If it does, define the conversion as the action that marks that future-sale step (such as a signup), not an immediate purchase.
3. 3Trace the chain from impression to click to conversion. Record clicks and impressions for a fixed time period, and compute CTR as clicks divided by impressions for the same period. Treat CTR as a health check on ad and keyword selection, not as the result.
4. 4Write each KPI goal (CTR and the others) as a target that must serve the conversion. Review every KPI against conversions. The source cautions that CTR correlates only weakly with conversion rates, so a CTR gain that does not move conversions is not a success.
5. 5Segment conversion results by ad rank. Rank affects both CTR and conversion rates, and searchers click near the top because of the serial position effect. A click from a high position and a click from a lower one are not equal evidence of ad quality.
6. 6Align the landing page to the searcher's state. Searchers expend the least effort, view few results, and spend limited time on a site. They arrive with uncertainty and want information, such as price, features, or feel. Put that information, and the conversion action, where it is reached with minimal effort.
7. 7Test changes with an A/B or multivariable design, and size the sample before you start. The source's formula is n = (zα/2 · σ / E)². For a quantitative analysis, plan on roughly 35–40 subjects as the minimum rule of thumb.
8. 8Check validity and reliability before acting. Ask whether the result generalizes (external validity), whether the measure is stable on repeat tests (reliability), and whether the sample is large enough to be statistically significant.

The source gives a sample-size calculation you can apply to a landing-page test. You want to compare two landing pages and need to know how much data to collect first. Set the confidence level at 95 percent, so α = 0.05 and zα/2 = 1.960. Choose a margin of error E = 5 and use a standard deviation σ = 15. Then n = (1.960 × 15 / 5)² = 5.88² = 34.5744, which rounds up to 35. You therefore need at least 35 observations, and the source's rule of thumb is 35–40 for quantitative work. Fewer than that, and the source warns the law of probability can produce large errors. The source's formula is for a quantitative measure with a known σ, so check that your conversion metric fits it before relying on 35. Run the test on conversions, not clicks. A variant that lifts CTR but not conversions has not succeeded, because early research found CTR a weak indicator of conversion.

• If a campaign's CTR rises but conversions do not, then treat it as a failure to optimize and look at the landing page and the keyword-to-ad match, not the ad's click appeal.
• If conversions differ between ad positions, then compare them within the same rank before judging an ad or keyword, since rank affects both CTR and conversion rate.
• If the goal is a future sale rather than an immediate one, then set the conversion to the interim action (signup, download) and judge KPIs against that.
• If you are about to call an A/B test, then confirm the sample size meets your computed n (the source's example gives 35, with 35–40 as a rule of thumb) before declaring a winner.
• If the landing page makes the searcher work to find the information or the action, then simplify it first, because searchers expend the least effort and spend limited time on a site.

• Treating CTR as the goal. The tell: reports lead with clicks and CTR while conversions go unreported, even though CTR correlates weakly with conversion.
• Leaving the conversion undefined or having several vague ones. The tell: different stakeholders cite different 'successes' for the same campaign.
• Reading positional bias as ad quality. The tell: top-ranked ads look best on clicks and conversions, and no one has controlled for rank.
• Calling A/B tests on too little data. The tell: a winner is declared from a handful of observations, below the 35–40 rule of thumb.
• Sending the searcher to a page that does not answer the information need (price, features, emotional appeal). The tell: clicks arrive but visits are brief and no action is taken.

• ▸Campaign / ad group: name the unit you are measuring.
• ▸Conversion action: one sentence naming the exact landing-page action (purchase, signup, download).
• ▸Timing of the sale: immediate or future; if future, name the interim action that counts.
• ▸Impressions and clicks (period): record both for the same time window, then compute CTR.
• ▸KPI targets: list each KPI goal and state how it ties back to the conversion.
• ▸Ad rank / position: note the position(s) the ad ran in so results can be compared like for like.
• ▸Searcher information need: what the page must show (price, features, feel) and where the conversion action sits.
• ▸Test design and sample size: variants compared, confidence level, E, σ, and the computed n.
• ▸Result and validity check: conversions per variant, whether n was met, and whether the result would hold across contexts and repeat tests.

1. 1Define ROI as one ratio before touching a bid: revenue the advertising generates compared with advertising cost. The source treats it as the terminal outcome. Every other metric (CTR, position, impressions) is an input to it. Write the formula down and fix the revenue and cost definitions for the whole account.
2. 2Set the expected ROI at the strategy level, alongside bidding strategy, budget allocation, account structure and effort goals. The source places expected ROI in that tier, shaped by branding, advertising factors (market size, audience size, budget) and the seven Ps and three Cs. Decide it before choosing tactics such as keyphrases, ad copy and scheduling.
3. 3Do the market research on how you will evaluate ROI before launch. In the frame-shop case, analyzing customers and planning the ROI evaluation both come before bidding. Decide what counts as revenue, and how it will be tied back to a keyphrase.
4. 4Convert expected ROI into a per-keyphrase bid. The bid is the maximum you would pay if the engine shows your ad for a query linked to that keyphrase. The source frames the core question as relating the price per keyphrase to what the business makes from the advertising. Estimate what a click on that keyphrase is worth to you, then bid at or below that value. Bids must also clear the engine's floor price.
5. 5Segment keyphrases into head (high volume, more expensive) and tail (low volume). The power law explains why some phrases cost much more than others in the same vertical. Compute ROI separately for each group, because head terms carry higher prices and tail terms may deliver the targeting efficiency that lowers cost relative to return.
6. 6Measure conversion and revenue, not clicks alone. Early research cited in the source finds CTR a weak indicator for ROI optimization because its correlation with conversion rates is often low. Track each keyphrase through to conversion and attribute revenue to it.
7. 7Review each campaign on a known scale of cost (advertising cost) and result (ROI), as sponsored-search analytics prescribes. Use the comparison to decide where to push and where to stop, and to avoid costly experiments that you cannot measure or analyze.
8. 8Balance effectiveness against efficiency. Effectiveness is how well the system accomplishes its goal. Efficiency is how it uses resources to get there. A campaign can satisfy one and fail the other, so check both before adjusting budgets.

The source's running case is the frame shop. Before bidding, the owner analyzes customers to target likely buyers, does market research on how advertising ROI will be evaluated, and picks the keyphrases that trigger ads. Then comes the question the source poses directly: how much should the shop pay per keyphrase, and how does that price relate to what the business makes from the advertising? The shop tells the engine the most it will pay when an ad shows for a query tied to a keyphrase, and the engine sets a floor price. The ROI logic then works as follows (illustrative figures, not from the source). Suppose a keyphrase brings 100 clicks at a $1.00 average cost, giving $100 of cost. If those clicks produce $300 of framing revenue, the ratio of revenue to cost is 3 to 1. A second keyphrase with a higher CTR but few conversions might produce $40 on the same $100. That is exactly the case where CTR misleads and ROI governs. The shop lowers the bid on the second phrase and keeps or raises the first, reading the auction price as its own statement of expected ROI.

• If a keyphrase shows a high CTR but weak conversion, then treat it as a poor ROI candidate and lower the bid or pause it, because CTR correlates only weakly with conversion.
• If a keyphrase's expected revenue per click is below the price needed to win the position, then do not chase the position, because the bid should reflect expected ROI from that slot.
• If a head keyphrase costs far more than comparable phrases, then require its measured ROI to justify the premium. Otherwise shift budget toward tail phrases, where targeting is more specific.
• If an experiment is too costly in time or money to measure and analyze, then do not run it until a cheaper way to read cost and result exists.
• If a change improves efficiency but lowers effectiveness (or the reverse), then compare both against the ROI target before keeping it.

• Optimizing CTR as if it were ROI. The tell: click rates improve while revenue per advertising dollar stays flat or falls.
• Bidding without a stated expected ROI. The tell: bids are set by position or competitor price, and no one can say what a click is worth to the business.
• Skipping conversion tracking per keyphrase. The tell: you can report cost and clicks but cannot attribute revenue to specific phrases.
• Treating ROI as a tactical afterthought rather than a strategy-level parameter. The tell: budget allocation and account structure were decided before any ROI target existed.
• Lumping head and tail terms into one average. The tell: a blended ROI looks acceptable while expensive head terms hide losses or tail terms hide gains.

• ▸Keyphrase: the exact phrase that triggers the ad.
• ▸Head or tail: classify by query volume, since head terms are higher volume and costlier.
• ▸Expected ROI target: the revenue-to-cost ratio you require, set at strategy level.
• ▸Estimated revenue per click: your expected revenue from a click on this phrase, based on market research.
• ▸Maximum bid: the highest price you will pay, derived from revenue per click and the ROI target.
• ▸Engine floor price: the platform minimum, to confirm the bid clears it.
• ▸Actual cost: advertising spend on this phrase for the review period.
• ▸Attributed revenue: revenue traced to this phrase through conversion tracking.
• ▸Measured ROI and action: compare attributed revenue with actual cost, then raise, hold, lower or pause.

1. 1Fix the campaign topic first. The source defines a campaign as a set of keyphrases, bids, and advertisements cognitively linked to one topic. Write the topic in one sentence before you list any keyphrase. A keyphrase that does not fit that sentence goes in another campaign or gets cut.
2. 2Name the transition you are buying. The source says a searcher becomes a potential consumer, and a consumer may become a potential customer (someone with a realistic chance of converting), and that this transition is the point of sponsored search. State what conversion means for this offering (a purchase, a newsletter signup) and what a searcher near that point would be trying to do.
3. 3Profile the searcher with the source's model of searcher intent. Figure 2.3 breaks it into an information need, cognitive factors, and situational factors. Add the affective state the construct definition names. For each topic, write one line per factor on what a convertible searcher knows, where they are, and how they feel.
4. 4Draft candidate keyphrases in the searcher's terms, not yours. A keyword is the advertiser's word and a term is the searcher's word, so your job is to choose keywords that link to terms searchers actually type. Write each candidate as two or more keywords, since a keyphrase is a set of two or more keywords. Next to each, write the query you expect it to match.
5. 5Run a relevance test on each candidate. Ask whether the information need behind the likely query is one your offering satisfies. The source defines relevance as how well a result meets the searcher's need. If the answer is no or partly, cut the candidate or narrow it. Do not keep it just because it has volume.
6. 6Check the keyphrase against the ad and the landing page. The search engine's review process checks that advertiser content is relevant to the targeted keyword. Google's quality score also draws on click-through rate, keyword relevancy, and landing-page quality. Write the ad and name the landing page for each keyphrase, and confirm that all three tell the same story.
7. 7Set the bid only after the keyphrase passes. Maximum cost-per-click is equal to or less than the keyphrase bid. The source's framing-shop example uses a $1.00 maximum per click as a starting point to be tuned once historical data exists. Treat early bids as provisional.
8. 8Log your reason for every keyphrase and review it against results. The source warns that doing something because it worked before holds up only until the context, situation, or environment changes. A written reason (the intent you expected) lets you see which assumption broke when performance moves.

The source's running example is a small framing shop that sets up an account, enters keywords, crafts ads, and sets a $1.00 maximum bid per click. The passages do not list its keywords, so the keyphrases below are illustrative. Topic sentence: people who want a specific item professionally framed. Candidate keyphrase A is 'custom picture framing'. The expected query matches an information need (find a framer) that the shop satisfies, so it passes the relevance test. Candidate B is a single broad word about art. It could come from someone studying art history or buying a poster, and the shop does not meet either need. That searcher is far from becoming a potential customer, so B is cut or moved to a separate topic. Ad and landing page for A both lead with custom framing, so the keyword, ad, and page agree. The $1.00 maximum is kept as a provisional bid. If clicks on A arrive but nobody converts, the logged intent assumption is the first thing you re-examine, not the bid.

• If a keyphrase's likely query expresses an information need your offering does not meet, cut it or move it to a different topic, because the source treats relevance as meeting the searcher's need.
• If you cannot name the situational and cognitive state of the searcher who types the query, do not add the keyphrase yet. You are guessing at intent.
• If a keyphrase cannot be matched to an ad and a landing page that repeat its promise, do not launch it, because keyword relevancy and landing-page quality both feed the quality score.
• If a keyphrase only works 'because it historically works', treat it as provisional. Context changes make historical data of little value.
• If a keyphrase fits no existing campaign topic, create a separate campaign instead of widening an existing one, because a campaign is keyphrases, bids, and ads linked to one topic.

• Picking keywords from the advertiser's vocabulary instead of searchers' terms. The tell is impressions with low click-through, because the query and the offer do not meet.
• Treating more keyphrases as better coverage. The tell is a long list in one campaign with no single topic sentence that every keyphrase serves.
• Chasing broad, high-volume words that attract searchers far from becoming potential customers. The tell is clicks without conversions, and a click-through rate that drags down quality score.
• Selecting keyphrases by checklist or tool output without the reason for each. The tell is that nobody can say why a keyphrase is there, so no one can diagnose a drop when the context shifts.
• Fixing a mismatch with a higher bid. The tell is that spend rises while click-through rate and conversion stay flat. The source notes quality factors make it more expensive to buy position without clicks.

• ▸Campaign topic sentence: write the one topic the keyphrase must serve.
• ▸Candidate keyphrase: two or more keywords, exactly as entered.
• ▸Expected searcher query terms: write the words you expect a searcher to type.
• ▸Information need: state what the searcher is trying to find or do.
• ▸Cognitive and situational factors: note what they likely know and the circumstances they are in, plus the affective state if you can infer it.
• ▸Offering fit: yes, partly, or no, with one line on how the offering satisfies the need.
• ▸Ad and landing page: name both and confirm each repeats the keyphrase's promise.
• ▸Provisional bid and max CPC: record the bid and the maximum cost-per-click, which must be equal to or below the bid.
• ▸Review note: record the intent assumption to check against the data and the date to check it.

1. 1List the keyphrases the ad is linked to before writing anything. In the framing-shop case these are "frames," "framing," and "custom framing." Signal quality is judged against the searcher's query terms, so the ad is graded against this list.
2. 2Sort every candidate word in the ad into the signal categories: branded term, query term, action term, location term, price term, quality term. Words that fit no category are candidates for noise.
3. 3Put at least one query-matching term in the ad, since the searcher is sifting the results page for cues that match what they typed. In the Faster Frame ad, "Custom Framing" does this job for the keyphrase "custom framing."
4. 4Add one concrete term from each remaining category you can support truthfully: a price or offer, an action cue, a quality assurance, and the brand name. Leave out any category you cannot back up. A missing signal is cheaper than a deceptive one.
5. 5Cap information load. The passages state that information helps only up to a point, after which more of it hurts the speed and accuracy of the searcher's response. Cut any term that does not clearly sit in a category.
6. 6Remember the ad is not the product and does not solve the consumer's problem. Its only task is to make the searcher perceive that it addresses their need and to earn the click. Write for that task.
7. 7Test word by word. Click and no-click behavior is measurable, and every word and aspect of display can be tested. Change one element at a time and compare click-through rates.
8. 8Check that the landing page delivers what the ad promises. Google's quality score drew on click-through rate, keyword relevancy, and landing-page quality, and it penalizes deceptive practices and poor sites.
9. 9Re-audit periodically. Signal and noise are contextual judgments, and today's noise can become tomorrow's signal, so a term that did not work earlier may deserve a retest.

The source's own ad for the framing shop reads: "Faster Frame / $50 Off Custom Framing Coupon / FASTER FRAME Satisfaction Guaranteed / www.fasterframe.com." It is linked to the keyphrases "frames," "framing," and "custom framing." Audit it by category. Brand: "Faster Frame" and "FASTER FRAME." Query term: "Custom Framing," which matches one of the keyphrases. Price: "$50 Off." Action: "Coupon." Quality: "Satisfaction Guaranteed." The ad has no location term. That is a gap, since the source frames the searcher's need as frame shops in our area. The audit gives two options: test a location term, or accept the omission. Because information load can slow and degrade response, any addition should replace a weaker word rather than extend the ad. Click-through rate on each variant then shows whether the added signal helps.

• If a keyphrase triggers the ad but the ad text contains none of its terms, then rewrite so the query term appears in the ad, because the searcher cannot connect the ad to their need.
• If the ad contains a signal in every category but click-through rate is low, then suspect information load and remove terms one at a time, since more information eventually hurts response.
• If a price, offer, or quality claim cannot be fulfilled on the landing page, then cut the claim, because deceptive practices are penalized in quality scoring and break the signal.
• If two variants differ in more than one word, then do not trust the comparison. Every word is testable, so change one element at a time.
• If a previously rejected term starts to look relevant to current searchers, then retest it, since noise and signal are contextual.

• Writing a clever ad that ignores the keyphrases. The tell is an ad that never repeats any term from the keyword list and shows a weak click-through rate.
• Stuffing every signal category into the ad. The tell is a crowded ad whose clicks drop as terms are added, which is the information-overload effect.
• Treating the ad as the product pitch, trying to close the sale in the ad text. The ad is not the product and does not directly solve the problem. The tell is long copy with no clear cue that earns the click.
• Making claims the landing page cannot support. The tell is clicks without conversions, plus quality-score penalties for relevancy or landing-page quality.
• Judging by opinion rather than measured behavior. The tell is ad copy that has never been tested word by word, even though every word can be.

• ▸Keyphrases linked to this ad: list each keyword or keyphrase the ad is triggered by.
• ▸Branded term(s): write the brand text that appears in the ad.
• ▸Query-term match: write which ad words repeat or match the keyphrase terms.
• ▸Price/offer term: write the price or offer cue, or 'none'.
• ▸Action term: write the action cue (e.g., coupon), or 'none'.
• ▸Location term: write the location cue, or 'none' and whether the query implies a local need.
• ▸Quality term: write the quality assurance, and confirm the landing page supports it.
• ▸Noise candidates: list any words that fit no category and are candidates to cut.
• ▸Test log: record the single change made, the variants compared, and the click-through result.

1. 1Record rank and position as two separate fields. Rank is the numerical label for your ad in the listing (1, 2, 3...). Position is the slot it occupies: north (above organic), south (below organic), east/right rail, or integrated among organic results. Two ads can hold the same rank in different positions, so a single 'position 3' note hides the difference.
2. 2Build the keyphrase list at campaign level before touching bids. For a framing shop the passages use 'frames,' 'framing,' and 'custom framing.' Rank is earned per keyphrase, so each one needs its own rank target.
3. 3Set a keyphrase bid for each keyphrase, then set the maximum cost-per-click at or below that bid. The bid is what you are willing to pay for serving the ad, typically per click. The maximum CPC can never exceed it.
4. 4Treat rank as a function of bid and quality score. The source's historical example shows how much the rule has changed. GoTo.com's 1998 first-price auction gave the top bidder the top position, with money as the transparent ranking factor. Do not assume money alone buys rank now; the construct definition adds quality score.
5. 5Expect price pressure to follow competition. The more businesses bid on a keyword, the higher the bid price needed to get your ad on the SERP. Check each keyphrase's contested level before deciding how high a rank you can afford.
6. 6Set a rank target tied to profit, not to the top slot. The goal is to acquire customers at a cost that still leaves a profit on what you sell. The top of the list draws the most traffic, but only buy it where the cost per acquired customer stays profitable.
7. 7Log observed rank and position per keyphrase and compare them against clicks and cost. The source says keyword advertising is an empirical field ('always testing'), with caveats for individual accounts, campaigns, keyphrases, and ads. Use your own data to find where lower ranks still pay.
8. 8Change one lever at a time: bid, or ad text. Re-record rank and position after each change so you can attribute any shift in clicks to the lever you moved.

Faster Frame, the source's framing shop, does not appear in the top ten organic results for 'frame shops' in its area. It opens a pay-per-click account and links the ad ('$50 Off Custom Framing Coupon... Satisfaction Guaranteed, www.fasterframe.com') to three keyphrases: 'frames,' 'framing,' and 'custom framing.' For each keyphrase it sets a keyphrase bid and a maximum CPC at or below that bid. Because several other businesses probably want the same keywords, higher competition pushes the required bid up, so Faster Frame cannot assume one bid level works across all three. It records, for each keyphrase, the rank its ad gets and the position (north, south, or east). It then compares clicks and cost per acquired customer across keyphrases. Where a higher rank costs more than a customer is worth, it stops raising the bid. Where the top slot yields profitable customers, it keeps competing for it. The source supplies the setup and the principle, not performance figures, so the rank-to-click comparison comes from the shop's own account data.

• If a keyphrase's rank is high but cost per acquired customer exceeds what the sale returns, lower the keyphrase bid (and the maximum CPC with it) and accept a lower rank.
• If many businesses bid on a keyphrase and the price keeps rising, treat top rank on that keyphrase as expensive. Shift budget to less contested keyphrases before chasing it.
• If rank stays low despite a bid you consider reasonable, check the non-bid factor. The definition says rank depends on bid and quality score, so review ad and keyphrase relevance rather than only raising the bid.
• If the same rank performs differently in different positions (north, south, east), analyze them separately. Rank is a label, position is a slot, and they are not interchangeable.
• If a theory-based expectation about rank and clicks conflicts with your account data, trust the test. The source says individual accounts, campaigns, keyphrases, and ads carry exceptions.

• Treating rank and position as the same thing. The tell: your reports show 'rank 2' with no note of whether the ad sat north, south, or east.
• Assuming the highest bid always wins the top slot, as in GoTo.com's first-price model. The tell: you keep raising bids while rank barely moves, and you never examine ad quality.
• Setting the maximum CPC above the keyphrase bid. The tell: your account settings show a maximum CPC higher than the bid, which breaks the rule that it is equal to or less than the bid.
• Chasing the top slot regardless of profit. The tell: rank improves and clicks rise, but cost per acquired customer exceeds what the products and services return.
• Applying one bid and one rank target to every keyphrase. The tell: 'frames,' 'framing,' and 'custom framing' share identical bids even though their competitive pressure differs.

• ▸Campaign / keyphrase: Name the campaign and the one keyphrase this row covers (e.g., 'custom framing').
• ▸Keyphrase bid: Enter the amount you are willing to pay for serving the ad on this keyphrase.
• ▸Maximum CPC: Enter the most you pay per click. It must be equal to or less than the keyphrase bid.
• ▸Observed rank: Enter the numerical label your ad received in the listing.
• ▸Observed position: Mark north (above organic), south (below organic), east (right rail), or integrated.
• ▸Competitive pressure: Note how many businesses appear to bid on this keyphrase and whether the required bid is rising.
• ▸Clicks and cost: Record clicks received and total cost at this rank and position.
• ▸Cost per acquired customer vs. profit: Compare what a customer costs to the profit on the sale. Mark whether it is profitable.
• ▸Next single change: State the one lever you will change (bid or ad text) and the date, then re-log rank and position.

1. 1Set the unit of decision first: a bid belongs to one keyphrase inside one campaign (a set of keyphrases, bids, and ads tied to a topic). Write down the topic, the keyphrase, and the ad it triggers before you type a number.
2. 2Define the bid precisely: it is the maximum you will pay the search engine per click. Your actual cost per click equals the bid, sometimes lower, never higher. Treat the bid as a ceiling, not a forecast of spend.
3. 3Anchor the ceiling to the goal of all advertising in the source: acquire new customers at a cost that still leaves a profit on what you sell them. Decide the most you can pay for a customer, then work backward to a per-click ceiling.
4. 4Expect competition to push the price. The more businesses that bid on a keyphrase, the higher the price to get your ad onto the SERP, where screen real estate is the scarce resource. Set the first bid with that demand in mind.
5. 5Enter the bid with its other settings: period of activation, language, geographic limits, and matching degree (exact to very loose). These shape which queries the bid is exposed to, so set them deliberately.
6. 6Activate, then collect historical data. The source says the right cost is hard to determine at first and that you fine-tune bids and budget once data exists. Don't judge a bid on a few clicks.
7. 7At each billing period, read the statistics the search engine provides. Adjust bids, maximum costs per period, keyphrases, and match degree. Change one lever at a time so you know which change moved the result.
8. 8Review bids per keyphrase and per ad placement. Remember that ads served on other Web sites are typically priced lower per click than SERP ads, so don't compare their costs directly.

The source's framing shop, Faster Frame, signs up for pay-per-click sponsored search. It picks the keyphrases "frames," "framing," and "custom framing," and writes an ad offering a $50 off custom framing coupon. It then sets a maximum bid of $1.00 per click. That $1.00 is a ceiling: a click may cost $1.00 or less, never more. Accounts like this go live within minutes, but the source warns that the setup hides complex questions. The shop doesn't yet know what a customer is worth or how often clicks convert, so $1.00 is a starting guess. After the first billing period it reads the click statistics and uses that historical data to fine-tune the bid and budget. The shop's goal stays fixed: acquire customers at a cost that still leaves a profit on the framing it sells.

• If the actual cost per click you're paying is below your bid, leave the bid alone, because the bid is only a ceiling and you are not overpaying.
• If many businesses bid on your keyphrase, expect to pay more for visibility, so decide whether the keyphrase is worth that cost rather than chasing position.
• If you lack historical data, set a provisional bid you can afford to lose, run it, and fine-tune only once statistics accumulate.
• If the cost of acquiring a customer through a keyphrase exceeds the profit on what that customer buys, lower the bid or drop the keyphrase.
• If you're comparing cost across placements, separate SERP clicks from clicks on other Web sites, because their prices differ.

• Treating the bid as the price you pay. The tell: you expect spend to equal bid times clicks and are puzzled when actual cost per click comes in lower.
• Setting bids from competitor prices alone, with no profit-per-customer ceiling. The tell: clicks arrive but customer acquisition cost exceeds margin.
• Judging bids too early. The tell: you change bids after a handful of clicks, before any meaningful historical data exists.
• Changing bid, keyphrase, match degree, and budget at once. The tell: results shift but you can't say which change caused it.
• Assuming one bid fits every keyphrase in a campaign. The tell: the same $1.00 sits on "frames," "framing," and "custom framing" despite different intent and demand.

• ▸Campaign topic: the topic that links this set of keyphrases, bids, and ads.
• ▸Keyphrase: the exact two-or-more-word phrase you are bidding on.
• ▸Linked ad: the headline or offer shown when the phrase triggers.
• ▸Maximum bid per click: the ceiling you will pay, entered in currency.
• ▸Affordable customer acquisition cost: the most you can spend to win one customer and still profit.
• ▸Match degree and limits: exact to loose matching, plus activation period, language, and geography.
• ▸Observed actual cost per click: from billing statistics; should be at or below the bid.
• ▸Clicks and outcomes this period: clicks plus conversions to compare against the ceiling.
• ▸Next adjustment: one lever you will change (bid, match, budget, keyphrase) and why.

1. 1Define the unit of analysis as one keyphrase-ad-landing page combination. The score attaches to the combination, not to the keyword alone. Build one row per combination before you touch any bids.
2. 2Name the three inputs the passages cite: click-through rate, ad/keyword relevance, and landing-page quality. The bid amount is a fourth factor, but you control it separately. Record all four per row.
3. 3Measure CTR per keyphrase and per ad, not per campaign. The source says to evaluate every keyphrase, ad, term in an ad, image, and display URL, and to compare cost and results for each keyphrase. A campaign-level average hides the combinations that drag the score.
4. 4Check relevance by asking whether the ad is shown only because the searcher's keyword triggered it, and whether the ad text promises what that keyword implies. If you cannot say how the ad answers the query, split the keyphrase into its own tighter ad group.
5. 5Audit the landing page against the ad's promise. Because the engine penalizes poor sites and deceptive practices, the page must deliver the ad's claim. Fix mismatches between ad and page before you raise a bid.
6. 6Match your claims to the product type. For experience and credence goods, the searcher's uncertainty is higher and quality is harder to judge. Searchers then use different query terms, so tune the keyphrase and ad to the uncertainty-reduction motive, not just the product name.
7. 7Treat bid and quality together when you act. A low-quality combination can still win position, but only at higher cost, because you cannot just buy the top spot with no clicks. Compare the cost of buying position with the cost of improving the ad and page.
8. 8Re-test one change at a time (ad term, image, display URL, page element) and judge it by the pay-off, not by taste. The source says the goal is the ads and keyphrases that pay, even if you find the wording odd or the images unattractive. Stop spending when the next dollar costs more than a dollar of sales, per the diminishing-returns note.

The clearest real case in the passages is the shift from Overture to Google. Overture ranked ads by bid price alone. Google started in 2002 with Overture's pay-per-click model but ranked ads on a quality score derived from the bid amount and click-through rate. Keyword relevancy and landing-page quality were added later. The effect is that an advertiser with a high bid but no clicks no longer sits in the top spot, or can stay there only at greater expense. An advertiser with a relevant ad and a good site gets more clicks and gains position, which also raises the search engine's revenue because users click relevant ads more. Overture later moved to second-price in 2002 and to quality-based bidding, its Quality Index, in early 2007. Apply the same logic to your account. Take two ads on the same keyphrase, where one has a top bid and a vague ad pointing to a generic page, and the other has a lower bid and an ad matched to the page. The source gives no formula, but the mechanism says the second ad can match or beat the first's position at lower cost.

• If CTR on a keyphrase is low while the bid is high, then rewrite the ad or split the keyphrase before raising the bid, because bid alone does not buy position.
• If clicks are healthy but conversions are poor, then audit the landing page against the ad's promise, since page quality is a scored factor and also drives sales.
• If an ad is triggered by a keyword it does not speak to, then remove or isolate that keyphrase. Relevance means the ad appears only when the searcher's keyword warrants it.
• If the cost needed to hold a position exceeds the revenue it generates, then stop raising bids, per diminishing marginal returns, and improve quality or cut the keyphrase.
• If a change affects the ad, page, or keyphrase, then test one element at a time and keep it only if the numbers improve, regardless of your aesthetic opinion.

• Bidding up to fix position: you raise bids on a low-CTR combination and your costs climb while position barely moves. The tell is rising cost per click with flat or falling click-through rate.
• Judging ads by taste: you keep or kill copy and images because you like or dislike them. The tell is that no ad variant has recorded metrics behind it.
• Broad ad groups with generic ads: one ad serves many keyphrases, so relevance is weak for most of them. The tell is a few keyphrases carrying the CTR while the rest drag.
• Ad-page mismatch: the ad promises something the landing page does not deliver, which is the poor-site or deceptive pattern the quality score penalizes. The tell is a decent CTR with weak conversions and expensive position.
• Campaign-level reporting: you look at account averages and miss the weak combinations. The tell is that you cannot name the cost and results for each keyphrase.

• ▸Keyphrase: the exact term that triggers the ad.
• ▸Ad text and display URL: paste the current version being tested.
• ▸Landing page URL: the page the ad sends to.
• ▸Click-through rate: clicks divided by impressions for this combination.
• ▸Relevance check: does the ad speak to the keyphrase, and is it shown only when the keyword triggers it? Yes/no plus note.
• ▸Landing-page check: does the page deliver the ad's claim, and is the site free of deceptive or poor-quality elements? Yes/no plus note.
• ▸Current bid and average cost per click: record both to see what position is costing.
• ▸Conversions and cost per conversion: results for this keyphrase only.
• ▸Next single change and decision: one test (ad term, image, display URL, or page element), then keep, revise, or cut based on the numbers.