How to Evaluate an Agency’s Case Studies: Seven Checks, Applied to Our Own Records

What an SEO case study is supposed to explain

An SEO case study is supposed to explain what a business’s search visibility was before the work, what it was after, over which dates, on which surface, measured how, and what changed in between. AI Search Engine Optimization publishes eight of them under a fixed set of rules: every figure is a seven-day average of daily rank checks, every count carries its denominator, Google’s organic results and its map pack are reported separately, map-pack positions count only in the three that are visible, and the distance from the business’s address to the search point is stated when it is large. Those rules exist because the case studies an owner usually reads, and the ones ranking for this search, do not have them. The field for “SEO case studies” today is a software vendor’s list of 23 studies, an agency’s list headed by a “279% traffic explosion in 90 days,” a forum thread asking for real examples, and Google’s own success stories, none of which shows a denominator, a distance or a statement of what the result does not include. One of ours, as published: a paint protection film specialist in England, 64 to 108 of 123 tracked terms in Google’s visible three-position map pack, in the five weeks to September 5, 2026.

The forum thread is the honest one: a marketer asking for examples of SEO case studies that are not sales pages. The answer is that a case study becomes evidence the moment it can be checked, and it stays a sales page for as long as it cannot. This article is the checklist for telling the two apart, tested on the records this agency publishes, because a checklist that its author’s own work fails is not worth reading.

Read that record in full: the window runs from the week of July 30 to August 5 to the week of August 30 to September 5, 2026; the figures are seven-day averages of the same 123 terms at both ends; the terms at first place rose from 8 to 75 over the same five weeks; and the surface is the map pack, reported apart from the organic results. Every element of that sentence is one of the checks below.

The seven checks, and where most case studies fail

A credible SEO case study passes seven checks, and AI Search Engine Optimization applies all seven to every record before it is published: the count has a denominator; the window is dated and uses averages rather than single days; the surface is named, and map-pack positions are counted only where a searcher can see them; the place of measurement is stated when it is far from the business; the work that produced the change is described; the result is attributed to a business the reader could recognize by what it does; and the record says plainly what it does not claim. The table lists each check with what a passing record shows and what a failing one shows instead.

Most published case studies fail checks 1, 3, 4 and 7 at once, because a percentage with no denominator, on an unnamed surface, measured from the business’s front door, with nothing said about what stayed flat, is the format that makes any result look large. The four checks that are hardest to pass are the four that decide whether the number means anything.

Check 1 applied: every count carries its denominator

A count without a denominator cannot be evaluated, so every figure this agency publishes carries one. The cell tower lease consultancy record reads “0 to 30 organic terms in Google’s top ten” and then states that 39 terms were tracked in January and 83 by September, and that the 39 original terms moved from an average position of 82.8 to 25.0 with 14 of them in the top ten. The family law firm in Montana reads “4 to 53 terms in the visible map pack,” and the denominator is 70, the same 70 terms at both ends. A reader can weigh 53 of 70 against 30 of 83 and know they are different kinds of result; the same two records reported as “rankings up 1,225 percent” and “top-ten rankings from zero” would be impossible to compare and easy to inflate.

• Ask: how many terms were tracked, and were they the same terms at both ends of the window?
• Ask: were the terms chosen before the work or after the result, when the ones that moved were known?
• Ask: what happened to the terms that did not move, and are they in the count?

Check 3 applied: the map pack counts only where a searcher can see it

A map-pack position below third is not visible in Google’s results, so AI Search Engine Optimization does not count it, and a case study that does is counting positions no customer ever saw. Google shows three map results; a business at position seven has moved from invisible to invisible. Every map-pack figure on this site is therefore a count of terms in the visible three, or a count of terms at first place, never an average position across the whole map list, which is the figure that makes a move from 40th to 8th look like progress. The Montana family law record is the plainest example of the rule and of check 7 together: 4 to 53 of 70 terms in the visible three, 0 to 20 at first place, and the same record states that the firm’s organic positions for those terms softened over the same year while the map pack improved. It is presented as a map-pack record, because that is what it is.

• A map-pack “average position” is not a customer-facing number; positions 4 and below are not shown.
• A record that reports only the surface that improved is hiding the one that did not.

Check 4 applied: where the result was measured, and how far from the business

A local result is easiest to earn right around the business’s own address and hardest to earn far from it, so a case study should say where the measurement was taken, and this agency’s records state the distance whenever it is large. The mobile outboard repair record was measured from a Fort Lauderdale search point about 19 miles in a straight line from the business’s base in Pembroke Pines, and on that basis the terms at first place in the map pack rose from 2 to 9 and the terms in the visible three from 5 to 12, of 58 tracked. The bathtub reglazing record was measured for Los Angeles searches from a business about 90 miles away in Simi Valley with no Business Profile in the area, so no map-pack result was ever available, and the four commercial keywords moved from an average organic position of 5.0 on February 1 to 1.75 on August 27, 2026, in the organic results alone. The special education law record was measured statewide across California, against every competitor in the state, and the organic terms in the top three moved from 1 to 11 of 26 between September 2025 and September 2026. A case study that omits the place of measurement has usually measured from the easiest place there is.

Why a traffic percentage is the weakest number a case study can lead with

A traffic percentage is the weakest headline a case study can carry, because it has no denominator, no surface and no place, and a small site can post a very large one from a very small base. AI Search Engine Optimization does not lead any record with one. A 279 percent traffic increase in 90 days can be 40 visits becoming 150, most of them from a blog post about a topic the business does not sell, and nothing in the percentage tells the reader which. The ranking record, by contrast, names the terms, the count, the window and the surface, and the reader can see whether the terms that moved are the ones a buyer types before calling. Traffic and rankings are also not the number the owner feels: none of this agency’s eight records claims a measured increase in inquiries or revenue, and the hub says so in its own words, because rank tracking cannot measure them and a case study that claims them without a separate measurement is asserting, not reporting.

• Ask for the base: a percentage from a base under a thousand visits describes noise.
• Ask which terms drove it: a traffic gain on non-commercial terms is a publishing result, not a business one.
• Ask what was measured in inquiries, and how; if the answer is “rankings went up,” the inquiry claim is not a measurement.

Check 7 applied: what this agency’s own records do not claim

A case study that says what it does not claim is the one to trust, and the case-study hub on this site carries a standing statement to that effect: the records are Google ranking measurements; they do not establish a measured increase in AI citations, qualified inquiries, sales or return on investment; they do not isolate the contribution of any single change; where one surface improved and another did not, both are stated; and client identities are private. That last point is deliberate. Every business on the hub is named by what it does and where, a family law firm in Montana or a mobile outboard repair business in Central Florida, because a logo proves nothing a reader can check and a category tells the reader whether the market resembles their own. The AI visibility question is answered separately, on the AI answer benchmark page, with its own method and denominators, rather than being folded into a ranking record where it would not belong.

Using a case study in the hiring conversation

A case study is most useful when it becomes a set of questions for the agency presenting it, and AI Search Engine Optimization answers each of the seven on the strategy call before it asks for any work. Bring the record and ask what the denominator was, what dates the window covers and whether the figures are averages, which surface was measured and how the map pack was counted, where the measurement was taken, what was actually done in what order, what kind of business it was, and what did not move. An agency that can answer all seven from its own tracking data can do the same for your site; one that cannot is showing you a page built to persuade. The AI visibility audit that begins each engagement here records your own baseline under the same seven checks, so that the case study written about your business a year from now can be read the same way.