A Claim Evidence Checklist for SEO, PPC, and AI Search Case Studies

Marketing case studies fail in a predictable way. Someone writes a real result honestly, then it passes through a summary, then a slide, then a headline, and by the time a buyer reads it the baseline has gone, the period has gone, the method has gone, and what remains is a percentage with no way to check it. The work was real. The proof is not.

This is the evidence checklist we apply before any performance claim goes on a page. It covers SEO, PPC, local and AI-search claims, the review workflow we use in a Canadian context, and the structure of a published study. It contains no search volumes, rankings, traffic figures, campaign results or AI visibility numbers: fresh Search Console data was unavailable when this was written and stale figures were not reused.

The governing principle in Canada

The Competition Bureau's primary guidance on performance claims is straightforward in its direction: a performance claim must be substantiated by adequate and proper testing before the claim is made. The sequence is the substance of it. Evidence assembled after a claim has been published, in response to being challenged, does not satisfy the principle.

Two practical consequences follow. First, the substantiation record has to exist as an artefact, not as a recollection. Second, the claim you publish has to be no broader than the record supports, which usually means writing a narrower sentence than the marketing instinct wants. This article summarises the principle for marketing practice and is not legal advice; for a specific claim on a specific page, take advice.

The claim ledger

Before writing a study, list every assertion it will make, each with its source, the date the source was checked, and the person who verified it. Claims with no source do not get written; claims sourced to a third party get attributed rather than adopted.

The ledger's value is that it surfaces the weak claims while they are still cheap to remove. It is also the document you want to exist if the claim is ever questioned. We keep a fuller treatment in our marketing claim ledger guide, and the library of reusable proof artefacts in our marketing proof library.

The seven-part evidence test

Apply all seven to every figure. A figure that fails any one of them gets rewritten or removed, not softened with a hedge.

Channel-specific checks

The seven tests are universal; what counts as adequate evidence differs by channel.

SEO claims. Ranking claims need the query, the location, the device, the date and the measurement method, because position varies across all of them. Traffic claims need the property, the filter set and confirmation that tracking did not change during the period. Any claim spanning a known platform update should say so rather than implying the work caused the whole movement.

PPC claims. Cost and conversion claims need the conversion definition, because a redefinition alone can move a cost-per-acquisition figure dramatically. State whether brand and non-brand were separated, whether budgets changed, and whether the period includes a bidding or platform change.

Local claims. Profile actions and map visibility vary by the searcher's location, so a local claim needs the measurement geography stated. Exposure metrics such as views belong in an exposure lane and never stand in for calls, enquiries or booked work.

AI-search claims. These need the most caution. Appearance in AI-generated answers varies by query, session and surface, and Google's own guidance on people-first and AI-generated content is about content quality rather than a visibility guarantee. Describe what was observed, on which queries, on what date and by what method. Avoid share-of-voice percentages that imply stable measurement. Our GEO work is scoped on that basis.

Third-party reporting stays labelled

Industry survey and news reporting is useful context and is not your evidence. A September 18 Search Engine Journal survey report, for example, is third-party reporting: it can be cited as what that source reported, on that date, with its own methodology, and it cannot be converted into a claim about your results or about the market generally.

The rule we apply is that the sentence must carry its source. "Search Engine Journal reported on September 18 that ..." is publishable. The same figure with the attribution dropped is not.

Structure of a published case study

A consistent structure makes the omissions obvious, which is the point.

The limits section is the one that earns trust with experienced buyers. A study that names its own confounds reads as work; a study with no limits section reads as marketing. Our own portfolio is written to this structure.

The review workflow

Four passes, in order, each with a different person's attention if possible.

First, the practitioner assembles the source records and drafts the claim ledger. Second, a reviewer who did not do the work checks each figure against its export and fails anything not reproducible. Third, the client approves in writing what may be published, including whether they are named. Fourth, a final read against the seven tests, with the specific question: is any sentence broader than the record supports?

Recheck published studies on a schedule. Claims age: a figure that was accurate two years ago and is presented in the present tense becomes a misrepresentation without anyone editing a word. Our content analyzer is useful in the final pass for spotting the vague intensifiers that tend to creep in where a figure was removed.

The compact checklist

Before publishing any performance claim, confirm: the baseline exists and is recorded; the metric is defined in the study itself; both periods are named and comparable; the method and tool are stated; concurrent changes are disclosed; permission is in writing; the source export is filed and retrievable; third-party figures are attributed with their date; nothing is stated in the present tense that is no longer true; and the study says plainly what it does not prove.

What this checklist will not do

It will not turn weak results into strong ones, and it will not substitute for legal advice on a specific claim. What it does is make the difference between a claim you can defend and one you cannot visible before publication, which is the only point at which the difference is cheap.