How to Report AI Search Visibility Without Overclaiming
Teams increasingly want to understand and report their AI search visibility, meaning how their websites appear in AI-assisted discovery. A useful report separates observable conditions, such as crawl access, index eligibility, website changes, and available referral or search data, from outcomes controlled by external systems. Record what was tested, what changed, and what the evidence can and cannot show.
Peppermint Global describes The Carcass as the machine-legible layer that connects the whole system and makes its meaning intelligible to AI, a deliberately engineered architectural layer grounded in design, linguistics, semiotics, and information structure. For reporting, use the Carcass Method to document what was reviewed and which changes were made, not as a universal score or a prediction of AI citations.
What Each Metric Can and Cannot Tell You
For Google Search reporting, use the dedicated Generative AI performance report in Search Console, which reports impressions from AI Overviews and AI Mode. Google states that the report rolled out worldwide as of August 31, 2026; some properties may not show it if they have insufficient impressions.
Do not ignore established SEO metrics or rely on a single third-party AI visibility score. AI responses can vary by query, location, time, platform, and system version. Treat manual answer checks as a sample, record the exact conditions, and avoid presenting them as universal market share or proof that a website change caused a result.
What to Check Before You Report
Before reporting any results, check the site against the four layers of the Carcass Method: Surface, Structure, Data, and The Carcass underneath them. Then report outcomes separately.
Surface
Review whether each page makes its subject, audience, purpose, and key terms clear to a person, and whether it answers real customer questions with specific, verifiable information. Record the exact wording that needs clarification and any key claim that only appears in images or video.
Structure
Check that each important topic has one clear page, that pages link to their supporting evidence with descriptive anchors, and that visitors and crawlers can move from a claim to its proof. Note orphan pages and pages that compete for the same question.
Data
Check whether the company, products, founders, locations, and related offerings are described consistently across relevant pages and public profiles. Where names or facts differ, identify the source of truth and document the correction rather than assuming an AI system will reconcile conflicting information.
The Carcass
Check whether the meaning of the other three layers reaches machines intact. Confirm that important pages can be crawled and indexed, return expected status codes, keep key text in the initial HTML, and express the same entities and relationships in their markup and structured data. Record the URLs tested and the date; technical eligibility is not proof that a page will be selected for an AI answer.
Outcomes, Reported Separately
Outcomes are not a fifth layer; they show whether the four layers are working. Repeat a stable set of answer checks and record available referral traffic, search performance, qualified leads, and conversions where appropriate. Note the platform, query, date, and observed result. These checks are directional observations, not a universal score or a representative measure of every AI response.
Build a Baseline Before Setting Targets
Start with a well-defined set of key questions that buyers might have. The questions should cover the category, the problem that the product is solving, comparisons with alternative solutions, the details of implementation, pricing where this is available, and common objections. Use actual questions that customers have asked during sales calls, in support tickets, through search traffic, and as part of product research, not just questions that would make the brand appear favorable.
Record the date, the platform, the exact wording, the market where relevant, and what was returned. In which results did the company appear and with what description? Which sources were used and did they mix up the company with another entity? Run the same set of questions over time. This is not a universal standard, but it gives the team a clear internal benchmark.
Keep the set stable so results can be compared over time, and add new questions when customer needs or the product change. One good answer does not mean the brand will keep appearing, and one missing result is not a failure; these systems can return different answers on successive runs.
Build a Report Around Evidence
| Report Field | What to Record | Why It Matters |
|---|---|---|
| Test conditions | Date, platform, prompt/query set, market, and tool | Makes comparisons interpretable |
| Implementation changes | Pages, templates, technical fixes, and content changes | Shows what the team directly changed |
| Observed signals | Search performance, available referral data, and repeatable answer observations | Separates different types of evidence |
| Limitations | Missing data, changing prompts/models, and attribution gaps | Prevents conclusions beyond the evidence |
| Next action | Priority, owner, and review date | Turns reporting into a work plan |
Use a Consistent Review Method
For each reporting period, record the page or query reviewed, date, platform or tool, observation, evidence link, action taken, and next review date. Keep the test conditions consistent enough to compare periods, and note any changes to the site, tools, or query set.
For example, a report might state that three product pages were changed to make pricing terms and integration requirements clearer, two crawl issues were resolved, and the same 20 prompts were tested before and after the release. It should record the dates, platforms, and observed responses, then note that this small test set is directional evidence, not a representative measure of all AI search activity.
Report Changes, Limitations, and Next Steps
Separate implementation indicators from outcomes. Implementation indicators include resolved crawl blocks, improved internal links, clearer service definitions, corrected facts, and important copy being available in HTML. Outcomes may include organic performance, qualified leads, conversions, and observations from a repeatable AI answer test set. State the source and date for every metric.
List every change made during the measurement period. When the site structure, content, tracking setup, and test prompts all change at once, it is hard to attribute an outcome to any one change. If changes roll out in stages, record the date of each major update.
Avoid unattributed percentages such as “AI-ready websites are cited 80% more often” unless a relevant study explains the sample, timeframe, comparison group, and measurement method. If no credible benchmark fits the site, report a transparent baseline and the changes observed instead.
A Practical Reporting Template
Whatever template the team uses, name the source for each item rather than relying on a summary label. For example, distinguish a crawl test from an indexing observation, an organic-search click from a referral session, and a manually observed AI answer from a measured website visit. These signals answer different questions and should not be combined into a single total without a clear, defensible method.
When the team cannot access a reliable data source, state that limitation plainly. Do not estimate missing data as though it were measured. If a third-party platform provides a visibility score, document its methodology, query sample, update frequency, and known limits before comparing it with another tool or period. A change in a vendor’s model or sampling method may affect the score even when the website has not changed.
Separate the report’s findings from its recommendations. A finding describes evidence. For example, a key page returns an unexpected status code, a product name differs between pages, or a test prompt returned a different answer on a later date. A recommendation identifies the next action, owner, priority, and how the team will verify completion. This makes the report useful to designers, developers, content owners, and business stakeholders, and keeps expectations for any single fix realistic.
Review cadence should reflect the site’s release schedule and business needs. Re-test after significant changes to templates, navigation, product information, or rendering behavior, and use a consistent periodic review for ongoing monitoring. Keep a record of the exact test conditions so that future comparisons are meaningful.
Conclusion
A useful AI-search report does not collapse several different signals into one impressive-looking number. It records the site conditions reviewed, the tests performed, the data source, the period covered, and the limits of the evidence.
The Carcass Method gives teams an architectural way to think about machine-legible meaning. The practical next step is to make that idea testable: define the checks, document the evidence, measure a baseline, and revisit the same questions after meaningful changes. That approach gives a team a clearer view of progress and keeps the framework honest about what it can show.
For help improving website structure and AI-search readiness, explore Peppermint Global’s AEO-optimized website service.
FAQ
What Are AI Search Readiness Benchmarks?
AI search readiness benchmarks are defined checks across the four layers of the Carcass Method: Surface, Structure, Data, and The Carcass. They give a team a repeatable internal baseline for reporting, not a universal industry score.
Can a Website Guarantee Citations in AI Answers?
No. Technical eligibility and good content do not guarantee that a search engine or answer system will crawl, index, retrieve, or cite a page.
Which Metrics Should a Team Track?
Track findings across the four layers, then report outcomes separately: AI impressions, organic performance, referral traffic, repeated answer checks, and relevant business outcomes. Choose metrics based on the site’s goals and available data.
Does the Carcass Method Provide a Public AI Readiness Score?
No. The Carcass Method is an architectural framework, not a public or universal score. A team can build an internal scorecard from defined checks, evidence, and a transparent rating scale, but it should present the result as its own internal measure.
