There is no single, stable “AI ranking” to report
A CEO asks whether the company is visible in ChatGPT, Google AI and Copilot. The dashboard answers with one score: 62 out of 100. It looks precise, but nobody can explain which questions, markets, engines or answer types produced it.
That is the central measurement problem. AI answers can change with wording, location, product state, available web sources and user context. A business may be cited as an expert for one question, named as a provider for another and absent from a third. Rolling those situations into one opaque score hides the decision the data should support.
Platform reporting is uneven. Google's generative-AI Search Console reports expose impressions, pages, countries, devices and trends for supported Google experiences. Bing AI Performance reports citations, cited pages and sample grounding queries across supported Microsoft experiences. Neither covers every answer engine or business result.
My verdict: build a repeatable evidence system that keeps visibility, representation and commercial movement separate. Use it to decide what to repair. Do not manufacture attribution.
The AI Visibility Evidence Ladder
I use six layers because each answers a different owner question. Moving upward increases commercial relevance, but it also increases uncertainty. The lower layers show what the search surfaces exposed; the upper layers show what buyers and the business did afterwards.
Are we observing the right decisions?
Define buyer questions by problem, market and decision stage before collecting results.
Does the business appear?
Record named, linked, recommended or absent for each question and engine.
What evidence supports the answer?
Capture owned pages, independent sources and competing domains that receive citations.
Is the description accurate?
Check offer, audience, location, proof and limitations against current business facts.
Does visibility create movement?
Observe AI referrals, branded search, useful page journeys and suitable enquiries.
Does demand become valuable?
Review accepted opportunities, sales progress and revenue without forcing causal credit.
Start with buyer questions, not a software-generated keyword list
Select 15–30 questions a suitable customer might ask while identifying a problem, comparing approaches or choosing a provider. Include the priority countries and plain-language variations that materially change the decision. Keep the core set stable for a quarter, and document the engine, date, location and signed-in state where relevant.
Separate mentions, citations and recommendations
A mention names the business. A citation links to a source. A recommendation places the business into a shortlist or proposed action. They are not interchangeable: useful expertise may be cited without the company being recommended.
That distinction also keeps this measurement guide separate from the AI Citation Readiness System, which explains how to build citable evidence, and the wider Business Discoverability Stack, which covers access and entity clarity.
The owner scorecard: five measures and the decision each supports
| Measure | Plain-English calculation | Owner decision | Do not claim |
|---|---|---|---|
| Presence rate | Buyer questions where the business is named ÷ questions checked | Whether the category association is emerging | A permanent rank or guaranteed reach |
| Source coverage | Questions citing an owned or credible earned source ÷ questions checked | Which evidence assets need protection or improvement | That every citation recommends the business |
| Representation accuracy | Appearances with correct material facts ÷ total appearances reviewed | Whether identity, offer or external records need correction | That positive wording is objective endorsement |
| Qualified action | Suitable AI referrals, branded visits or enquiries observed in the period | Whether visibility connects to a useful buyer journey | That every direct or branded visit came from AI |
| Commercial progression | AI-referred or credibly influenced enquiries that reach accepted sales stages | Whether to expand, hold or redirect investment | Revenue causation from an unobserved answer |
Keep counts beside every rate. One appearance in two prompts is 50%, but not a reliable trend. Segment by engine, market and question type; publish the formula and sample if a board report uses a composite score.
Microsoft states that Bing citation count does not indicate placement, authority or ranking. OpenAI says ChatGPT search referrals include utm_source=chatgpt.com. These are useful pieces—not a complete attribution model.
What the evidence can and cannot tell a CEO
A repeatable prompt panel can reveal stronger presence, outdated descriptions or trusted competitor sources. It cannot estimate total audience exposure unless the platform supplies impressions for that surface.
Referral traffic is a lower-bound signal. A buyer may remember the brand and return through search or direct. Treat branded demand and assisted journeys as context, not automatic credit.
Be cautious with sentiment and “position.” Conversational answers do not consistently behave like ten blue links. Preserve the answer when a material decision depends on interpretation.
Priority pages are missing, weak or not supported by credible sources.
Align the website, profiles and independent records around current facts.
The observed prompts may lack commercial intent or the next step is weak.
Protect the cited evidence and test adjacent buyer decisions without overexpanding.
A practical monthly review and quarterly decision
Run the stable panel
Record presence, citations, description and answer evidence under consistent conditions.
Check platform data
Review available Google impressions, Bing citations, analytics referrals and branded demand.
Choose one repair
Prioritise access, evidence, representation or conversion rather than changing everything.
Expand, hold or redirect
Connect the trend to suitable enquiries and sales context before reallocating investment.
The Marketing Proof Stack separates activity from demand and pipeline. The Search-to-Revenue Portfolio tests the wider discovery investment.
See my AI growth service, growth partnership services, case studies and operator background. The next step is a scoped diagnostic—not a ranking promise.
Sources and evidence notes
Sources were checked on 20 August 2026. The framework is original ThomPerformance analysis. Priority is qualitative; no search volume, benchmark, client result or causal revenue claim is presented.
Frequently asked questions
What is AI search visibility?
AI search visibility is the observable presence of a business or source inside AI-generated answers for a defined set of buyer questions. Useful measurement records whether the business appears, how it is described, what is cited and whether suitable buyers take a next step. It is not one permanent rank.
What is the best metric for AI search visibility?
There is no universal best metric. Start with presence rate across a stable buyer-question set, then add source coverage and representation accuracy. Connect those leading signals to referrals, branded demand, suitable enquiries and qualified pipeline where evidence exists. Keep the formula and scope visible if using a composite score.
Can Google Search Console measure AI visibility?
Google announced dedicated generative-AI performance reports in Search Console on 3 June 2026 for a subset of websites. They show impressions, pages, countries, devices and trends for supported Google features—not every AI engine or proof of revenue.
Can I track traffic from ChatGPT?
OpenAI says ChatGPT search referral URLs include utm_source=chatgpt.com, so analytics tools can identify those visits. The evidence is incomplete: a buyer may later search the name or visit directly, so a missing referral is not proof of zero influence.
How often should a business review AI search visibility?
Use a monthly operational review and a quarterly business decision. Monthly checks reveal access, accuracy and source changes without reacting to every answer variation. Quarterly reviews are better for connecting visibility trends to branded demand, suitable enquiries and pipeline.
Measure evidence, not an imaginary league table
AI visibility is useful when leaders see which questions, sources and descriptions change—and whether suitable customers move afterwards. Keep the panel stable, expose the sample and separate platform visibility from commercial outcomes.
Which buyer decision would matter most if your business were accurately represented in the answer?