Last checked: 10 June 2026. This page describes how Fortitude Sentinel produces every score, chart and published report, including the limits of the method. We re-verify and re-stamp it quarterly. If you find an error, email us and we will correct it and note the correction.
What we measure
We measure the answers five AI engines give to buying questions: ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews. We chose these five because they are where buyer questions are actually asked, and we include Claude because several widely used alternatives do not. For each answer, we record which brands appear, how they are framed, and what sources are cited, and we keep the record over time. The unit of measurement is always the answer to a question a buyer would plausibly ask, not a keyword.
How prompts work
Every category we cover has a canonical library of buyer-intent prompts, written and owned by us. These are the questions a real buyer asks at the point of choosing: who to shortlist, what to compare, what something should cost, who is credible for a specific need. The canonical library is fixed per category and versioned, so results are comparable over time; it is the basis of all public data.
Customers can add their own custom prompts inside their account. Custom prompts and their results stay private to that customer. They are never added to the canonical library, never used in public data, and never visible to anyone else. The two pools do not mix.
How often we run
Cadence is tied to plan and published wherever the data appears. Sentinel Free includes a monthly refresh, and every chart and report we produce states the dates of the runs behind it, so you are never reading an undated number. Because AI engines vary their answers between runs, we treat each run as a sample and read movement across runs rather than from any single result.
How brands are identified
We identify brands by entity matching, not string matching. An AI engine might refer to the same company by its trading name, legal name, product name or an abbreviation, and a naive text search either misses these or false-matches unrelated terms. Entity matching resolves mentions to a canonical brand entity before anything is scored, which is also what lets us score named competitors accurately in the same answers.
What the scores mean
Sentinel reports three things, each against named competitors rather than in the abstract.
| Score | What it answers |
|---|---|
| Visibility | How often you appear in answers to your category's buyer-intent prompts |
| Share of voice | How your presence compares with named competitors in the same answers |
| Sentiment | How you are framed when you do appear, against those same competitors |
The competitor baseline is the point. A visibility number on its own cannot tell you whether you are improving or the whole category is shifting; the same number read against named rivals can. Our piece on benchmarking AI visibility against competitors covers why this matters when evaluating any provider, including us.
What the Registry is, and the two-pipeline rule
The Registry is our longitudinal record of how AI engines answer the canonical buyer-intent prompts in each category, and it is governed by one strict rule: public data comes only from our own canonical runs, never from customer accounts. We call this the two-pipeline rule. Pipeline one is the canonical library, run by us, on our prompts, producing the Registry and everything we publish. Pipeline two is customer accounts, including custom prompts, which stay private and feed nothing public.
Because the Registry is built from our own observations, it can describe brands that have never been customers, in the same way other research firms such as Ahrefs and SimilarWeb sell observations of companies that never signed up. New customers receive up to twelve months of their category's Registry history at signup, which is why an engagement with us does not start blind. The Registry's history is deepest in UK B2B categories today, with US and EU coverage expanding.
How the public indexes and reports are produced
Everything we publish is produced from aggregated Registry data and edited by a human before release. The publishing programme consists of quarterly Category Indexes, monthly movers reports and findings pieces. Each piece carries its methodology and the dates of the underlying runs. Nothing is fabricated, interpolated to fill gaps, or generated without a person checking it against the data. Public indexes in this market are not new (Profound and Semrush both publish them); ours is distinct in being the UK B2B record, and this page is the methodology behind all of it.
Data ethics and GDPR
We hold two kinds of data and treat them differently. Data about people (account holders, contacts) is personal data: it is held under GDPR and erasable on request. Data about brands is corporate information derived from public sources, namely the answers AI engines give to questions anyone could ask; it describes companies, not individuals, and is the same in kind as any published market research observation. We do not use customer account data in anything public, per the two-pipeline rule above.
Limitations
Every method has limits and these are ours, stated plainly. AI answers vary: the same engine can answer the same prompt differently on different days, so single results mean little and trends mean more. Our measurements are samples, not censuses: we observe a structured slice of engine behaviour, not every answer every engine gives. History depth varies by category: the Registry is deepest in UK B2B categories today, and newer categories carry shorter histories, with US and EU coverage expanding. And attribution is never absolute: movement in a score reflects engine updates and market activity as well as anyone's work, which is why we score against named competitors and publish dates on everything.
For how to use measurement like this in a buying decision, see tool, agency or both, the ten questions to ask any GEO agency, our UK B2B roundup, and how to track whether AI is recommending your business. The case for us specifically lives on why Fortitude, not here.
FAQ
Which AI engines does Fortitude Sentinel track? Five: ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews. Brands in their answers are identified by entity matching and scored on visibility, share of voice and sentiment against named competitors.
How is AI visibility measured? By running a fixed, versioned library of buyer-intent prompts against AI engines on a regular cadence, recording which brands appear in the answers and how they are framed, and reading movement across runs rather than from any single result, because engine answers vary day to day.
Does Fortitude use customer data in its public reports? No. Public data comes only from our own canonical prompt runs under the two-pipeline rule. Customer accounts, including custom prompts and their results, stay private and are never used in the Registry, the Category Indexes or any published report.
Is the Registry GDPR compliant? Personal data about people is held under GDPR and erasable on request. The Registry itself records brand-level information derived from public sources, which is corporate data about companies rather than data about individuals.
How far back does Registry history go? New customers receive up to twelve months of their category's history at signup. Depth varies by category: it is deepest in UK B2B categories today, with US and EU coverage expanding.
See the method applied to your own category. The free AI Visibility Check runs your category's buying questions across the five engines and shows you the result, with the dates attached, exactly as this page describes.