AI Visibility Index
How the AI Visibility Index is measured (version 1)
Version 1 of the AI Visibility Index method: buyer questions are put to ChatGPT, Perplexity and Google AI Overview, every answer is stored verbatim, and a company's visibility is the share of category answers that name it. It covers 144 answers about skincare brands in Canada and 166 answers about wealth management firms in Canada, collected up to 6 October 2026.
310 answers
Data as of MethodologyCorrections
What is measured
The index measures what AI engines say. For each category and country, it records which companies ChatGPT, Perplexity and Google AI Overview name when asked the questions buyers ask, and which sources they cite. It does not measure the companies: no figure here rates a company's products, service, advice, conduct or financial position.
Questions
Each category has a fixed set of questions written to match how buyers ask an AI assistant for a recommendation, grouped into topics. A question that names no company is a category question; only these count toward visibility. A question that names a company is a brand search for that company and is reported on its own. The full question set for every category is a downloadable file.
| Category | Category questions | Brand searches | Category answers counted | Collected | Not counted (empty) |
|---|---|---|---|---|---|
| Skincare brands, Canada | 50 | 0 | 144 | 6 October 2026 to 6 October 2026 | 6 |
| Wealth management firms, Canada | 28 | 8 | 166 | 2 October 2026 to 3 October 2026 | 2 |
Engines and collection
Questions are run through Searchable, a third-party monitoring platform, against ChatGPT, Perplexity and Google AI Overview, with each question's location set to the country it is about (or to the city it names). Each answer is stored exactly as returned, with its timestamp, the engine, the question as sent and every link it cites. The collector does not report the engine's model version, so the access date stands in for it. Each study is a snapshot: once collected, its answers are frozen and are not re-run.
How a mention is counted
- The collector extracts the companies each answer names. Those names are then resolved to one canonical company each (see entity resolution).
- An answer counts once for a company, however many times it names it.
- Position is the order in which the answer text first names each company. The company at position 1 is named first.
- An answer with no text is not counted, and is listed above as not counted.
Entity resolution
The engines use many strings for one company: a legal suffix, a former name, a division's name or a short form. Every string is mapped to one canonical company in a reviewed alias file kept under version control, with the reason for each merge. Distinct businesses that share a parent stay separate. When a merge retires a company's page, its old address redirects to the company it was merged into. Merges are listed in the corrections log.
Definitions
| Figure | Definition |
|---|---|
| Visibility | Category answers that name the company, divided by all category answers, to one decimal place. |
| Rank | Companies ordered by visibility, then by name. Companies with the same visibility share a place ("joint 56th"). |
| Share by engine or topic | The same ratio, counted only over that engine's or that topic's category answers. |
| Named first | Category answers in which the company is at position 1. |
| Co-occurrence | Category answers that name both of two companies. |
| Share of mentions | A company's mentions over all company mentions, one mention per company per answer. |
| Citations | Links in an answer, counted by registrable domain. Own-site citations are those to the website of a company named in the category. |
| Engine agreement | Category questions on which every engine names at least one company in common. |
What gets a page
A page is published only where the data can answer it. Below these bars no page is generated.
| Page | Minimum evidence |
|---|---|
| Company | Named in at least 1 category answer |
| Question | At least 3 engine answers |
| Topic | At least 3 category questions |
| Engine | At least 3 category answers from that engine |
| Head-to-head | Both companies named in the same answer at least 3 times |
| City | At least 8 category answers tied to the city |
| Source | Cited at least 10 times across all categories |
Reproducibility
Every answer behind a published figure is stored with its identifier, and each question page lists the identifiers of its answers. Any figure on the site can be recomputed from the stored answers.
Known limitations
- AI answers vary from one run to the next. A study is a sample on the dates shown, and a figure from it describes those answers on those dates.
- Each category has a few dozen questions. A company named in one or two answers can move several places with a single answer.
- The question set shapes the result. A company strong on questions the set does not ask will look weaker than it is in practice.
- Company names are extracted by the collector. A company it fails to extract from an answer is not counted for that answer.
- Engine versions are not reported by the collector, so a change of model inside a collection window cannot be detected.
Corrections
Errors in entity resolution, transcription or citation attribution are corrected within 5 business days of a report to Anton Sopov at [email protected]. The process and every correction made are on the corrections page.
Change log
- Version 1, 9 October 2026: first published version.