Public research
The AI Visibility Index
Recurring public research measuring how often AI assistants recommend businesses across a defined universe, UNIVERSE, e.g. 200 South Yorkshire B2B firms across 10 sectors. Published quarterly with full methodology, downloadable raw data and confidence intervals on every figure. No business in the Index is a client, and none pays to be included or excluded.
Current edition: [Q/YEAR] · Download the full dataset (CSV) · Methodology in full →
Last updated 31 July 2026
Why this exists
Profound's authority in this category came from publishing research at scale, not from its software. The UK and regional equivalents are unclaimed.
More practically: almost every statistic quoted in AEO sales conversations comes from US enterprise data. If you run a manufacturing business in Rotherham, a study of Fortune 500 brand visibility in ChatGPT tells you very little. This Index is an attempt to produce numbers that apply to the businesses actually reading them.
Choosing the universe
Decide this before the first run and then don't change it, because comparability across editions is the entire value.
Selection criteria to publish:
- Geography: [defined area]
- Sector coverage: [n] sectors, [n] businesses per sector
- Inclusion rule: objective and checkable, e.g. Companies House SIC code plus registered address plus active trading status
- Exclusions: clients of Question.Marketing (declared, listed by name)
- Sample size: [n] businesses
State the selection rule precisely enough that a third party could reconstruct your sample. That's the difference between research and a marketing asset shaped like research.
Methodology
- Prompt set: [n] prompts per sector, designed around real buying intent rather than brand names. Published in full in the appendix.
- Sampling depth: ~[n] runs per prompt per platform per edition.
- Platforms: ChatGPT, Google AI Overviews, Google AI Mode, plus a supplementary panel across Gemini, Claude, Perplexity and Copilot.
- Statistics: Wilson 95% confidence intervals throughout. Species-accumulation modelling for competitor discovery.
- Locale: [country, language, and how location is controlled]
- Field window: [dates]
Headline findings
Populate after the first run. Structure below.
,
Finding 1, figure with interval, what it means and what it doesn't mean
Pending first edition
,
Finding 2, figure with interval, what it means and what it doesn't mean
Pending first edition
,
Finding 3, figure with interval, what it means and what it doesn't mean
Pending first edition
Sector tables
One table per sector: business, share of voice per platform with intervals, rank within sector.
| Business | ChatGPT | AI Overviews | AI Mode | Rank in sector |
|---|---|---|---|---|
| Pending first edition | , | , | , | , |
What this data cannot tell you
- It measures mention, not traffic or revenue. A business with high share of voice may be getting no AI referrals at all.
- It measures our prompt set, not every question a buyer might ask. The prompts are published so you can judge whether they resemble your buyers' questions.
- It cannot explain why a business is visible. Correlation with observable characteristics is offered where we have it, as correlation.
- It is a snapshot. These platforms change without notice, and an edition-to-edition shift may reflect a platform update rather than anything any business did.
- Personalisation and geography introduce variance we can only partly control.
Editions
| Edition | Published | Data |
|---|---|---|
| [Q/YEAR] | [date] | [CSV] |
Data is published under CC BY 4.0, reuse it with attribution.