How Findable measures AI visibility
We ask AI systems the kinds of questions a real customer might ask when looking for a local business, record which businesses are named and which sources are cited, then use the same measurement approach again later to see what changed.
The short answer
How do you measure whether AI visibility has actually improved?
By asking the same questions twice and counting. A fixed set of customer-style questions is approved and frozen before any work starts. Each is asked on the scored engines, ChatGPT and Gemini, across the planned baseline runs, with Google AI Overview recorded where Google shows one. For every answer, Findable records whether the business was named, which other businesses were named, and which sources were cited.
Four weeks later the identical frozen set is asked again, on the same engines, and recorded the same way. Improvement is the difference between the two counts of how often the business was named, like for like. A question that was reworded, added or dropped is a new measurement, not a comparison.
One AI response is never enough, because answers vary between runs and between engines. That is why the measurement is a set of questions and repeated observations rather than a screenshot, and why the engine is recorded with every result. The seven steps below are that process in order.
The measurement pipeline
From a question to a comparison
Seven steps, in the order they happen. The same set of questions goes in at the start and again at the end, which is what makes the comparison a measurement.
Each step is described in full in the sections below.
-
Step 1
Choose customer-style questions
Built around the business type, its location and the services or specialisms supplied for the business.
-
Step 2
Approve and freeze the set
A paid client reviews the questions first. The approved text and order are frozen for the baseline.
-
Step 3
Run the same questions on the engines
ChatGPT and Gemini are the scored engines, with Google AI Overview included where Google showed one.
-
Step 4
Record named, cited, listed
Whether the business appears as a recommendation, which sources the engine relied on, and where it is present.
-
Step 5
Record the other names and sources
The engine is recorded with the result, so which system said what is never blurred into one score.
-
Step 6
Make the changes
Four weeks pass between the baseline and the remeasurement.
-
Step 7
Replay the exact approved set and compare
The remeasurement replays the frozen baseline questions so the before-and-after comparison is like for like.
What we measure
Named, cited, listed
Three signals, recorded separately, because one does not imply the next.
- 01 Named
- A business is named when it appears as a business recommendation in an answer to one of the measured customer questions. The implementation records that signal from the answer and, where applicable, a source title that names the business.
- 02 Citation
- A citation is a source returned with, or relied on while producing, an answer. Citations tell us what the engine read; they are not automatically recommendations.
- 03 Listed
- A business is listed when it is present on a source such as a directory or website. Being listed can make a business available to be read, but it does not by itself mean the business was named.
The method, in detail
How the measurement is built
-
01
How questions are chosen
Questions are built around the business type, its location and the services or specialisms supplied for the business. They are phrased as customer-style searches, with local questions naming the relevant town where a local scope is being measured. The question generation includes safeguards for one clear intent, the real trade, the supplied place and non-invented services. We do not publish a claim here about a fixed question count or a universal weighting scheme because those vary by measurement purpose.
-
02
AI systems measured
The research describes measurements from ChatGPT and Gemini, with Google AI Overview included where Google showed one. ChatGPT and Gemini are the scored engines used in the core audit reporting. Availability and returned results can differ between systems, so the engine is recorded with the result rather than treated as a single interchangeable score.
-
03
Client baselines and remeasurement
A paid client reviews an approved question set before the baseline starts. That approved text and order are frozen for the baseline measurement, which runs the same set across the planned baseline runs. Four weeks later, the remeasurement replays the frozen baseline questions so the before-and-after comparison is like for like. A future question change is a new measurement cycle, not a silent edit to the existing comparison.
-
04
Why results can vary
AI answers are not fixed search-result positions. An answer can vary between runs, engines can return different sources, and Google AI Overview is not present for every query. A citation shows a source in the returned evidence; it does not guarantee that the business will be recommended, contacted or chosen.
Google’s AI Overviews are recorded rather than scored, because Google does not show one for every query. How that recording works, and how it relates to conventional Google search, is on the Google AI visibility page.
The research dataset
What the public research describes
The public research currently describes 941 UK businesses across 17 trades and 223 towns, with 61,848 individual citations logged. The figures are Findable's own measurements. They describe what was observed in this dataset; they are not a universal ranking of every business or every AI answer.
The public research does not currently publish a complete collection date range or a formal data dictionary. We will add those details when the underlying records support them.
- UK businesses audited
- 941
- trades
- 17
- towns
- 223
- AI citations logged
- 61,848
Limitations
What this data does not prove
- L1 Observational findings do not prove that one factor caused an AI answer.
- L2 A listing or citation is not the same thing as a recommendation.
- L3 Improved visibility does not guarantee enquiries, leads or sales.
- L4 AI visibility is not a permanent position; engines and sources change.
See where your business stands
The public findings explain the method. The free check applies the same kind of question to your business and shows what AI says today.