How to measure your brand’s visibility in AI answers.
When people ask an AI assistant which company to choose, a few names come back. Measuring AI visibility means finding out, on a schedule, whether yours is one of them, what the answers say about you, and which sources they draw on.
Published 13 October 2026 · Updated 14 October 2026 · By Visibian
Why it is worth measuring.
Search reports count impressions, clicks, visits and where your pages appear in the results. Those numbers describe a list of links. An AI answer works differently: it names a few brands, compares them, describes them and cites the pages it drew on. None of that shows up in a search report.
Measuring it shows how assistants present you to people who are deciding, and where that picture is wrong or missing. It does not replace search measurement. It sits beside it.
1. Start with the questions your customers ask.
Build the question set from what customers ask, not from an arbitrary list of prompts. Look at sales calls, support tickets and what people type into the search box on your site and your competitors’ sites.
Group the questions by the stage of the decision:
- Criteria: “What should I look for when buying a…?”
- Shortlist: “Who are the best providers of…?”
- Comparison: “A or B, which is better for…?”
- Validation: “Can I use this brand for…?”
Keep the wording the same each time you measure, and set the market and language each question is asked in. Answers change with wording and location, so a reworded question is a new measurement.
2. Measure more than mentions.
A mention is a start, and only one part of the picture. For each question, record:
- Visibility: the share of answers that name you. Asked 100 times and named 30 times is 30%.
- Position: where you appear when you are named. First is worth more than fifth.
- Share of voice: your share of all the brand names in the answers, which shows how crowded the field is.
- Sources: which pages and sites each answer cites, whether yours are among them, and which kinds of page come up most, such as news articles or documentation.
- Accuracy: whether what the answer says about you is true and current, and whether it describes the right company.
- Sentiment: how warmly you are described, and whether the answer mentions limitations or criticism.
Agree on the definitions before the first report. For example, decide whether being named somewhere in an answer counts the same as being first in a list of recommendations. Fixed definitions make changes over time easy to read.
3. Expect the answers to vary.
Ask the same question twice and you can get 2 different answers. One reading proves little. Ask each question many times, across the assistants your customers use, and put a margin of error around each number, such as “24 to 38%”. When the margins of 2 periods overlap, nothing has changed yet. When they are clear of each other, something did.
4. Compare with the right competitors.
Visibility only means something next to the alternatives a customer would consider. Include competitors at the category level, including those that solve the same need with a different product. Measure them on the same questions, the same assistants and the same definitions as yourself.
The gaps then point to work. Are competitors named on a topic where you are not? Are you missing from questions about the uses you were built for? Each gap is a page to check, or an explanation you have not yet given.
5. Look at the sources the answers use.
When an answer cites its sources, look at what they are: your own pages, retailers, review sites, industry voices, documentation or competitors’ sites. Do they explain your category, your offer and what sets you apart, accurately and recently?
Use what you find to update your own pages, to make sure accurate third-party pages exist, and to keep the facts about you the same everywhere. Sharing what you know with the sites your field reads builds the kind of presence that assistants pick up.
A cited source is evidence, not proof. An answer can cite a page and still draw its view of you from elsewhere. Use source data to decide where to look, not to explain every answer.
6. Measure, change, measure again.
Take a baseline with your question set across the assistants that matter to your audience. Find the gaps: missing mentions, wrong facts, absent sources. Make a change, to your own pages or to what you publish elsewhere. Then ask the same questions again after a fair interval and compare, margins included.
Set what you see beside search traffic, visits from AI assistants, lead quality and sales. Answers also change for reasons you do not control, such as a model update, so one change rarely explains one shift on its own. Treat it as an ongoing series of observations and adjustments.
7. Make your site easy to understand.
No measurement guarantees that an assistant will recommend you. What helps most is clear, accurate information. When you review your site, look for:
- Pages that say plainly what you offer, who it is for and what makes it different
- Straight answers to the questions customers ask
- Accurate facts, with sources and dates where they matter
- Comparisons, examples, documentation and original research that support what you claim
- An accessible, well-built site with descriptive titles, clear headings and links that are easy to follow
- The same information on your site as everywhere else that describes you
These are the same things that help search and help customers. Write for the people who are deciding, and the assistants have something accurate to repeat.
A short checklist.
- We have a written set of real customer questions.
- We ask them on a schedule, across the assistants our customers use.
- We track visibility, position, competitors, sources and accuracy.
- We read every number with its margin of error.
- We have a baseline, and we repeat the measurement over time.
- We choose improvements by customer value and evidence.
- We compare AI visibility with visits and business results.
Measured for you, every day.
Visibian Brand Intelligence asks your questions to ChatGPT, Google Gemini, Google AI Mode, Perplexity, Microsoft Copilot, Grok and Claude, every day or every week. It reports visibility, position, share of voice and sentiment, each with its margin. It also shows the sources the answers draw on, flags answers that contradict your facts, and explains the answers that left you out.