What share of AI answers means
It is the AI search version of an old question: when buyers ask, how often are we in the answer?
The unit is the answer. One question asked five times gives five answers, and each one either names your brand or does not. Your share is the named answers as a percentage of all of them.
In our own reports the closest measure is Share of Memory. You will also see it called share of voice in AI search, or brand share in AI answers. Some versions divide by competitor mentions instead of answers, so check what is being counted.
How to measure it in five steps
- Fix the question set. Write the questions a buyer asks before choosing a supplier in your category. Leave your brand name out of them. Our starting point is ten. Keep the list the same every time you measure.
- Ask each chatbot several times. Chatbot answers change from one run to the next, so one answer is one reading. Google's AI Overviews and AI Mode give one answer per run, so repeat the whole set on a schedule instead. We ask the chatbots each question several times and Google's AI surfaces once per run (our methodology).
- Count the answers that name you. A clear yes or no for each answer. Decide your rule for near misses, such as a product name without the company name, and keep to it.
- Divide by the answers you collected. Named answers divided by all answers, as a percentage. State both numbers beside it.
- Keep each platform separate. One figure for ChatGPT, one for Claude, one for Google AI Mode, and so on.
A worked example
These numbers are made up to show the sum. They are not from a real brand.
- 10 questions, each asked 5 times on one chatbot: 50 answers.
- Your brand is named in 12 of them.
- Share of AI answers on that platform: 12 out of 50, which is 24%.
Write it as "24% (12 of 50 answers, 10 questions, November 2026)". A percentage with no count behind it cannot be checked or repeated.
Named, recommended and cited are three counts
An answer can involve your brand in three ways. Count them separately.
- Named. Your brand appears in the answer. This is share of AI answers.
- Recommended. The answer puts you forward as a choice, not only as a passing mention. We report this as Recommendation Rate.
- Cited. One of your pages is in the answer's sources. A page can be cited while the brand goes unnamed (SearchIntel Study 02).
Four mistakes that distort the number
- Blending the platforms into one score. In SearchIntel Study 01, covering 100 UK buying questions in June 2026, 46.1% of recommended brands were named by only one of six platforms. An average hides where you are missing.
- Comparing platforms as if they were alike. In the same study Claude named 6.2 brands in an average answer and Google AI Overviews named 3.3. Our reading, not a finding of the study: a longer list has more places on it, so a brand could show a higher share on that platform for that reason alone.
- Putting your name in the question. "Is [brand] any good?" will usually name you, because your name is in the question. Mixing those questions with category questions inflates the figure. Keep them in separate sets.
- Comparing two brands on different question sets. A share only means something against the same questions. Measure competitors on your list, in the same runs.
How to read the result
This part is our practice.
- Read it by question. The total tells you less than the list of questions where you are never named. Those are the gaps to work on.
- Read who was named when you were not. Where an answer names other brands, count them the same way.
- Read it over time. Answers move on their own. In the first travel edition of our tracker (fixed UK travel questions, data to 28 September 2026), mean name overlap between eligible consecutive answer pairs was 48% for Google AI Overviews (117 pairs), 47% for AI Mode (139 pairs) and 55% for Gemini (80 pairs). Overlap is the names in both answers as a share of the names in either, averaged across the pairs. Those figures do not cover ChatGPT. Compare the same question set month to month before calling a change real.
Where to start
For one platform and a few questions, do it by hand: see how to check if ChatGPT recommends your business. For a first reading across platforms, our free AI brand checker runs up to five of your questions on five platforms and shows which of them named you. For the full count every month, see AI visibility monitoring.
Frequently asked questions
What is a good share of AI answers?
It depends on the questions and the platform, so a single target would mislead. Compare your share with competitors on the same questions, and with your own figure last month.
Is share of AI answers the same as share of voice?
It borrows the idea of share of voice. Here the total is answers collected, counted per platform, not competitors' mentions.
How many questions do I need?
Our practice is to start with ten, which shows where the gaps are. A larger set gives a steadier figure. Whatever the size, keep the set fixed so each month compares with the last.
Does being cited count as being named?
No. A citation is a link to your page in the answer's sources. Being named means the brand appears in the answer itself. Count them separately.