AI Brand Visibility: How to Measure It Without Fooling Yourself
AI brand visibility is how often, and how prominently, AI platforms like ChatGPT, Perplexity and Claude name your brand when someone asks about your category. It is easy to measure badly. Ask questions that contain your name, pool every platform into one number, and you get a figure that looks healthy and tells you little.
This guide covers the four measurements that hold up and three ways they get inflated. The worked example is our AI visibility audit of Nike, Adidas, Puma and Under Armour, run on 15 October 2026.
What AI brand visibility measures
An AI answer to "best soccer cleats" is a short list, sometimes with links. Visibility is whether you are on it, where, against whom, and whether it sends the reader to you:
| Measurement | The question it answers |
|---|---|
| Mentioned | Did the answer name you at all? |
| Named first | Were you named before the competitors you track? |
| Share of brand mentions | Of all mentions of you and those competitors, how many were you? |
| Links your site | Did the answer cite a page on your domain? |
Each one means something only when you read it per platform and per question type. A total across platforms hides which platform moved. A total across question types hides whether AI found you or was simply asked about you.
Our audit asked five questions a shopper types without naming a brand, such as "best running shoes 2026", on ChatGPT, Perplexity and Claude: 15 answers, every brand counted on the same ones. Gemini is left out because a token limit on our side cut its answers short, and Google AI Overviews were not in this run.
Read memory answers and search answers separately
ChatGPT and Claude answered with web access off, from what the model learned in training. Perplexity searched the web live.
| Platform | How it answered | Nike named first | Under Armour mentioned | Under Armour's share of brand mentions | Links to any of the four brand sites |
|---|---|---|---|---|---|
| ChatGPT | Memory | 5 of 5 | 2 of 5 | 6% | None (no links at all) |
| Claude | Memory | 5 of 5 | 4 of 5 | 12% | None (no links at all) |
| Perplexity | Live search | 3 of 5 | 3 of 5 | 13% | 1 of 75 citations (nike.com) |
The two memory platforms named Nike first in all ten answers. On the search platform, Adidas led two of five, and Nike was missing from "best running shoes 2026" entirely. Pool the three and that disappears.
It matters because you can affect one and not the other. A memory answer is frozen at the model's training cutoff: the ChatGPT model we queried, gpt-5.6-luna, lists a "Feb 16, 2026 knowledge cutoff". Nothing you publish after that date reaches its memory answers until a new model ships. A search answer is built from pages fetched when the question is asked. That is where it counts whether AI crawlers can reach your site and what your pages say.
One platform can give both kinds of answer. OpenAI's help page on ChatGPT search says "ChatGPT may search the web automatically when your question would benefit from current information." Record which kind of answer you got, not only which platform gave it.
Links follow the same split. ChatGPT and Claude cited nothing because they did not search, so a citation rate pooled across all three divides by answers that could never link to anyone.
Measure on unbranded questions
A question that names the brand guarantees the brand gets mentioned. We also asked three branded questions per brand, such as "who is the CEO of Under Armour" and "Under Armour vs Nike". Every brand was mentioned in all 9 of its branded answers on these three platforms. On the unbranded questions, Puma and Under Armour were mentioned in 9 of 15.
Links inflate the same way. Perplexity linked to underarmour.com 14 times in its three branded answers, and 0 times across 75 citations in the five unbranded ones. For Puma it was 34 and 0.
Branded questions test whether AI describes you correctly. Unbranded questions test whether AI finds you. Mix them and a brand nobody recommends can still post a high mention rate. Branded questions belong in an accuracy check, like the one in what ChatGPT says about your brand.
Treat a fixed prompt set as a sample
Your prompts are not what the platform searches for. The same OpenAI page states that "ChatGPT search typically rewrites your query into one or more targeted queries that it sends those providers." Google says AI Overviews and AI Mode "may use a 'query fan-out' technique — issuing multiple related searches across subtopics and data sources — to develop a response."
So no prompt set sees exactly what a buyer sees. A fixed set gives you a repeatable sample: the same questions on the same platforms, dated, so this month compares with last month. Read each result as a finding about that question. Under Armour was absent on all three platforms for "best running shoes 2026": a fact about one question, not a verdict on its running range.
Use the set to measure, not as copy to write toward: the platform searches for its own rewritten queries, not yours.
Use a blended score as a summary, then read the counts
A single 0–100 score is useful: it shows at a glance whether things moved between audits, and 99Visibility shows one. But it summarises the counts above, and the counts are where the finding is.
Pooling carries a hidden weight. ChatGPT's unbranded answers ran about 9,100 characters each, against about 3,700 for Perplexity and 2,300 for Claude. They also named the brands far more often: of the 192 mentions of the four brands in the 15 answers, 118 came from ChatGPT. Under Armour's pooled share of 8% sits close to its 6% on ChatGPT, and well under its 12% on Claude and 13% on Perplexity. The pooled number mostly describes one platform.
So read in this order: the score for direction, the per-platform counts for which platform moved, and the answers themselves for why. For the inputs behind our score, see how we calculate AI visibility metrics.
What to look for in AI brand visibility tools
Before you pay for one of the AI brand visibility tools on the market, check that you can see what its number is made of:
- Which platforms, and how each one answered. Every answer should say whether it came from memory or a live search.
- Raw answers you can read. If you can't open the answer, you can't check the count.
- Dated model versions. Answers change when the model does, so an undated result can't be compared with the next.
- Competitors measured on the same answers. Share of mentions compares like with like only when every brand is counted on the same answer.
- Accuracy checks against your facts. A mention with a wrong fact is not visibility you want. Checking by hand, we found Claude writing "UA owns popular fitness-tracking apps" and naming MyFitnessPal, which Under Armour sold to Francisco Partners in December 2020.
- Crawler access checks. Your robots.txt is half of it. OpenAI's ChatGPT search page says: "To make a website eligible for inclusion, allow OAI-Searchbot to crawl the site and confirm that the website host or content delivery network allows traffic from OpenAI's published searchbot IP addresses." A firewall can block what robots.txt allows.
Be wary of any tool that sells markup, files or phrasing to raise your score: none has credible evidence that it makes AI favour a page. What holds up is being reachable and being right. See how LLMs prioritise information about brands.
99Visibility audits five AI platforms (ChatGPT, Gemini, Perplexity, Google AI Overviews and Claude), tracks your competitors alongside you, and checks what the answers say against the brand facts you enter. Your first audit is free, with no credit card, and covers ChatGPT and Google AI Overviews. See the plans for all five platforms. Inside the dashboard, Agent V reads your audits and your site and drafts the fix, such as page copy for you to paste. It never edits your site itself.
FAQ
What is AI brand visibility?
How often, and how prominently, AI platforms name your brand when people ask about your category. It comes down to four counts: mentioned, named first, share of brand mentions and links to your site.
How do you measure brand visibility in AI search?
Ask each platform a fixed set of questions that don't contain your name. Record the four counts per platform and whether each answer came from memory or a live search. Date every run and keep the questions the same.
Is one AI visibility score enough?
As a summary, yes: it shows direction between audits. It can't show which platform or question moved it, so read the per-platform counts before you act.
See where AI names you, and where it doesn't. Run a free AI brand visibility audit: no credit card.
Want more of our research in Google Search and AI Mode? Add us as a preferred source.
Get your AI visibility score
One number from 0 to 100: how often AI mentions you, how accurately, in what tone, and whether it cites you, scaled by the market weight of the platforms that name you.
Start Free Audit46/100
Named on 3 of 5 platforms× 0.76 coverage
Related Articles
We Ran an AI Visibility Audit on Nike, Adidas, Puma and Under Armour. Here's What AI Got Wrong
We asked ChatGPT, Perplexity and Claude the questions sportswear shoppers ask, then checked the answers by hand. Nike was named first in 13 of 15 answers. Under Armour was named first in none, and Claude says it still owns MyFitnessPal.
GEO Metrics: How to Measure AI Visibility, Share of Voice and ROI
The GEO metrics worth tracking, how to measure each one, what no tool can measure, and how to compare competitors and estimate ROI without inventing numbers.
LLM Brand Visibility: How AI Models Decide What to Say About You
An AI answer about your brand passes four stages: memory, crawler access, retrieval and generation. What each AI vendor does at each one, and the stage you control.