Ask ChatGPT "what’s the best product analytics tool for a seed-stage startup?" and you get a short answer that names three or four products. AI visibility is whether your product is one of them, how often that happens across the questions your buyers ask, and which sources the assistant points to when it answers.
This guide defines the term, explains why it matters for a small company, and draws a clear line between what you can measure and what nobody can. It ends with two ways to measure it: by hand this week, or with a tool.
AI visibility, defined
AI visibility is the share of relevant AI answers that mention or cite your brand. "Relevant" means answers to questions a potential buyer would ask, such as "best CRM for a two-person agency" or "Notion alternatives for engineering docs". "AI answers" means the responses people read in assistants like ChatGPT, Claude and Perplexity, and in Google’s AI Overviews above search results.
In practice it has three parts:
- Presence. Does the answer name your brand at all? Measured as a mention rate.
- Evidence. Does the answer link to your own site as a source? Measured as an owned citation rate, plus the list of other domains the assistant cites.
- Position against competitors. When the assistant talks about your category, how much of that talk is about you compared with the brands you compete with? Measured as tracked share of voice.
Why AI visibility matters for founders
A growing share of software research starts in a chat box. A buyer types a question, reads a shortlist and opens one or two of the names. If you are not on the list, you never learn that buyer existed. Nothing shows up in your analytics, and Google Search Console has no report for it.
The answer is also a summary of what the web says about your category. If review sites, Reddit threads and comparison posts describe you as "an enterprise analytics suite" while you sell to seed-stage teams, the assistant repeats that description to the exact people you want to reach. You can only fix a description you have seen.
Three things make this worth tracking even for a two-person company:
- Shortlists form before anyone visits your site, so your landing page never gets a chance to make the case.
- Competitors who get named keep getting named, because the pages that mention them get cited and linked more often.
- You can act on it. The sources an assistant cites for a prompt are a concrete list of places to earn a mention, fix a description or publish a better page.
What you can measure
Everything measurable in AI visibility comes from sampling: you ask a fixed set of questions, save the answers and count what is in them. These are the numbers that hold up.
| Metric | Question it answers | How it is counted |
|---|---|---|
| Mention rate | How often does the assistant name us? | Answers that mention any name or alias of your brand ÷ answers successfully captured |
| Owned citation rate | How often does it link to our site? | Answers that cite at least one URL on your domain ÷ answers successfully captured |
| Cited domains | Which sites shape answers in our category? | Sources attributed by the assistant, grouped by domain or page |
| Tracked share of voice | How much of the category talk is about us? | Your mentions ÷ mentions of every brand you track, in the same answers |
| Gaps | Where do competitors appear without us? | Answers where a tracked competitor is mentioned and you are not |
The denominator matters as much as the number. An answer that failed to load, hit a login wall or returned no AI Overview tells you nothing about your brand, so it should be left out instead of counted as a miss. Our methodology page spells out exactly how AEOmeter counts each one.
What you can’t measure
Plenty of dashboards show numbers that look like SEO metrics. Be skeptical of these:
- Impressions. No AI provider publishes how many people asked a question or how many times your name appeared in their answers. Nobody outside OpenAI, Anthropic, Perplexity or Google can see other people’s chats. An "AI impressions" figure is an estimate stacked on assumptions.
- A global rank. Answers change with the exact wording, the engine, the run, the account’s memory and settings, and location. There is no fixed "position 1 in ChatGPT" that every user sees. Our guide on how to rank in ChatGPT goes deeper on this.
- Prompt search volume. Providers don’t publish what people type. Keyword volume from SEO tools is a useful proxy for demand, nothing more.
- Cause and effect. If an answer mentions you and cites a review site, that does not prove the review site caused the mention. Treat cited sources as strong hints about where to look.
What remains is honest and useful: a consistent sample of real answers, repeated over time, with every number traceable to the answer it came from.
How AI visibility differs from search rankings
| Classic search | AI answers | |
|---|---|---|
| What the user sees | A ranked list of links | A written answer naming a few brands, sometimes with sources |
| Stability | Positions move slowly | The same prompt can produce a different answer on the next run |
| First-party data | Search Console shows impressions and clicks | None from the providers |
| What you track | Position per keyword | Mentions, citations and share across a prompt set |
| Main lever | Your own pages ranking | What many sources say about you, including your own pages |
The two overlap. Google says its AI features are rooted in its core Search ranking systems, and ChatGPT, Claude and Perplexity can all search the web before answering. Pages that rank well and get referenced elsewhere are more likely to be read when an answer is put together, so good SEO work still pays off here.
How to measure AI visibility by hand
You can get a first baseline in an afternoon with a spreadsheet.
- 1
Write 10 to 20 buyer prompts.
Mix category questions ("best product analytics tool for a seed-stage startup"), comparisons ("Mixpanel vs PostHog for a B2B SaaS") and alternatives ("Amplitude alternatives for small teams"). Our guide on choosing prompts has a full method.
- 2
List your brand and three competitors.
Include every name the assistant might use: product name, company name, common misspellings and your domain.
- 3
Use a clean session.
In ChatGPT, turn on temporary chat so memory and past chats don’t color the answer. In other assistants, use their private or incognito mode where it exists, or a logged-out window.
- 4
Ask each prompt in each assistant.
Paste the exact same wording every time. Save the answer text and every source link it shows.
- 5
Record what you see.
One row per prompt and assistant: mentioned yes or no, which competitors were named, which URLs were cited and whether any of them are on your domain.
- 6
Repeat on a schedule.
Run the same set weekly with identical wording. Change a prompt and that row starts a new baseline.
Twenty prompts across four assistants is 80 answers per round. That is enough to see patterns and small enough to do by hand a few times. The free AI visibility checker gives you prompts and a scoring sheet for exactly this, and our guide to tracking brand mentions in ChatGPT shows the full spreadsheet.
How to measure it with a tool
A tracking tool runs the same loop for you on a schedule. When you compare tools, ask four questions:
- Where do the answers come from? Some tools query model APIs. API output can differ from what people see in the consumer apps, which add web search, memory and their own interface.
- Can you read the original answer? Every number should open the answer text and sources it was counted from.
- How are failures counted? A blocked or timed-out check should be excluded and shown, never silently counted as "not mentioned".
- What happens when you edit a prompt? Trends should only compare answers to the same wording.
AEOmeter runs your prompts in the consumer apps of ChatGPT, Claude, Perplexity and Google AI Overviews through a Chrome extension, signed in with your own accounts, using each engine’s private mode where it can. Every metric links back to the answer and its cited sources. The trade-off is that your Chrome has to be open for scheduled checks to run. How it works covers the details.
What to do with the numbers
- Low mention rate on category prompts: open the answers and list the domains cited. Those review sites, directories and comparison posts are where your category gets described.
- A competitor appears and you don’t: read that answer’s sources. If a "best tools" list includes them and not you, that is a specific page to get onto.
- You are mentioned but described wrong: find the page that should state the fact plainly, usually your homepage, pricing page or docs, and fix it there first.
- Mentioned but never cited: publish pages an assistant can quote, like clear comparisons and use-case pages. The answer engine optimization guide walks through this.
FAQ
Is AI visibility the same as AEO?
They are related. Answer engine optimization is the work you do to appear in AI answers. AI visibility is the result you measure to see whether that work is landing.
Can I see how many people saw my brand in ChatGPT?
No. OpenAI and the other providers do not share impression or query data with brands. You can measure how often your brand appears in a sample of answers, which is a different and more honest number.
Why does the same prompt give different answers?
AI assistants generate text with some randomness, may run a fresh web search each time and adjust to account settings. That is why you track rates over many answers instead of one screenshot.
Which AI assistants should I track?
The ones your buyers use. For most software founders that means ChatGPT, plus Google AI Overviews because they appear in ordinary search. Perplexity and Claude are worth adding if your buyers are technical or research-heavy. See our engine pages for each one.
How often should I measure?
Weekly is enough by hand. With a tool, a daily schedule gives you more answers to average over. Either way, keep prompt wording fixed so you compare like with like.
