How to measure your visibility in AI engines - and in classic search
AI assistants now answer the questions your customers used to type into Google. Here is a practical method to find out whether they mention you - and what to do when they don't.
Search did not disappear - it moved into the answer
For twenty years, "being visible" meant one thing: rank on page one. That rule still holds, but it now covers only part of the journey. A growing share of buying research starts in a chat window, where the assistant reads the sources for the user and returns a single recommendation - often a shortlist of three or four brands, with no links clicked at all.
That changes the question you need to answer. It is no longer only "where do I rank for this keyword?" but "when someone asks an assistant for the best option in my category, am I one of the names it says?" Those two questions have different answers surprisingly often: plenty of companies that rank well in Google are never mentioned by Claude, and plenty of smaller brands are named by Perplexity because a handful of well-cited comparison pages describe them clearly.
Both matter. Classic SEO still drives the clicks that convert; AI visibility increasingly decides whether you make the consideration set at all. The practical work is to measure both, separately.
Classic SEO vs. AI visibility, side by side
The mental model transfers, the metrics do not.
| Classic SEO | AI visibility | |
|---|---|---|
| What is measured | Position in a list of blue links | Whether you appear inside a written answer |
| Unit of demand | Keywords | Prompts and follow-up questions |
| Result surface | A ranked SERP | One synthesised recommendation |
| Competitive signal | Who outranks you | Who gets named instead of you |
| Trust drivers | Backlinks, authority, technical health | Cited sources, consistent descriptions, third-party mentions |
| Volatility | Fairly stable day to day | Answer varies per run, per engine, per phrasing |
A six-step method you can run yourself
- 1
Define the prompts your buyers actually use
Start from the jobs your customers hire you for, not from your product name. 15–30 natural prompts ("best tool for X", "alternatives to Y", "how do I solve Z") beat a long keyword list.
- 2
Pick the engines that matter for your market
ChatGPT, Gemini, Perplexity and Claude answer the same prompt very differently - they use different retrieval sources and different training cut-offs. Measuring only one gives a misleading picture.
- 3
Run each prompt repeatedly
AI answers are probabilistic. One run is an anecdote. The same prompt run daily over weeks turns into a mention rate you can trust and compare.
- 4
Record mention, position and context
Three separate things: were you named at all, where in the answer, and how were you described. A first-place mention with a wrong description costs you more than a neutral third place.
- 5
Benchmark against real competitors
Your absolute score means little on its own. What matters is share of voice: of all brands named for your prompt set, how often is it you?
- 6
Track it over time and act on the gaps
Visibility moves when engines refresh their sources, when a competitor publishes, or when your own pages change. Only a time series shows whether your work is paying off.
Curious what this looks like applied to a real brand or a whole market? See the use cases.
What the numbers look like
Mention rate per engine - the share of runs in which your brand appears in the answer, next to the best-performing brand in the same prompt set.
Illustrative example, not measured data. The pattern - large gaps between engines for the same brand - is the point.
Share of voice beats a lone score
Of every brand named across your prompt set, how often is it you? This is the number that moves when a competitor invests in content - and the one worth reporting internally.
- Competitor A34%
- Your brand26%
- Competitor B22%
- Everyone else18%
Illustrative example.
Doing it by hand, and why it stops scaling
You can absolutely start manually. Open four assistants, paste twenty prompts into each, and log the results in a spreadsheet. That single afternoon is genuinely useful - it usually produces at least one uncomfortable surprise.
The problem is the second week. Twenty prompts across four engines is 80 runs; doing that daily to smooth out the randomness is 560 runs a week, each needing to be read, classified for mention and position, and compared against the last one. Manual sampling also has no memory: you cannot tell whether today's missing mention is a real drop or just variance.
That repetition is exactly the part worth automating - which is why we built Seeqer.
The measurement loop, run for you every day
- Daily scans across ChatGPT, Gemini, Perplexity and Claude with a visibility score per engine.
- Competitor benchmarking you define, so share of voice is measured against the brands you actually lose deals to.
- Mention, position and context captured per run - not just a yes/no.
- AI advice per query explaining what to change to close a specific ranking gap.
- Market research mode: general category prompts, to see who the engines recommend when nobody names a brand.
- Email reports and an API, so the numbers land where your team already works.
Setup takes one field: your website address. Seeqer reads the site, works out your segment and competitors, and proposes the prompt set for you.
See how the AI engines answer for your brand
Enter your website address and get your first visibility report the same day. 7-day free trial, no credit card.