Measure AI visibility
Which brands assistants name when a buyer asks, and which pages they built the answer from.
Most buyers now ask an assistant — or read the AI box above the search results — before they visit anyone's site. AI visibility asks the questions a buyer would ask, on a schedule, and records which brands each assistant names and which pages it read to answer.
What it needs
Each assistant is measured only if you've given Seolyticize a way to ask it. With nothing set up, the page says so rather than reporting the client as invisible.
Assistant | How it's asked |
|---|---|
Google AI Overview, Google AI Mode | Read from the results page through Your own proxies (Settings → Search data). Needs residential addresses; datacenter addresses are blocked. |
ChatGPT, Perplexity, Microsoft Copilot | The logged-out consumer product a buyer uses, through your proxies — or the official API with an OpenAI / Perplexity key. |
Claude, Gemini, Grok, Meta AI, DeepSeek | The official API, with your key under Settings → AI assistants. |
Each reading is labelled with how it was taken, and an assistant that answered from training rather than a live search is marked as such.
Set it up for a client
- Add the questions
On the client's AI visibility page, under Questions, add what a buyer would ask (for example best dentist in Soho), with the market as a two-letter country code. Not sure? Suggest questions drafts some from the client's site.
- Add the brands
Under Brands to look for, add the client and whoever they compete with: name, domain, and any other names it goes by.
- Pick a schedule and assistants
Under Schedule, choose Every day, Every week, Every month or Off, and tick the assistants to ask. The panel shows how many answers each check makes and the most Claude can cost per check and per month; other assistants bill your own accounts.
- Try one now
Ask once puts a single question to the assistants straight away.
Reading the results
Share of answers — the share each brand took, and how many questions it was named in at all.
Prompt results — each question and assistant in one check. Open a row to read the answer.
What the assistants read — the pages behind the answers over the last 90 days.
Where a rival is and you are not — sources read for questions whose answers named a competitor and not the client. It means the assistant didn't mention the client, not that the page leaves them out.
What kind of pages they read — forums, reviews and so on. Forum-led categories are won by being talked about, review-led ones by review volume.
How much the assistants agree — sources each assistant read that no other did.
Who is looking these brands up — Wikipedia article views over twelve months: a name's direction of travel, not a market size.
Readable by assistants, on the client's Content health page, checks the other side: whether assistants can read and cite the client's own site.