Citino teamAI visibility tracking software should tell you, for the questions your buyers ask, whether ChatGPT, Google AI Overviews, Perplexity and other engines name you, how they describe you, which pages they rely on, and whether the changes you make move any of it. This checklist sets out what that takes, layer by layer.
It is a requirements list for AI search analytics. For the questions to ask a vendor in a demo, see how to choose an AI visibility tool.
1. The answers themselves
Everything else is computed from the answers, so the software should collect them carefully and keep them.
- Every engine your buyers use. At least ChatGPT, Google AI Overviews and Perplexity. Check which engines are included and which cost extra.
- Fixed questions, fixed wording. Language models are sensitive to small changes in a prompt (Sclar et al.), so a reworded question should start a new series, not continue an old one.
- A regular schedule. Daily runs at a fixed time. Weekly checks take months to separate a change from noise.
- A clean session. No account history or personalisation, so the answers reflect what a new buyer would see.
- The full answer, stored. The complete text, the brands named in order, and the pages cited, so any number can be traced back to the answers behind it.
2. The four core numbers
| Metric | What it should report | What to check |
|---|---|---|
| Visibility | Share of answers that name you | Reported per engine, with the number of answers behind it |
| Share of voice | Your share of all brand mentions | A fixed competitor list; branded questions excluded |
| Position | How early you are named | Counted only over answers that name you |
| Sentiment | How you are described | Scored over answers that name you, with the wording visible |
Each number should come with a margin of error, calculated over questions rather than over answers. Answers to the same question move together, so treating them as independent makes a result look more precise than it is. How to measure AI visibility explains the method, and how to measure share of voice applies it to share of voice.
3. The sources behind the answers
Engines that search the web cite what they read. Google says AI Overviews and AI Mode may run multiple related searches across subtopics to build one response, and Perplexity says each answer includes numbered citations to its sources. The software should turn those citations into something you can act on:
- Every cited page, per answer. Which engine cited it, for which questions, and how often.
- Who owns each page. Your own pages, third-party pages you can influence, and competitor pages. The order you can act on them in follows from that.
- When each page was first cited. So you know whether an engine has read a page since you changed it.
4. The facts stated about you
Being named is not enough if the answer gets your price or features wrong. The software should let you state the facts that matter, such as price, plans and integrations, and check each engine’s answers against them: correct, different, wrong or not stated, with the page the wrong version came from. How to fix what AI assistants get wrong covers what to do with the results.
5. The visits and what they are worth
AI answers send some buyers to your site. The software should connect to your web analytics and show the visits that came from AI assistants, matched to the questions that sent them where that can be done reliably. Visits that cannot be matched should be shown as unmatched rather than spread across questions to look tidy.
A buyer who reads an answer and visits later through another route will not be counted, so treat the result as a floor, not the full value.
6. Whether your changes worked
Most tools tell you what to fix. Fewer can tell you whether the fix worked. For that, the software needs to:
- Record which questions a change targets, chosen before results come in.
- Wait until an engine has read the changed page.
- Measure the change on those questions, minus the median change of competitors over the same days.
- Compare what is left with its margin of error, and give a verdict.
How to tell whether an AI search fix worked sets out the steps.
A checklist for AI visibility tracking software
| Layer | Minimum | Complete |
|---|---|---|
| Answers | Daily runs on your main engines | Full text, brand order and citations stored for every answer |
| Numbers | Visibility per engine | All four metrics, per engine, each with a margin of error and its n |
| Sources | A list of cited pages | Ownership, frequency and first-citation date per page |
| Facts | Spot checks by hand | Every stated fact checked against yours, with the source of each error |
| Visits | AI referrals in your analytics | Visits matched to questions, with unmatched visits shown |
| Fixes | A list of suggestions | A verdict on each change, net of what moved every brand |
If a tool cannot show you the answer behind a number, the number is a claim, not a measurement.
Where Citino fits
Citino is an AI search analytics tool. It asks ChatGPT, Google AI Overviews and Perplexity your buyers’ questions every day, with Gemini, Google AI Mode and Claude as add-ons, keeps every answer with the brands it named and the pages it cited, and reports each number with a 95% margin of error bootstrapped over prompts, as the methodology sets out. Everyone tells you what to fix. Citino tells you whether it worked. See how Citino measures.