AI Visibility: the answers
The AI Tracker is part of AI Visibility now: one tool, one setup and one page. This page is about its Answers tab – what AI assistants and the places buyers ask tell a buyer about your category and about you: how often they name you when a buyer asks without naming you, where you sit in the lists they give, your share of the conversation against your competitors, which sources they cite, and what they get wrong about you.
Open it from Studios → AI Visibility. Old links to the AI Tracker open the same page.
How it measures
The assistants are asked the way a buyer asks them: plainly, with nothing about your brand in the question or around it. Two kinds of question:
- Questions that do not name you ("What are the best options for …?", "Which … would you recommend for …?", "What are the best alternatives to …?"). Whether you come up is your visibility – the number that matters.
- Questions that name you ("What is … and what does it do?", "Is … any good?"). These read what the assistants believe about you.
Named or not, position in a list, the competitors mentioned and the sources cited are counted from the answers, so the same answer always scores the same. A capitalised name of several words is matched as written, so a phrase like "the AI CMO role" is not counted as your brand. Tone and misstatements are read by an AI judge against your business profile, and the report says so where it shows them.
Set it up
- Brand, Website and What a buyer calls it come from your profile. The category is the one to get right: category questions use it, and it should be what a buyer would type.
- Competitors are counted in every answer for your share of voice.
- Questions: suggested ones are filled in; edit, remove or add your own (up to 12), or find them from what buyers search. Each shows whether it names you.
- Where to ask: the answer a buyer gets when the assistant searches, with the pages it cites – ChatGPT, Google AI Overview and AI Mode, Gemini and Claude (with web search, asked through its API, not the Claude app) are ticked; Perplexity is there to tick. Copilot can't be asked from outside. The AI Overview is counted on its own row, because Google does not show one for every question.
- Also ask from memory: the models' own answers, without searching the web. None is ticked.
Run the check asks every question in every place you picked, together with the site and agents check of AI Visibility. For another company, or with only the models from memory ticked, the button reads Ask the assistants and asks the answers alone. A run usually takes one to three minutes. You can leave the page; the run carries on and is saved under History.
What it costs
- Assistants cost one credit per answer.
- Where buyers ask is priced on what each read costs. The most it can cost is held before the run; when the run ends you pay for what it read, and the rest comes back to your credits. Claude costs the most, because every Claude answer runs its own web search.
The price shows beside the button, with both parts when a run has both. If none of the answers comes back, nothing is charged.
Read the answers
The Answers tab opens with Mentions, Cited and Share of voice, each with one row per place: a count, "n/a" for a place not asked, and "not read" when every read of it failed – never 0 for either. Then:
- Visibility – named in how many of the answers to questions that do not name you.
- Average position – where you sat in the ranked lists you were named in.
- Share of voice – your mentions against each competitor's, in those answers.
- Wrong about you – statements your own profile contradicts, each quoted with the correction and the assistant that said it.
- Every question, every assistant – a grid: filled means named, a number is your place in the list; on questions that name you, the answer's tone. Click a cell for the answer, word for word, with your name marked.
- By assistant and Sources the assistants cite – which assistants know you, and the sites they lean on. Being written about on those sites helps the next answer find you; a full check reads whether each cited page names you.
- Over time – from the second run on, visibility per run as a line.
Calls that fail are counted as failed, never as "not named".
Download the answers gives every answer as a spreadsheet. Run it again reuses the questions, assistants and places of the run you are looking at.
The pages to make that the answers point to are under What to do, in one list with the check's other fixes, each with Write it, Copy brief, Add as a task and Ask about this.
Check every week, under What to do, is the brand's one weekly check, with an alert on what changed; its price shows beside the switch. A brand gets one run and one alert a week, not two.