AI Brand Visibility Tracker

Ask the questions your buyers actually ask, and find out whether Google's AI answers name your brand, how early, whether they link you, and which competitors get named instead.

Search used to end with ten blue links and a ranking you could measure. An AI answer ends with a recommendation, and either you are in it or you are not. This measures that.

Two surfaces, not one

Google has two AI answers and they are different products. The AI Overview is the summary above the blue links: short, opinionated, three or four sources. AI Mode is the conversational tab: long, structured, often a breakdown by use case with different brands in it.

They disagree often enough that averaging them would hide the useful part, so every question is asked on both and each gets its own row. A brand can lead the Overview and be absent from AI Mode for the same question, and that is a finding, not noise.

Neither needs an API key from you.

What a row tells you

• brand_mentioned and brand_mentions — is the brand named, and how often
• brand_rank — 1 means the answer names you before any other brand. This is the number to watch week to week
• brand_first_position — where in the answer, as a character offset, so earlier is lower
• brand_context — the sentence you appear in, which is where you find out whether being named was good news
• brand_cited and brand_citation_urls — whether the answer linked your site

brand_mentioned and brand_cited are deliberately separate. An answer can name you without linking you, and can link your domain without naming you. One is reputation, the other is traffic, and a single "visible" flag would hide which of the two is missing.

• competitors_mentioned — who was named instead, each with the domain it cited and where it appears
• competitor_count and share_of_voice — your slice of every brand mention in that one answer

How competitors are found

Give a list under Known competitors and they are matched by name. Leave it empty and the run works them out: it takes the brand name out of every domain the answer cited and keeps the ones the answer also names in its text.

That is grounded on the citations on purpose. Pulling capitalised words out of prose finds "The Best", "However" and every product category at least as often as it finds a rival. A cited domain that also gets named is a real competitor by both tests.

Each competitor carries the domain it came from, because a token can read as an ordinary word: scrape is the name behind scrape.do, and only the domain tells you which it is.

Questions with no answer are not failures

Google decides per request whether to offer an AI Overview at all. A question it declines comes back with answered: false, is not billed, and is counted separately in the scorecard as unanswered.

This matters more than it sounds. Folding a refusal to answer into your visibility rate would mix two different problems: "the AI does not recommend us" and "the AI does not answer this question at all" have different fixes, and only one of them is about your brand.

The scorecard

summary in the key-value store is the run's own answer:

• visibilityRate per surface — answers naming you, over answers given
• citationRate per surface — answers linking you
• averageRank — where you land among the brands that do get named
• topCompetitors — who shows up most across the whole prompt set
• promptsWithoutBrand — the actionable list. The questions that got a real answer with no mention of you

Rates are over answered checks only, for the reason above.

Writing good prompts

This is the input that decides everything else. A question nobody asks tells you nothing, however good the measurement is.

Write them the way a buyer types them, not the way you describe yourself. "best ecommerce price monitoring tool" is a question someone asks; "enterprise retail pricing intelligence platform" is a phrase from your own homepage. Include the comparison questions ("X vs Y"), the ones with a qualifier ("for small business", "free"), and the ones where you expect to lose. The last group is where the value is.

Country and language

AI answers are regional. The same question names different brands in the US than in the Netherlands, so Country and Language are separate inputs and both are carried on every row. If you sell in several markets, run them separately and keep the rows apart.

Errors

Code	Meaning
bad_input	No brand name or no prompts
engine_error	One surface could not be read for one question

An engine_error on one row does not affect the others; every question and surface is measured independently.

Related actors

google-ai-overview-scraper and google-ai-mode-scraper return the raw answers without the brand measurement, if you want to do your own analysis. For the blue links behind them, see the SERP actors.