What your customers ask ChatGPT about you, and how to find out
Nobody can read ChatGPT's query logs. What you can do is ask the questions your buyers ask, across every assistant, on a schedule, and measure the answers. How AI brand visibility is actually tracked.
No access to real user prompts
Nobody outside the model provider can read the prompts users send to an AI assistant about your brand. Any tool that claims to show the actual prompts people typed about your company is presenting an estimate rather than a record.
The only honest method is to solicit responses on a schedule across ChatGPT, Claude, Gemini and Google AI Overviews and count whether each answer names your brand, a competitor or neither.
There is no rank to track
Traditional search engines deliver a ranked list of links where higher positions signal relevance. In contrast, each AI assistant returns exactly one answer per query, so rank-based metrics have no meaning.
We define visibility as the share of answers that name your brand.
Grounded search versus raw model output
Raw model API calls generate prose from internal weights without source attribution. That output may mention brands or features but does not cite the documents used. Grounded search retrieves web pages before composing an answer and returns both the answer text and a list of cited URLs.
We track four default sources: ChatGPT, Claude, Gemini and Google AI Overviews. That default set reflects a coverage decision rather than a statement on quality. Additional engines such as Perplexity, Copilot and Google AI Mode can be configured explicitly. On one real run across engines we recorded 11 cited URLs from ChatGPT, 15 from Perplexity, 15 from Google AI Mode, 5 from Copilot and prose with no structured sources from Gemini.
Citations give you the information you need because they name the page you have to beat. Every stored answer is kept whole with its list of citations so you can reopen the exact paragraph months later. When a vendor shows you a citations report, the question to ask is which execution path produced it. Storing each URL alongside the answer lets you plan content to address that snippet.
AI overview and organic results
Google AI Overviews appear inside the standard organic results payload rather than through a separate endpoint. A single request returns both the ten classic search results and the AI Overview answer when one is available. This dual view shows which pages rank in traditional results and which pages an assistant highlights.
A page that ranks fourth can supply the AI Overview citation and a page that ranks first can be absent from the overview entirely.
Reliable mention detection
Detecting whether an answer mentions your brand or a competitor requires more than a simple text search. String matching over-reports when your brand name doubles as a common word or appears in unrelated sentences. A classifier under-reports when the brand appears only in a citation title or is implied indirectly.
We apply both methods in two stages. First, a classifier assesses each tracked entity and labels sentiment and intent. Second, a deterministic text match scans the response text plus all citation URLs and titles and flips any missed mentions to true.
We store a record for every tracked entity on every answer, including cases where it was not mentioned. Those explicit absences provide the data needed for our competitor gap analysis.
The gap is the number that tells you what to publish
Prompt-level gap measures the percentage of responses to a specific question where a competitor appears and your brand does not. Sorting prompts by gap produces a priority list of questions you need to address first.
Share of voice aggregates across all prompts and engines, calculating the proportion of total mentions belonging to your brand versus each competitor. Gap directs your content strategy by revealing which questions you lose and whom you lose them to, and share of voice tracks your overall visibility trend over time.
Sampling cadence and measurement limits
AI answers do not shift on a daily cycle, so daily sampling mostly buys noise and API cost. Queries run by default every two days with each prompt tracked weekly in two batch windows to spread runs rather than stack them. You can switch to daily tracking by changing the run interval in project settings.
We do not score mention position within an answer because we lack a defensible method. Sentiment is captured at the mention level but is not aggregated in our standard reports.
Onboarding
Setup requires your domain. We crawl your site to identify your brand name, product function, buyer segments and a starting keyword set. We then suggest competitors based on your site content and a live search for alternatives.
You review and correct the entity list before any measurements run. A wrong or missing competitor silently distorts every gap number afterwards and every downstream metric depends on accurate entities.
A new project comes configured with your brand entity, the four default sources and a starter set of template prompts. These prompts cover what is [brand], how does [brand] compare to alternatives, what are the best features of [brand], is [brand] worth using and what is the best [category] solution. Replace them with the questions your sales calls open with, including the unflattering ones.
Pricing
Growth is $99 a month and Pro is $399 a month with higher caps, daily tracking and the technical SEO audit engine; both include a 14 day free trial. You can start a 14 day trial or see what is in each plan.