Buyers now ask an assistant before they open a search result. ScrapeWise runs your prompt list against ChatGPT, Gemini, Google AI Mode and Google AI Overviews on a schedule, and returns each answer with the sources it cited, so you can count how often you are named instead of guessing.
Someone on the team asks ChatGPT a question, sees a competitor named first, and sends a screenshot to Slack. That is one answer, on one day, from one account. It is not a measurement, and you cannot tell from it whether anything is getting better or worse.
ChatGPT, Gemini, Google AI Mode and Google AI Overviews are separate systems with separate source habits. Being recommended by one tells you nothing about the other three, and most teams are only watching the one they personally use.
There is no position 1 to 10 in an answer. There is a paragraph that either names you or does not, in some order, with a set of cited links underneath. Your existing rank tracker has nothing to report on it.
Ask the same question twice and the wording changes. A single check cannot separate a real shift in how the model talks about your category from normal variation, which is the whole reason this has to be a scheduled series rather than a spot check.
Your page can sit in the source list under an answer that recommends somebody else. Two different outcomes, two different fixes, and counting only one of them hides the problem.
The raw answer is a wall of text. Nobody can report on that. Turning it into share of voice means normalising brand names, de-duplicating cited domains and counting the same way every run, every engine.
AI answer surfaces covered: ChatGPT, Gemini, Google AI Mode, Google AI Overviews
Per 1,000 Google AI Overviews. Google AI Mode is EUR 1.50, ChatGPT and Gemini EUR 3.75 per 1,000 answers
Free requests on every new account, so you can check your own prompts first
Give us the questions your buyers ask and the brands to count, including spelling variants. We help configure the first run so tracking starts immediately.
Every prompt goes to ChatGPT, Gemini, Google AI Mode and Google AI Overviews on your schedule. Each answer is saved with the sources it cited.
Answers are reduced to countable brand fields, so you see which prompts name you, which engine you are missing from, and whether last month's change held.
This is what happens between a buyer asking an assistant which tool to use and your team seeing a number they can act on. One prompt, four AI surfaces, one scheduled run.
Each engine is called with the same prompt in the same run: ChatGPT, Gemini, Google AI Mode and the AI Overview on a Google search. Each call returns one row containing the answer text and the sources that answer cited. Google AI Overviews carry a state field, because the overview is not shown for every query, and a missing overview is a real result rather than an error.
The raw answer is prose. On top of it we apply the same cleaning work that any price feed gets: brand spellings are normalised so your name, your legal entity and your common misspelling count as one brand, cited links are reduced to domains and de-duplicated, and the order brands appear in is recorded. Being named in the answer and being cited as a source are kept as two separate fields, because they are two different problems.
One prompt on one day is an anecdote. The same twenty prompts, on the same schedule, across four engines, is a series. Share of voice is then simply the share of those prompts where each brand is named, per engine, over time. The rows land in a CSV or Excel export, the REST API, or your dashboard.
AI visibility looks easy until you run it weekly across four engines and a hundred prompts. These are the failure modes that turn a tracker into a wrong number, and the fields that keep each one honest.
The same prompt returns different wording, and sometimes a different brand order, on consecutive runs. A tracker that reports a single check as a change will produce alarm every week. The fix is structural, not clever: keep every run, date every row, and only call something a move when it holds across runs.
These get collapsed into one number constantly, and they have opposite fixes. If competitors are named in the prose while your page sits in the citation list, your content is being used but not recommended. If you are neither, you are not in the consideration set at all. Keeping both fields separate is what makes the report actionable.
Google does not show an AI Overview for every query, and AI Mode and the classic result page are different surfaces again. An empty overview recorded as a zero looks like lost visibility. It has to be recorded as not triggered, which is itself useful: a prompt that stops producing an overview has changed shape.
An assistant answers a German buyer and an American buyer differently, and the cited sources differ with them. On the two Google surfaces both the market and the language are part of the request, so both can be pinned. Gemini takes a language but no market. ChatGPT takes neither, so for those runs the row records the market the run was set up for and you keep one run per market rather than blending them.
AI visibility fails loudly when an engine returns nothing, and quietly when it returns something that reads convincing and is not representative. Quiet is the expensive kind, because it goes into a slide.
An assistant you have used for months has context about you. Asking it whether your product is any good is the least reliable test available, and it is the one almost every team runs first.
Buyers do not all type the same sentence. Visibility on the single phrase your team cares about can be high while the twenty phrasings customers actually use never mention you. The unit of measurement has to be a prompt list, not a prompt.
Being named third in a list of five is not rank three. There is no fixed slate. Treating order as a ranking invites week-to-week movement charts that measure sentence construction rather than preference.
You supply the prompts, the engines and the brands you care about. You receive one dated row per prompt per engine, with the answer, the cited sources and the brand fields derived from them.
The questions your buyers actually ask. Start with the phrasings that already appear in sales calls and support tickets rather than your keyword list.
best competitor price monitoring tool for ecommercehow do I track competitor prices automaticallyalternatives to a price monitoring platformChatGPT, Gemini, Google AI Mode and Google AI Overviews. All four by default, because a brand that is strong in one is regularly absent from another.
Your brand plus the competitors you want counted, with their spelling variants and legal entity names. This is what makes a mention countable instead of a thing a human has to read.
Your Brand, Your Brand OU, YourBrand.ai, Competitor A, Competitor BPinned in the request on Google AI Mode and AI Overviews. ChatGPT and Gemini take no market parameter, so it labels the run instead — keep one run per market and you still compare like with like over time.
How often the whole prompt list is re-asked. Weekly is enough for a trend; daily is for a launch or a category in motion.
One row per prompt per engine per run, 12 columns each
Delivered by REST API, CSV or Excel export, or into your dashboard. The answer and the cited sources come straight from the engine. The brand fields are derived from them by the same cleaning and matching step that normalises a price feed, so they are reproducible rather than read off a screen by hand.
One row per prompt per engine, dated. This is the prompt above as it lands in your system. Example values, with the brands anonymised.
| prompt_id | engine | brand_named | name_position | your_domain_cited | state | checked_at |
|---|---|---|---|---|---|---|
| P-014 | ChatGPT | yes | 3 of 5 | yes | answered | 2026-10-06 |
| P-014 | Gemini | no | no | answered | 2026-10-06 | |
| P-014 | Google AI Mode | yes | 1 of 4 | yes | answered | 2026-10-06 |
| P-014 | Google AI Overview | not triggered | 2026-10-06 |
Rows on their own do not tell you where to act. Once every answer is reduced to countable brand fields, the same rows roll up into the only AI visibility number worth reporting: the share of your prompt list where each brand is named, per engine. These are sample figures over a twenty-prompt list, to show the shape of the report. Each cell counts the prompts where that engine named the brand. The AI Overview column counts against the sixteen prompts where an overview appeared at all — the other four are recorded as not triggered, not as zero visibility. You build the view from the CSV or Excel export or the REST API in your own sheet or BI tool, or let our managed data team deliver it.
| Brand | ChatGPT (of 20) | Gemini (of 20) | AI Mode (of 20) | AI Overview (of 16 shown) |
|---|---|---|---|---|
| Competitor A | 18 | 17 | 14 | 12 |
| Competitor B | 12 | 13 | 11 | 10 |
| Your brand | 11 | 3 | 12 | 6 |
| Competitor C | 5 | 6 | 4 | 5 |
ChatGPT, Gemini, Google AI Mode and Google AI Overviews asked the same questions in the same run, so you can see which surface you are missing from instead of assuming they agree.
Every row keeps the full answer text and the links the engine cited, so you can tell whether you are being recommended, merely used as a source, or absent.
Brand names normalised, cited links reduced to domains, nothing invented that the engine did not return. Share of voice that holds up when somebody asks how it was measured.
Put the questions your buyers ask into one scheduled run instead of checking an assistant yourself. ScrapeWise sends the same prompt list to ChatGPT, Gemini, Google AI Mode and Google AI Overviews daily, weekly, monthly or on demand, and keeps every run.
Each row holds the answer text plus the links that answer cited. That is what lets you separate being recommended in the prose from being quoted as a source, which is the difference between a positioning problem and a content problem.
There is no position one in an answer. The number that works is the share of your prompt list where each brand gets named, per engine, tracked over runs. Clean the brand names once and the count is reproducible from one run to the next, by anybody who asks how it was measured.
Start tracking what ChatGPT, Gemini, Google AI Mode and Google AI Overviews tell your buyers. Same prompts, every engine, on a schedule, delivered as rows you can count.
Common questions about tracking brand visibility in AI answers with ScrapeWise.
It is a scheduled record of what AI assistants answer when buyers ask about your category. Instead of typing a question into ChatGPT yourself, your whole prompt list is sent to ChatGPT, Gemini, Google AI Mode and Google AI Overviews on a schedule, and each answer is saved with the sources it cited, so you can count how often each brand is named rather than relying on screenshots.