How AnswerHawk measures brands in AI answers

We send the prompts you choose to 8 AI engines on a fixed schedule, in a fresh session, and keep every answer. This page explains how we collect them, how to read the numbers and what we do not measure.

Updated November 20, 2026

How we collect answers

You choose the prompts: the questions your buyers ask AI about your market. We send each one exactly as written and never rewrite it.

  • A fresh session every time. Each prompt runs in a new session, signed out, with no account, no chat history, no memory and no custom instructions.
  • Public interface or official API. ChatGPT, Gemini, Google AI Overviews and Google AI Mode are read from their public, signed-out interface: the same page a visitor without an account gets. Perplexity, Claude, Grok and DeepSeek run through their official API.
  • On a schedule. Three times a week by default. You can switch any prompt to daily or weekly. Some engines run at most once a week per prompt, and the app shows when.
  • Your market. Each project has a country and a language. Where the engine supports it, prompts run with them.
  • Every answer is kept. The full text, the date, the engine, the model the engine reports (when it reports one) and the sources it cites. Nothing is summarized or thrown away.

A collection that fails (the engine is down, times out or sends nothing back) is tried again and never uses your answers. One engine failing never stops the others.

What a signed-in user sees and what we measure

An answer in your own ChatGPT or Gemini app can differ from the one we collect. Apps add things a clean session does not have:

  • memory and past conversations;
  • custom instructions and connected accounts;
  • a precise location;
  • the model a paid plan unlocks, or one the user picked;
  • tests the engine runs on part of its users.

We measure the baseline: what a new person with no history gets for the same question in your market. It is the fairest comparison between you and your competitors, and it can be repeated run after run. If your signed-in app shows something else, both are real: yours is a personalized answer.

Answers vary: why we show percentages, trends and sample size

Ask the same question twice and an AI engine can name other brands, in another order, with other sources. One answer proves little, so AnswerHawk works like a survey, not a screenshot:

  • Percentages, not single answers. Visibility is the share of answers that name your brand: 6 of 10 answers is 60%.
  • Trends over weeks. Each run adds answers. The chart shows the direction, which is steadier than any single day.
  • The sample next to every percentage. You always see how many answers a number comes from. Under 20 answers we mark it as a small sample: one or two answers move it a lot, so read it as a direction, not an exact figure.
  • Position as an average. When an answer lists brands, we record where yours appears and average it over the answers that name you.

More prompts, more engines and a higher frequency give larger samples and steadier numbers.

Engines we cover and what each one measures

8 engines. Most answer with a live web search and cite the pages they used. DeepSeek is measured without web search.

Engine How we read it Web search Sources cited
ChatGPT Public interface, signed out Yes Yes
Gemini Public interface, signed out Yes Yes
Perplexity Official API Yes Yes
Claude Official API Yes Yes
Grok Official API Yes Yes
DeepSeek Official API No No
AI Overviews Google Search, signed out Yes Yes
AI Mode Public interface, signed out Yes Yes

Each answer stores how it was collected. Starter tracks ChatGPT, Gemini and Google AI Overviews; Growth and Agency track all 8.

With web search, an answer reflects what the engine finds on the web that day, and the sources it cites show which pages shape it. Without web search, the answer comes from what the model learned in training: it changes slowly and cites no sources. Some engines also report the searches they ran (the query fan-out), and we keep those too.

How we identify brands and sources

Each answer is read automatically into a fixed, validated format. We look for:

  • Brands: your brand and its other names, your competitors and any other brand the answer names, with the position where each appears and the tone used about it (positive, neutral or negative).
  • Sources: the pages the engine cites, grouped by site, and whether your site is one of them.

Names are checked against your brand, its other names and your competitors, so a common word is not counted as a brand by accident. The original answer is stored apart from what we read in it, so old answers can be read again when this step improves. You can open any answer in the app and see what was counted.

What we do not measure

  • Real traffic. We do not see visits, clicks or sales that come from AI apps. Your analytics tool is the place for that.
  • Real people’s conversations. No tool can see what people ask in their own chats. We see the answers to the prompts you define.
  • A guaranteed position. AI answers have no fixed ranking to buy or keep. We show how often and where you appear in a sample, and what to work on. Nobody can promise a position.
  • How many people ask a question. That number is not public. Any figure for it is an estimate.
  • Every possible wording. We measure the prompts you track. Other wording can get another answer, so track the questions that matter to you, in your buyers’ words.

Read AnswerHawk as a repeated sample of what AI engines answer about your market. To see a first sample for free, try the AI Visibility Checker. Plans and limits are on the pricing page.

Questions about the method

Is AnswerHawk’s data accurate?

It is exact about what it measures: the answers the engines gave to your prompts, collected in a clean session and stored in full. Answers vary between runs, so each percentage shows its sample size. Read trends over weeks rather than a single run.

Why does my ChatGPT show a different answer?

Your app uses your memory, past chats, custom instructions, location and the model of your plan. We ask in a fresh signed-out session, the baseline a new buyer gets. Both answers are real; ours can be compared run after run.

How often are answers collected?

Three times a week by default, or daily or weekly if you change it. Some engines run at most once a week per prompt. Every run is dated in the app.

Do you use the engines’ apps or their APIs?

ChatGPT, Gemini, Google AI Overviews and Google AI Mode are read from their public, signed-out interface. Perplexity, Claude, Grok and DeepSeek run through their official API. DeepSeek is measured without web search.

Can AnswerHawk show traffic from ChatGPT?

No. AnswerHawk samples answers to the prompts you track. It does not see people’s conversations or measure visits from AI apps.

Does a failed collection use my answers?

No. If an engine fails or times out, that collection does not count, and the other engines still run on schedule.

See where you stand in AI answers.

Choose a plan and see your first answers in minutes.