How to track your brand's visibility in ChatGPT and other AI answers

Measure how often ChatGPT, Gemini, Perplexity and Google AI name and cite your brand: the questions to ask, the engines and markets, the metrics and the API calls.

What AI visibility measures

Ask ChatGPT for the best CRM for a small team and the answer names a few brands and cites a few pages. AI visibility is how often your brand is among them. It is measured over many answers, because the same question gets a different answer from one run to the next.

Rates matter because AI answers are full of brands. In the September 2026 AI Index, 82% of 479 answers to buying questions named at least one brand, and the engines disagreed on which ones.

Step 1. Choose the questions your buyers ask

A visibility number is only as good as the questions behind it. Track what a buyer asks before they know your brand: recommendations, comparisons and the problems your product solves. Leave your brand name out of the prompts, because a prompt that names your brand returns your brand by request and its rate stays near 100%.

Start from demand. Prompt Research turns one topic into up to 50 questions people ask AI assistants, ranked by monthly search demand in Korea or the US, and returns monitorSet, the prompts to track first. Group the prompts by sub-field and buying stage so each group reads as its own result.

• 10 to 30 prompts per market is a practical start. More prompts make every rate firmer.
• Write prompts the way people type them into an assistant, in the market's language.
• Keep one prompt set per market: Korean prompts for Korea, English for the US.

Step 2. Pick the engines and markets

Each engine is its own channel. ChatGPT, Gemini, Perplexity, Bing Copilot, Google AI Overviews and Google AI Mode answer the same question with different brands and different sources. In Korea, Naver AI Brief and Naver AI Tab sit inside the country's largest search engine. Track the engines your buyers use.

Answers also change with the country and the place. Set the country of the market you sell in. For clinics, restaurants and stores, ask from the place too: located tasks on ChatGPT, Gemini and the Google surfaces answer for a station, a district or a street address, as the local GEO study shows.

Step 3. Collect the answers

Each answer is one task. Submit the prompt, the engine and the country:

curl -X POST https://api.querying.ai/v1/async/task \
  -H "Authorization: Bearer $QUERYING_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "taskType": "CHATGPT",
    "payload": {
      "prompt": "What is the best CRM for a 10-person sales team?",
      "country": "US"
    }
  }'

The completed task returns the answer in text and markdown, the pages it cites in sources and the searches ChatGPT ran in searchQueries. Poll GET /v1/async/task/{id} or set webhook.url to receive it. To send a whole prompt set at once, POST /v1/async/task/batch takes up to 500 tasks.

Step 4. Score every answer the same way

Apply the same rules to every answer:

• Mentioned: the answer names your brand under any of its names. List every alias, including short forms and other scripts, such as Round Lab and 라운드랩.
• Cited: one of the answer's source URLs sits on one of your domains.
• Position: where the first mention appears in the answer text.
• Competitors: the other tracked brands the answer names.

A monitor does all of this. It sends every prompt to every engine on a schedule, scores each answer once and keeps the answer behind every number. Create one with your prompts, aliases, domains, competitors, country and schedule:

curl -X POST https://api.querying.ai/v1/monitors \
  -H "Authorization: Bearer $QUERYING_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "CRM, US",
    "engines": ["CHATGPT", "GEMINI", "PERPLEXITY"],
    "prompts": [
      "best CRM for a 10-person sales team",
      "easiest CRM for a small business"
    ],
    "aliases": ["Acme", "Acme CRM"],
    "domains": ["acme.com"],
    "competitors": [{ "name": "Globex" }, { "name": "Initech" }],
    "country": "US",
    "intervalHours": 24,
    "alertBelowPct": 30
  }'

Step 5. Run on a schedule and read the trend

Run the same set daily or weekly and compare equal periods. A monitor reports mention and citation rates for the window and for the window before it, per engine and per prompt, and reports the change when the two periods are comparable. It flags any window with fewer than 30 scored answers.

A run costs the number of prompts times each engine's credits. Twelve prompts on ChatGPT (2 credits) and Gemini (1 credit) cost 36 credits a run, about 1,080 credits a month on a daily schedule. Set alertBelowPct and the monitor emails you when the mention rate falls below it.

Step 6. Turn the results into work

• Open the answers behind a change. When a rate moves, read what the engines said and which brands took the place.
• Find the sources that win. The monitor lists the domains and pages the answers cite most for your prompts. Those pages are where the engines read about your category: improve your own, publish where they cite, and earn mentions on the pages they trust.
• Work through the competitor-only answers. Each one is a prompt and an engine where a competitor appears and your brand is absent: the most direct list of gaps.
• Measure again after you publish. Compare the period after a change with the period before it, on the same prompts and engines.