Hawkscan

Prompt volume

Prompt volume is an estimate of how often people ask AI assistants such as ChatGPT, Gemini or Microsoft Copilot about a given topic over a period. Unlike Google search volume, OpenAI, Google and Microsoft don't publish it, so tools model it from the conversations of opt-in user panels.

Why prompt volume is hard to know

Prompt volume would be the AI version of keyword search volume, but the companies that hold the data don't publish it question by question:

  • OpenAI publishes research on overall usage, such as its study of 1.5 million ChatGPT conversations, sorted into broad categories by automated tools.
  • Google's Generative AI performance report in Search Console groups your impressions in AI Overviews and AI Mode by page, country, date or device, not by query.
  • Microsoft's AI Performance report in Bing Webmaster Tools shows grounding queries, which "are not full user questions or prompts", from a sample of citation activity.

How tools estimate prompt volume

Tools that sell prompt volumes estimate them, typically from panels of users who agree to share their AI conversations. None of them sees the engines' own logs: Google states that "No third-party tool has access to our internal ranking or AI systems." In broad terms, a panel estimate works in three steps:

  1. Observe the prompts sent by a panel of consenting users.
  2. Group them by topic, since people rarely word the same need the same way.
  3. Extrapolate from the panel to the whole user base of each engine.

Each step adds uncertainty. A panel reflects the people willing to share their conversations, and a topic-level number hides how differently each person phrases the question.

Formula

A simplified panel estimate:

Estimated prompt volume = prompts on the topic seen in the panel × (users of the engine ÷ users in the panel)

Worked example (fictional values). A panel of 50,000 users sends 30 prompts about payroll software for restaurants in a month. If the engine has 10 million users in the same market, the estimate is 30 × (10,000,000 ÷ 50,000) = 6,000 prompts a month. With only 30 observations, a handful of panelists more or less moves the estimate by hundreds.

How to choose prompts without volume data

You don't need a volume to know a question matters. Start from your buyers' own words, in sales calls, support tickets and your Search Console queries. Bing's grounding queries also show the searches behind the AI answers that cite you. Then cover each intent once instead of every phrasing: Google notes that its AI systems "can understand synonyms and general meanings of what someone is seeking." That ability is semantic search, which matches meaning rather than exact words.

Hawkscan shows the full answers to each prompt you track on ChatGPT, Microsoft Copilot, Gemini and Google AI Mode, so you judge a prompt by what the engines say, not by an estimated count.

Sources

  1. Google Search Console Help, Generative AI performance report (Search)
  2. Microsoft Bing, AI Performance in Bing Webmaster Tools
  3. OpenAI, How people are using ChatGPT ()
  4. Google Search Central, Optimizing your website for generative AI features on Google Search ()

Your AI visibility starts here.

Your customers are already turning to AI for recommendations. Hawkscan gives you the full picture, so you always know where you stand and where to grow.