Obility Editorial · · 4 min read
No assistant publishes how often a prompt is asked
Every SEO plan starts from search volume, so the first thing a team wants to know about AI answers is the obvious number: how many people ask ChatGPT about our category? Several tools now print a figure next to each prompt. It sits where the monthly search volume column used to sit, and it gets read the same way. It is a different kind of number.
Google publishes keyword volume through Keyword Planner. OpenAI, Anthropic, Google’s Gemini team and Perplexity publish no equivalent count for prompts. The companies that sell prompt volume say so themselves. Otterly, which offers an “intent volume” metric, wrote in May 2025 that none of the major AI platforms provide search volume data. Semrush’s documentation says individual prompts are often too specific and unique to measure directly. So every prompt volume figure is a model built from some other data, and the useful question is which data.
Each vendor builds the number from a different source
The vendors’ own documentation describes quite different methods. Otterly says its estimate starts from Google’s search volume for related keywords and then applies the ratio of each AI engine’s monthly visits and its growth trend. That is Google demand, rescaled by the size of ChatGPT or Perplexity. It tells you how many people search a topic on Google, adjusted for audience size, and nothing about how people phrase that question to an assistant.
Semrush says it sources prompts from AI search clickstream data and from Google’s keyword dataset for AI Overviews, groups them into topics, removes duplicates, simplifies the wording, and reports volume for the topic rather than for any single prompt. Profound describes its Prompt Volumes dataset as built on hundreds of millions of real, anonymized conversations.
These are reasonable engineering choices, and they guarantee that two tools will disagree about the same question. They observe different people, and they merge differently worded prompts into topics by different rules, so the size of a topic depends partly on where the vendor drew its edges. Before relying on a figure, ask whose conversations or clicks are in the data, which countries and assistants it covers, and how prompts are grouped. Semrush itself describes its AI metrics as directional signals and says no platform can provide exact numbers.
OpenAI’s own data shows how small the buying slice is
The most authoritative public evidence about what people ask ChatGPT comes from OpenAI. In a September 2025 NBER working paper titled How People Use ChatGPT, OpenAI researchers and academic economists classified a sample of about 1.1 million conversations. By July 2025, they report, users were sending 18 billion messages a week. Seeking Information grew from 14 percent of messages in July 2024 to 24 percent a year later. Questions about purchasable products made up 2.1 percent of messages across the sample period.
Two things follow for a marketer. A small share of a very large total is still a great many questions, so demand is rarely the constraint. Relevance is: whether the questions being asked are the ones your buyers ask. And the paper covers the Free, Plus and Pro consumer plans only. Business, Enterprise and Education accounts were excluded. A software buyer researching vendors on a workplace account sits outside OpenAI’s own published sample, so ask any volume vendor whether workplace accounts appear in its data at all.
Use volume to compare topics, not to count visitors
Prompt volume is useful in one narrow way: comparing topics inside the same tool, under the same method, over time. If one tool shows integration questions in your category rising faster than pricing questions, that ordering is worth acting on. Setting a prompt figure beside a Google keyword figure, or one vendor’s number beside another’s, compares two different models and tells you very little.
Do not forecast traffic from it either. A question put to an assistant is not a visit. The person may read the answer, take the shortlist, and never click anything.
For deciding which prompts to track, weight them with evidence you own: sales call notes, support tickets, site search terms and the question-shaped queries in Search Console. These show which questions your buyers ask, in their own words, without depending on someone else’s sample. Use outside volume as a tie-breaker between prompts you already believe matter, not as the reason a prompt is on the list.
Measure presence on the prompts you chose
Once the list exists, volume stops being the useful number. What matters for each tracked prompt is whether your brand appears, how the answer describes it, which competitors it names and which sources it cites, observed repeatedly under recorded conditions. In Obility, Answer Engine Insights groups prompts by audience, buying stage and market and keeps the exact wording and context attached to every observation, so a change in the prompt is never mistaken for a change in your visibility.
A prompt volume figure can help you choose the first twenty prompts to watch. It cannot tell you whether you are winning them, and it cannot be checked against any count the assistants publish. Treat it as a modelled estimate with unknown error, and if an assistant ever starts publishing first-party prompt counts, revisit everything above.