There is a new number doing the rounds in marketing meetings: prompt volume. How many people are supposedly asking ChatGPT about your industry, your product, your competitors. Plenty of tools will happily sell you a dashboard full of it.
Here is the catch. Nobody actually has that data. The AI platforms do not publish usage figures, so every prompt volume number you have ever seen is an estimate. A guess wearing a lanyard.
We love a good spreadsheet at Fresh Digital. But building an AI search strategy on prompt volume is like writing a sales forecast from horoscopes. So let’s look at why the number misleads, and what to use instead.
Key Takeaways
- Prompt volume is modelled guesswork, not measured data, so it should never decide what content you create.
- AI answers vary wildly between users and sessions, which makes SEO-style rank tracking close to meaningless in AI search.
- A January 2026 study by SparkToro and Gumshoe.ai found the brands named in AI answers are essentially random from one response to the next.
- Your customers’ real questions, pulled from sales calls, reviews and forums, are a stronger signal than any vendor dashboard.
- Use GEO tools for direction and monitoring, then judge trends over three to six months rather than reacting to single swings.
Prompt Volume Is a Guess Dressed up as Data
Quick definition first. Generative engine optimisation (GEO) is the work of getting your business mentioned and cited in AI answers, in tools like ChatGPT, Perplexity and Google’s AI Overviews. Prompt volume is the GEO industry’s attempt at a keyword search volume equivalent: an estimate of how often people ask AI a particular question.
The trouble sits in the word estimate. Google keyword data works because Google measures real searches and shares the numbers, give or take. ChatGPT, Claude and the rest publish nothing of the sort. So GEO tools build their figures from two main sources, and both have problems.
Panels Skew Nerdy
Some platforms recruit consumer panels: people who agree to share their AI activity. That produces millions of real prompts, which sounds great. But people who opt in to having their chats tracked are not a fair sample of your customers. Panels lean tech-savvy and early-adopter, and the modelling then stretches that skewed sample across everybody else.
Api Answers Are Not the Real App
Other tools fire test prompts at the models through the API, the plumbing developers use, rather than the app on your customer’s phone. Early research suggests the two can give different answers. So a chunk of the “AI visibility” data on the market describes a version of the tool your buyers never touch.
AI Answers Will Not Sit Still
Even if the volume data were solid, there is a deeper problem. Large language models are probabilistic. Ask the same question twice and you can get two different answers. Ask from two different accounts and the gap widens again.
The research on this is striking. In January 2026, Rand Fishkin’s SparkToro and Gumshoe.ai ran 2,961 prompts through ChatGPT, Claude and Google’s AI tools using 600 volunteers. The chance of two responses naming the same list of brands was under 1 in 100. The chance of the same list in the same order was under 1 in 1,000. Fishkin’s verdict was blunt: any tool selling an “AI ranking position” is making it up.
It does not settle down over time either. Profound, one of the bigger AI visibility platforms, published research showing that the domains cited for identical prompts change substantially month to month. The industry calls this citation drift. We call it a very good reason not to panic over a one-month dip, or pop the champagne over a one-month spike.
The short version: there is no rank to track. Anyone promising you position three in ChatGPT is selling a screenshot, not a strategy.
The Data You Already Own Beats the Dashboard
So if prompt volume cannot lead, what should? The unfashionable answer is your customers. The way real buyers describe their problems is the closest thing you will get to genuine prompt data, because those are the very people typing the prompts.
Three places to dig:
- Your own conversations. Sales calls, support tickets, onboarding chats and emails. The objection your team hears every week is almost certainly being put to ChatGPT as well.
- Public watering holes. Reddit threads, niche forums, LinkedIn comments, Trustpilot and G2 reviews. These capture unfiltered questions in plain language, which is exactly how people talk to AI.
- Your search data. Search Console queries and site search logs show the wording people already use to find you. AI prompts tend to be those same questions written out as full sentences.
A question that keeps cropping up in a subreddit makes a better content brief than any vendor-curated prompt list. It is real and it is written in your buyer’s own words. That is the raw material our Fresh Content team builds from, and it works for Google and AI engines alike.
Think in Clusters, Not One-Off Prompts
People do not ask AI one tidy question. They ramble, follow up and rephrase. So don’t chase individual prompts. Group them by intent and build depth around the theme.
Say you sell accountancy software. “How do I choose accounting software for a small UK business” sits in a cluster with questions about pricing, VAT, switching providers and integrations. Cover the cluster properly and you become a credible source for the whole conversation, however a buyer happens to phrase it.
This is topical authority, and it is the same discipline that wins traditional organic search. That overlap is the good news. Solid SEO foundations and AI search visibility are mostly the same work. One strategy, measured in two places.
What GEO Tools Are Actually Good For
None of this means bin the tools. It means demote them. They are decent at direction and dreadful at decisions.
Used well, they can flag topic gaps, show whether your brand appears in relevant AI conversations at all, and give a rough share of voice against competitors over time. Used badly, they set your content calendar based on modelled numbers nobody can verify.
The monitoring routine we recommend instead:
- Write down 20 to 30 prompts that reflect what your ideal customers genuinely ask.
- Run them at least monthly across the platforms your audience uses: ChatGPT, Perplexity and Google’s AI Overviews.
- Log when your brand, your content or a competitor shows up.
- Judge the trend over three to six months. Ignore single-month wobbles, because citation drift guarantees plenty of them.
Monitoring informs. Your understanding of your customers decides.
FAQs
Is Prompt Volume the Same as Keyword Search Volume?
No. Keyword volume comes from searches Google actually measures. Prompt volume is modelled from opt-in panels and API tests, because no AI platform publishes usage data. Treat it as a rough hint, never a fact.
Should We Ignore GEO Tools Completely?
No, just stop letting them lead. They are handy for spotting topic gaps and tracking whether your brand appears in AI answers over time. The decisions about what to create should come from real customer language, not modelled estimates.
How Do We Measure AI Search Success Without Rankings?
Build a fixed set of 20 to 30 customer-style prompts, check them monthly across the main AI platforms, and log brand mentions and citations. Pair that with referral traffic from AI tools and, most importantly, leads and sales. Look at the direction over a quarter or two, not the week-to-week noise.
AI search is noisy and full of confident people selling made-up numbers. Cut through it with the boring stuff that works: know your customers, answer their actual questions and check the results on a sensible schedule. Fancy a hand with that? We’re friendly, and we don’t bite.