Blog · · By HeardOf
Where does "ChatGPT search volume" come from? What five vendors say about their prompt data
The short answer
None of the five vendors' pages we read on 25 September 2026 says OpenAI supplies it: Profound licenses panel conversations, Semrush cites clickstream, Ahrefs and AthenaHQ use keyword data, and Conductor calls the metric misleading.
Where does ChatGPT search volume come from?
From vendors' own data and models; none of the pages we read says it comes from the engines, whose prompt logs, Conductor's article says, are private. On the pages we read on 25 September 2026, Profound says it licenses conversations from consumer panels, Semrush names "AI search clickstream data and Google's keyword dataset", Ahrefs builds its prompts from its keyword database, AthenaHQ's API reference names a "keyword-volume provider", and Conductor argues that any AI prompt volume metric "you are being sold is fundamentally misleading".
None of the pages we read uses the phrase "ChatGPT search volume"; Profound's launch post says "AI Search Volume" and Semrush says "AI Volume", each an estimate of how often people ask an assistant about a topic. Each vendor describes where its figure comes from in its own words, and the table puts those words side by side and ranks none of them.
| Vendor | Pages read, with date | Where the page says the data comes from | What the figure is called on its pages |
|---|---|---|---|
| Profound | Introducing Prompt Volumes, 16 Dec 2024, historical; Prompt Volumes feature page, undated; help article, shown as updated four months before our read | "licenses conversations from multiple, double-opt-in consumer panels of real answer engine users"; "Both" to "Is that data direct from users, modeled, or both?" (feature page); data sources "ChatGPT, Gemini, Claude, and Perplexity" (help) | Prompt Volumes |
| Semrush | Knowledge-base articles 1607, 1596, 1597 and 997; Semrush One page; all undated | "AI search clickstream data and Google's keyword dataset for AI Overviews"; topic volume from "third-party data on real AI interactions" and its machine learning models (1607) | AI Volume: "The estimated number of search queries the topic gets on a selected AI Platform" (1597) |
| Ahrefs | Brand Radar page, undated; help article 11064852, updated 3 Sep 2026 per its structured data | "People Also Ask questions from keyword queries in its keyword index" (help); "Every prompt is modeled from real user searches in Ahrefs' keyword database" (Brand Radar) | Estimated Impressions, weighted "by the real search volume behind each prompt" (Brand Radar) |
| AthenaHQ | Estimating AI Prompt Volume Across Platforms, 7 Jun 2025, historical; QVEM article, 10 Jan 2026, historical; API reference, undated | "both public and private data" (2025 article); "Open datasets, API feeds, research publications", "Premium data providers, industry partnerships", "AthenaHQ user analytics, first-party tracking" (QVEM article); "The keyword-volume provider" (API) | "Estimated monthly search volume for this prompt" (API) |
| Conductor | An Honest Talk on AI Prompt Volumes, 17 Dec 2025 in its structured data, historical; shown as last updated 8 Jul 2026 | Offers no volume figure on this page; says vendors rely on "paid panels and browser extension data", and proposes Google Trends, Search Console and Keyword Planner data instead | "AI MSV" and "AI prompt volume", as the page names what others sell |
Does OpenAI publish ChatGPT search volume?
Not according to the two vendors' pages we read that address it. Conductor's article says the LLMs are "black boxes", that "Their prompt logs are private" and that "They do not provide public MSV data"; Otterly's help article of 17 July 2026 says "AI search engines like ChatGPT and Perplexity do not publish query data." Our guide to choosing the prompts you track quotes the same Otterly line. We did not look for an OpenAI page that says otherwise, so this is what two vendors say, not something we checked with OpenAI.
What does Profound say Prompt Volumes are built from?
Licensed panels, then modelling. Its Prompt Volumes page, undated, says "Profound licenses conversations from multiple, double-opt-in consumer panels of real answer engine users", and, asking itself whether the data is "direct from users, modeled, or both", answers "Both", with "statistical modeling that corrects for demographic and geographic biases" and "volumes scaled to reflect the full population". Its help article lists ChatGPT, Gemini, Claude and Perplexity as data sources and says "ChatGPT data is available from January 2025".
The post that launched the feature, "Introducing Prompt Volumes" of 16 December 2024, historical, describes "a huge proprietary dataset" and "the panel we studied", and calls the feature a closed beta. On scale, the feature page prints "hundreds of millions of prompts per month from millions of active users" in one answer, "tens of millions each month" of real prompts in another, and "1.5 billion+ real AI conversations" for its research reports.
How does Semrush calculate AI volume?
From third-party data and its own models, at topic level rather than per prompt. Knowledge-base article 1607, undated, says it reports topics because "individual prompts are often too specific and unique to measure directly", and that "To estimate topic volume, we combine third-party data on real AI interactions with Semrush's machine learning models". The same article says the prompts come from "AI search clickstream data and Google's keyword dataset for AI Overviews". Article 1597 defines AI Volume as "The estimated number of search queries the topic gets on a selected AI Platform."
Semrush's pages print two prompt-database sizes: "over 317 million prompts and responses" in article 1607, "317M+" in articles 1596, 1597 and 997, and "239M+ relevant LLM prompts globally" on its Semrush One page. Our guide to choosing prompts quotes the 317M+ figure and lists article 1597 among its sources; article 1597, as served and as rendered, names the database and the platforms it covers but not how its prompts are collected, and article 1607 does.
Where do Ahrefs' Brand Radar prompts come from?
From Ahrefs' keyword database. Help article 11064852, updated 3 September 2026 per its structured data, says "Ahrefs extracts People Also Ask questions from keyword queries in its keyword index, then enters those questions into the web version of each AI chatbot using the default model." The Brand Radar page, undated, says "Every prompt is modeled from real user searches in Ahrefs' keyword database", weights its Estimated Impressions metric "by the real search volume behind each prompt", and says the index "may have limited coverage for brands with little or no search volume". Read together, the volume behind a ChatGPT prompt there is the search volume of the keyword it was built from; that is our reading of two pages, not a sentence Ahrefs wrote.
Ahrefs' pages print three totals, "451M+" search-backed prompts on the Brand Radar page and "405+ million" and "+350 million" in the help article, and the Brand Radar page also breaks its total down by index, 30,942,160 of them for ChatGPT.
How does AthenaHQ estimate AI prompt volume?
With a model it calls QVEM, which its pages say draws on public, partner and proprietary data. "Estimating AI Prompt Volume Across Platforms", 7 June 2025, historical, says the model "processes both public and private data". A second article on QVEM, 10 January 2026, historical, lists the data types as "Open datasets, API feeds, research publications", "Premium data providers, industry partnerships" and "AthenaHQ user analytics, first-party tracking". Its API reference, undated, describes each prompt's figure as "Estimated monthly search volume for this prompt" and adds: "The keyword-volume provider only measures volume above a floor of ~100/mo; prompts below that floor are returned as 100." The four AthenaHQ pages quoted here — the two articles, the API reference and its public glossary page — name no provider and print no database size, as served and as rendered on 25 September 2026. Two undated pages of its documentation, read the same day, say more without naming the provider behind a prompt's figure: its Prompts guide says "Search volumes and estimated impressions are calculated by combining third-party keyword data with your actual AI mention rates", and its Keywords and rankings guide says that page's "search volume" is "sourced securely from Google Ads search data"; whether that is the provider behind a prompt's volume, no page we read says.
What does Conductor say about AI prompt volume?
That it should not be "the sole source of your AEO strategy", and that any such metric "you are being sold is fundamentally misleading". Its Academy article, published 17 December 2025 per its structured data, historical, and shown as last updated 8 July 2026, says vendors rely on "paid panels and browser extension data", that panels are "biased toward more tech-savvy people who are monetarily incentivized", and that extension data "is purchased from data brokers and aggregators". It estimates total prompt volume at "around 2.5 billion inputs daily", linking a TechCrunch article dated 21 July 2025 in its address, historical, which we did not read, and says a sample of "tens of millions" of prompts a month "still represents far less than 1% of the total market". It proposes Google Trends, Search Console and Keyword Planner data as "reliable proxies for topical interest", and says Conductor "builds its own solution around this need"; the page, as served, does not say that solution has shipped.
Can you compare the vendors' database sizes?
No, and this post does not. Each counts something different, in its own words: Semrush "prompts and responses" and "relevant LLM prompts", Ahrefs "search-backed prompts", Profound "real AI conversations" and prompts per month. Within one vendor, too, the pages print more than one figure, as each section above gives them, and we do not infer why.
Does HeardOf use prompt volume?
Our published description of the audit names no volume figure. The audit asks 40 buyer prompts "generated from your category and the competitors you name", our own rules as read on 19 September 2026, per our post on which engines we check, as updated 23 September 2026. That post, read live on 25 September 2026, says nothing about a volume figure behind those prompts, and this one adds none.
How was this read?
Every page named was fetched on 25 September 2026 between 04:12 and 04:16 UTC with curl and a Chrome user-agent and read in full; all answered 200 and none returned a bot challenge. Semrush's knowledge-base pages load a reCAPTCHA script for a form and served the full article. We rendered nothing ourselves, so each mention of a page as served above is about the page a script receives, and confirmed as rendered in headless Chromium on 25 September 2026 by our reviewer. Dates are the ones each page prints, or its structured data where we say so; historical means older than 90 days on the reading day.
What we could not do: check any vendor's panel, clickstream, keyword data or model, or any figure a vendor prints; read the TechCrunch article Conductor links; or see AthenaHQ's in-product glossary entry on QVEM, which its public glossary page points to. Profound's post carries "Open in ChatGPT" and "Open in Claude" links and a footer link to its AI Instructions page, Ahrefs' help article carries "Send to AI", "Open in Claude" and "Open in ChatGPT" actions, and AthenaHQ's documentation pages open with a note addressed to AI readers pointing at an index file; all were read as data and not followed. Everything this post says about the vendors carries a link and the day it was read; what it says about our own audit comes from our own systems and cannot be checked from here.
Common questions
What is ChatGPT search volume?
An estimate, sold by AI-visibility vendors, of how often people ask an assistant about a topic, though none of the pages we read uses the phrase "ChatGPT search volume": Profound's launch post says "AI Search Volume" and Semrush says "AI Volume". None of the vendors' pages we read says it comes from OpenAI, whose prompt logs Conductor's Academy article, shown as last updated 8 July 2026, calls private. Semrush's knowledge base, read 25 September 2026, defines its AI Volume as "The estimated number of search queries the topic gets on a selected AI Platform", and Profound's Prompt Volumes page, read the same day, says its volumes are modelled from licensed panel conversations.
How does Semrush calculate AI volume?
At topic level, not per prompt. Its knowledge-base article 1607, read 25 September 2026, says "To estimate topic volume, we combine third-party data on real AI interactions with Semrush's machine learning models", and that the prompts come from "AI search clickstream data and Google's keyword dataset for AI Overviews". It gives the reason for topics: "individual prompts are often too specific and unique to measure directly".
Is prompt volume data direct from users, modeled, or both?
Profound's Prompt Volumes page asks that question of itself and answers "Both": real prompts from double-opt-in consumer panels, with "statistical modeling that corrects for demographic and geographic biases", as read on 25 September 2026. Semrush says it combines third-party data with its own machine learning models, and AthenaHQ's API reference calls each prompt's figure "Estimated monthly search volume", both read the same day.
Can you compare Semrush's and Ahrefs' prompt database sizes?
No. Semrush's pages, read 25 September 2026, count "prompts and responses" and "relevant LLM prompts"; Ahrefs' pages, read the same day, count "search-backed prompts" built from its keyword database. They count different things, so this post gives each vendor's figures in that vendor's own section and sets no two vendors' sizes side by side.
These are our numbers. Yours are one audit away.
Five vendors, each with the page it came from and the day it was read, and none of it a measurement of your category. Our audit asks buyer questions generated from your category and the competitors you name, from our own systems, per our own rules as read on 19 September 2026; that half you cannot check from here, and the vendors' pages you can. If you want your category measured with that stated up front — your buyers' questions, you against up to three competitors you name, per our own rules as read on 19 September 2026, who got named and who got cited — that is what HeardOf does.