AI Prompt Volume: How to Estimate What People Actually Ask ChatGPT

Type "search volume" into a conversation about SEO and everyone in the room knows what you mean: a number from Google Keyword Planner, or Semrush, or Ahrefs, that tells you roughly how many times a month people search a given phrase. Type "prompt volume" into the same conversation and watch the room split. Half assume it works the same way. It doesn't, and the gap between what people assume and what actually exists is costing content teams real planning time right now.

This post is about closing that gap. What the major AI-visibility vendors actually publish about prompt frequency, what they don't, the proxy signals you can pull together yourself without paying for any of them, and a step-by-step way to turn all of that into a prioritized list of prompts worth tracking and writing for.

Why there's no public "search volume" for AI prompts

Google Keyword Planner exists because Google owns the query logs and chose, decades ago, to expose an aggregated, anonymized slice of them to advertisers. That's the whole reason keyword volume data exists at all: one company decided to publish it, inside a tool built to sell ads against it.

OpenAI doesn't run an ad auction against ChatGPT prompts. Neither does Anthropic, and Google hasn't built an equivalent public tool for AI Mode or Gemini prompts the way it did for Search. There's no first-party "Prompt Planner." No vendor, however good its dashboard looks, has access to OpenAI's or Anthropic's raw server logs. Every AI-visibility tool on the market is working from something else: panels of consenting users, its own index of the web scaled against known search patterns, or a sample of queries it runs itself and extrapolates from. None of that is the same thing as Google handing out its own logs, and the vendors who are honest about this say so in their own documentation, once you read past the homepage.

That distinction matters more than it sounds like it should. A modeled estimate can still be useful for prioritization. It just can't be quoted the way you'd quote a Keyword Planner number in a client deck without a caveat attached.

What the vendors actually offer, read from their own pages

Profound's Prompt Volumes

Profound's Prompt Volumes feature is explicitly built on consumer-panel data rather than platform logs. Profound's own page describes the source as "double opt-in consumer panels of real answer engine users," covering ChatGPT, Gemini, Claude, and Perplexity, with more engines described as coming. The company says it processes billions of data signals daily and handles hundreds of millions of prompts a month, and that it layers "statistical modeling that corrects for demographic and geographic biases" on top of the raw panel data to make the numbers representative of a wider population than the panel alone.

Worth flagging directly: panel-based data starts somewhere, and Profound's own page states that availability begins in January 2025. If you're trying to understand prompt behavior from before that window, this isn't the source. Geographic coverage, as of this writing, spans the US, Canada, Italy, Brazil, Germany, Australia, Spain, South Korea, France, and the UK, refreshed weekly with under a week of latency according to the product page.

Semrush's AI Visibility Index

Semrush's AI Visibility Index is a different kind of product, and it's worth being precise about what it is and isn't. It tracks brand mentions and domain citations inside AI-generated answers, drawing on what Semrush describes as 180 million-plus AI answers across 22 industries and 9 markets, covering ChatGPT, Google AI Mode, Google AI Overviews, and Gemini. That's mention-frequency and citation-share data, not a prompt-volume number. If you're looking at the Index expecting something like "how many people ask this question," you're looking at the wrong report; it answers "how often does a given brand show up when that question gets asked," which is a related but separate question. Semrush's own positioning distinguishes the two explicitly, noting that being mentioned and being cited are different kinds of visibility worth tracking separately.

Ahrefs Brand Radar

Ahrefs' Brand Radar sits closer to Profound's territory. It offers two tiers: Custom Prompts, where you define specific buyer questions and track daily responses (starting at $50/month as of October 2026, per Ahrefs' own pricing), and an AI Visibility Index tier at $199/month that includes 83 daily prompts and 2,500-plus monthly checks with no setup required. Ahrefs states its modeling draws on its own web index plus prompts modeled from real user searches already sitting in its keyword database, and it publishes specific scale numbers for AI Overviews (268.8 million monthly prompts, by its count) and AI Mode (91.7 million monthly), alongside coverage of ChatGPT, Gemini, Perplexity, and Copilot. The honest caveat here, stated plainly by Ahrefs itself: the tool works best once your brand already has some presence in AI answers, which is a real limitation if you're trying to use it to find brand-new opportunity space rather than monitor existing visibility.

Otterly's "AI search volume"

Otterly has published a blog post specifically framing this problem the way most readers already think about it: can you get something like a Google-style volume number for ChatGPT and Perplexity queries. The post states directly that "platforms like OpenAI/ChatGPT and Perplexity currently do not share any information or metrics" on query frequency, which is the same baseline fact behind everything else in this post, then describes Otterly as having introduced a method for estimating these volumes. The post doesn't disclose the mechanics of that estimate, and it describes the capability as early-stage. Treat it as a planning signal, not a number you'd defend in an audit.

A quick honest comparison

None of these is wrong to use. All of them are estimates built on top of something other than the platform's own logs, and the vendor pages say as much if you read the actual product documentation instead of the marketing header above it.

If you want a broader rundown of tools in this category beyond prompt-volume specifically, the best AI visibility tools roundup and Profound alternatives both cover adjacent ground.

The proxy methods, for when you don't want to pay for any of it

You don't need a dedicated prompt-volume tool to build a reasonable, prioritized list of what to write for. You need a handful of signals that correlate with how people phrase questions to a chatbot, stitched together carefully.

Google keyword volume for question-form queries. People who ask ChatGPT "what's the best accounting software for a five-person agency" are drawn from roughly the same population asking a messier version of the same thing on Google. Pull keyword volume for the question-form variants of your core topics (who, what, best, how to, vs., alternative to) from whatever keyword tool you already have access to. It's not a 1:1 translation to prompt frequency, but it's a real, measured number about real human curiosity on the same topic, which is more than most prompt-volume estimates can say for themselves.

People Also Ask boxes. These are Google's own harvested real-question data, shown to you for free on every relevant SERP. They tend to overlap heavily with the kind of specific, conversational phrasing that shows up in chatbot prompts, because both are drawing from the same underlying pool of actual human questions rather than keyword-tool-generated variants.

Reddit and Quora thread frequency. Search a topic on Reddit or Quora and count how often a specific question phrasing, or a close variant of it, recurs across threads and comments. This is slower than pulling a number from a dashboard, but it's genuine frequency data about how people phrase a question when nobody's optimizing their wording for a search box, which is closer to how people type into a chatbot than how they type into Google.

Your own site search queries and sales-call questions. This is the most underrated source on this list and it costs nothing. Whatever people type into your own site's search bar, and whatever your sales team gets asked on calls every week, is a near-perfect match for the real, specific, sometimes oddly phrased questions your actual buyers have. If a prospect asks your sales rep "does this integrate with our existing payroll system" four times this month, that's a stronger prompt-volume signal for your business specifically than any aggregate panel number from a tool that's never spoken to one of your customers.

Search Console query data. Look for question-form queries with meaningful impressions but a depressed click-through rate. That pattern (people are searching this, Google is showing something, nobody's clicking through) is often a sign the answer is already satisfied in a snippet or an AI Overview, which also tells you it's exactly the shape of question a chatbot would answer directly rather than send someone off-platform for.

GA4's AI Assistant channel and landing pages. GA4's default channel grouping includes a dedicated "AI Assistant" category, and Google's own documentation defines it precisely: the channel "by which users arrive at your site from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok," explicitly excluding Google's own AI Overviews and AI Mode traffic, which get their own classification. Pull the landing pages receiving AI Assistant traffic in GA4 and you're looking at pages that have already earned a citation or a link-out from a chatbot answer, which tells you two things at once: which of your existing pages are already matching some prompt out there, and by extension, roughly what that prompt probably looked like.

None of these six signals is a volume number on its own. Stacked together, cross-referenced, and weighted by how directly each one maps to your actual buyers (sales-call questions and your own site search should usually outweigh a generic PAA box), they get you a genuinely useful, defensible prioritization, which is the actual job here, not chasing a number that doesn't exist yet.

How query fan-out changes what you're even counting

Here's the complication that makes "prompt volume" a fuzzier concept than keyword volume ever was, even in principle. Query fan-out, the technique Google has publicly described for AI Mode, means a single prompt doesn't map to a single search. Google's own language, from its Search blog, is that AI Mode works by "breaking down your question into subtopics and issuing a multitude of queries simultaneously," so one prompt a person typed becomes several queries running behind the scenes, each one potentially surfacing a different page as a source.

That means the unit you're trying to count isn't stable. A prompt volume number, even a perfectly accurate one, tells you how often a question got asked. It doesn't tell you how many sub-queries that question generated, which of those sub-queries your content could realistically win, or how the model stitched the results back together into the answer a person actually saw. Two prompts with identical "volume" can fan out into completely different numbers of retrieval opportunities depending on how broad or narrow the question is. A prompt that sounds simple ("best CRM for a solo consultant") can fan out into several angles (pricing comparison, feature comparison, reviews, alternatives) that a prompt-volume number alone would never surface.

Practically, this means the goal of building a prompt list isn't just "rank these by frequency." It's building prompts broad enough that you can reasonably guess at the sub-topics a fan-out system might generate from them, then making sure your content actually answers each of those sub-topics as a self-contained, citable unit, not just the surface-level question.

A step-by-step method for building a prioritized prompt set

Step 1: Harvest raw candidates from every proxy source above. Don't filter yet. Pull question-form keyword data, PAA boxes, your own site search log, sales-call notes, Search Console impressions-high/CTR-low queries, and GA4 AI Assistant landing pages, and dump everything into one sheet. Expect 150 to 400 raw candidates for a mid-sized topic area.

Step 2: Dedupe and cluster by intent, not by exact wording. "Best project management tool for remote teams" and "what project management software works for distributed teams" are the same underlying prompt. Cluster aggressively around the actual question being asked, because a chatbot's fan-out process is going to collapse wording variants into the same retrieval intent anyway.

Step 3: Score each cluster on three things. A Google-volume proxy (from the question-form keyword pull), a buyer-proximity signal (does this match something your sales team or site search actually sees, weighted higher than generic web data), and a coverage gap (do you already have a page that directly answers this, or is this wide open). A cluster that scores high on buyer proximity and low on coverage is your highest-priority write.

Step 4: Validate a sample manually across two or three AI platforms. Before committing a content calendar to a cluster, actually type the top five or six candidate prompts into ChatGPT, Gemini, and Perplexity yourself and read what comes back. This catches the cases where a topic looks huge on paper but the models already answer it fine from general knowledge with no citation opportunity at all, versus topics where the models visibly struggle or cite thin sources, which is where you actually have room to win a citation.

Step 5: Lock the validated set as your tracking list. This becomes the fixed prompt set you run on a schedule, which is the same foundational list the KPIs that actually matter for AI visibility depends on; a prompt set that changes every month produces noise instead of a trend line. Revisit and prune it quarterly, not weekly.

A worked example (hypothetical, not a real brand)

To make Step 2 and Step 3 concrete, here's a hypothetical prioritization table for a made-up B2B SaaS product, a project-scheduling tool we'll call "Taskwell." None of the numbers below are real measured data; they're illustrative placeholders showing the shape of the exercise, not figures pulled from any tool.

The pattern worth noticing: the highest-priority items aren't necessarily the ones with the biggest Google-volume proxy. The QuickBooks integration question has no measurable Google search volume at all in this hypothetical, but it scores priority 1 because it's a direct, recurring, specific question from actual buyers with zero existing coverage. That's the whole point of weighting buyer-proximity signals above generic volume proxies: a prompt-volume number you can't get is a worse reason to skip a topic than a strong, repeated signal from the people who actually buy your product.

FAQ

Is there any tool that shows real ChatGPT search volume, not an estimate?

Not publicly, as of October 2026. Every vendor covered here, including Profound, Semrush, Ahrefs, and Otterly, builds its numbers from panels, sampled answers, or proprietary modeling rather than OpenAI's actual server logs, because OpenAI hasn't published that data or built a public tool exposing it the way Google built Keyword Planner.

Should I just use Google keyword volume as a stand-in for prompt volume?

It's a reasonable proxy for question-form, informational queries, not a direct substitute. The overlap is real because people researching the same decision tend to phrase similar questions whether they're typing into Google or a chatbot, but the populations aren't identical, and purely conversational or highly specific prompts (the "does X integrate with Y" type) often have no meaningful Google-volume signal at all despite being asked constantly in a chatbot.

How often should I refresh my prompt priority list?

Quarterly for most categories, monthly if you're in a fast-moving space where new competitors, features, or terminology show up often. Refreshing weekly tends to produce noise rather than a usable trend, since you're measuring small-sample fluctuation rather than real shifts in what people are asking.

Do I need a paid tool to get started, or can I do this with proxies alone?

You can build a genuinely useful prioritized list with proxies alone, especially if your own site search and sales-call data are rich (most B2B companies have more of this than they realize and never mine it). A paid tool becomes worth it once you want repeatable, scheduled tracking of how often your brand specifically shows up against that prompt set, which is a different job than building the initial list.

The actual output of this process shouldn't be a spreadsheet you admire once and never touch again. It's a living, quarterly-refreshed list of real questions, weighted by who's actually asking them, that tells your writers exactly what to cover next and tells your reporting exactly what to check against once that content is live.