Prompt Volumes: How to Discover What Your Customers Are Actually Asking LLMs

Every month, your potential customers type billions of questions into ChatGPT, Gemini, Claude, and Perplexity. ChatGPT alone processes 2.5 billion prompts per day as of mid-2025 , and the platform reached 900 million weekly active users by early 2026 . These aren't idle queries. They're product comparisons, purchasing deliberations, and problem-solving conversations that shape buying decisions before anyone visits your website. The problem: you can't see any of it. Tools like ChatGPT do not yet offer first-party analytics data to pull prompts from, the way Google Keyword Planner surfaces search queries . While Google Search Console tells you exactly which keywords drive clicks, LLM platforms offer no equivalent dashboard. Your customers are having consequential conversations about your category, and you're operating blind. This gap is what makes "prompt volumes" the emerging metric that matters. Prompt Volumes measures AI prompts and answers-what users ask assistants and how topics shift . Understanding this data is the foundation of any Generative Engine Optimization strategy, and the brands that build this intelligence now will shape how AI represents their category for years.

Why Traditional Keyword Data Misses 70% of What People Ask AI

The assumption that you can port your Google keyword strategy into AI optimization is dangerously wrong. Semrush's analysis of 80 million clickstream records found that only 30% of ChatGPT queries match traditional search patterns, and that users write longer, more detailed prompts averaging 23 words versus 4.2 words with search enabled . That 70% gap is enormous. Only 30% of ChatGPT prompts fell into standard intent categories like navigational, informational, commercial, and transactional-meaning 70% of the prompts are unique and rarely seen in classic search engines . People aren't typing "best CRM software." They're writing multi-sentence requests with constraints, context, and follow-up conditions. These longer prompts reflect a fundamentally different behavior: consulting rather than searching.

Traditional tools optimize for search queries and rankings in Google, but queries to Answer Engines are typically longer, contain follow-up questions, and are generative in nature . Someone asking ChatGPT for a CRM recommendation might include their team size, budget, integration requirements, and previous tool frustrations-all in one prompt. A four-word keyword can't capture that intent depth.

The Query Fan-Out Effect

The gap widens further when you account for how AI engines process queries internally. When a user types a question into AI Mode, the AI model breaks down that query into multiple search queries around related subtopics. Query fan-out looks at the "subintents" behind a search query-for example, "best sneakers for walking" could break into subqueries like "best sneakers for men," "best sneakers for walking in different seasons," and "sneakers for walking on a trail" . This matters because this multi-vector retrieval strategy forces LLMs to pull evidence from multiple passages and documents rather than relying on a single high-ranking page, resulting in a fundamental break from keyword-based ranking . Your content doesn't just need to match the original prompt. It needs to answer the invisible subqueries that the AI generates behind the scenes. Mike King of iPullRank built a free open-source tool called Qforia specifically to reverse-engineer these fan-outs. Qforia leverages the same Gemini model used by AI Overviews and AI Mode , generating 20–30 related subqueries for any seed prompt. In a small experiment by Semrush, optimizing for fan-out queries more than doubled AI citations in tracked prompts-from two to five .

The Tools and Data Sources That Actually Reveal Prompt Volume

No single tool gives you the complete picture. But combining several sources creates a workable intelligence layer.

Dedicated Prompt Volume Platforms

The first category is purpose-built tools that estimate how frequently specific questions appear across LLM conversations. Profound's Conversation Explorer is the most data-rich option for enterprises. It offers access to 400M+ anonymized conversations showing what users actually ask AI systems . Profound licenses conversations from multiple double-opt-in consumer panels, with data anonymized, aggregated, scrubbed of PII, and compliant with GDPR and CCPA. Scale is in the hundreds of millions of prompts per month from millions of active users . The platform currently has data from ChatGPT, Gemini, Claude, and Perplexity conversations, with additional engines coming, updated on a rolling basis with at-most one-week delay . AthenaHQ takes a different approach. The platform brings technical credentials with a founding team including ex-Google Search and DeepMind engineers, backed by Y Combinator . Its Query Volume Estimation uses proprietary models to predict AI search volume for different queries, helping prioritize tracking prompts with actual traffic potential . For teams on tighter budgets, Similarweb's Prompt Analysis stands out because most prompt tracking tools build their prompt libraries synthetically, generating questions from keyword research or predefined templates, whereas Similarweb's is built from real user queries . That distinction between synthetic and organic prompts is one most buyers overlook.

AI Visibility Toolkits From Established SEO Platforms

Semrush's AI Visibility Toolkit allows prompt tracking across LLMs, while Promptmonitor offers the most comprehensive platform coverage at 8+ LLMs including ChatGPT, Claude, Gemini, Perplexity, Google AI Overview, AI Mode, DeepSeek, and Grok, at an affordable $29/month . For a completely free option, Mangools AI Search Grader tracks 8 AI models with no paid tiers . Each tool has blind spots. Enterprise platforms like Profound provide actual conversation data but carry enterprise pricing. Lightweight trackers offer broad LLM coverage but rely on synthetic prompt generation. The practical move is to start with one tool and layer in others as you learn which prompt categories matter most for your business.

A Five-Step Workflow for Discovering Prompts That Matter

Knowing tools exist is different from knowing how to use them. Here's the practitioner workflow that produces actionable prompt intelligence.

Step 1: Mine Your Own First-Party Data

Start with what you already own. If your site has a search bar, it's already collecting queries that resemble prompts. Search providers like Algolia not only provide search technology for your website but also give you insights into what your website users search for . Internal site search logs, customer support tickets, and sales call transcripts contain the exact language your customers use when they're stuck, comparing, or evaluating.

Sales teams sit on a goldmine of prompt-like language. From early discovery calls to RFPs to demo recordings, buyers reveal what they're researching, struggling with, or trying to compare . Export your support ticket subjects from the past six months, group them by theme, and you'll have a starter set of prompts that no competitor can replicate because they originate from your actual buyers.

Step 2: Audit AI Responses for Your Category

Before tracking prompt volumes, you need to understand the current landscape. Go into ChatGPT, Claude, Perplexity, or Gemini and ask the questions your customers ask: "What's the best [product] for [use case]?" "Is [competitor] worth it?" "Where should I buy [category]?" Look at which sources show up in the answers. Map 5–10 of these queries and track which sources keep appearing . Document three things for each response: whether your brand appears, what the AI says about competitors, and which source URLs the AI cites. This manual audit takes 30–60 minutes but reveals more strategic insight than any dashboard, because you'll see exactly how the AI frames your category narrative.

Step 3: Expand With Community and Forum Intelligence

Reddit is a place where people ask honest, straightforward questions-the same kind of language they might use with ChatGPT. These questions often reflect how people naturally phrase prompts . And the connection goes deeper: AI models treat query-specific subreddits as subject matter experts. For any given prompt, Answer Engines choose 3–5 key subreddits to be the primary sources of truth . Tools like AnswerThePublic and AlsoAsked help here too. These tools show question clusters from Google search data that are visual, intuitive, and perfect for identifying "people also ask" style content-which often aligns with how users phrase prompts . For a more advanced approach, if you want prompt data from people not already in your funnel, platforms like Pollfish and Wyntner let you target specific ICP segments and ask them directly what they'd ask an AI when researching tools in your category .

Step 4: Classify Prompts by Scope, Intent, and Funnel Stage

Raw prompts become strategic only when classified. Capture three prompt scopes: brand mentions (your name, products, pricing, alternatives), category/problem prompts (jobs-to-be-done queries where you should appear), and competitor-comparison prompts (vs. questions that influence switching) .

Non-branded prompts are where the real competitive signal lives. These are category-level questions with no brand name attached. Whether your brand appears in prompts like these shows whether AI actually sees you as an authority in the category . Tag each prompt by journey stage: awareness ("What is [concept]?"), consideration ("Best tools for [use case]"), and decision ("Is [product] worth it?"). This forms your Prompt Funnel, aligned to the user journey from awareness to decision. Not all prompts are equally valuable -focus resources on consideration and decision prompts where purchase intent runs highest.

Step 5: Track Visibility Over Time, Not Single Snapshots

This step is where most teams stumble. A one-time check of whether ChatGPT mentions your brand is nearly meaningless. SparkToro research found that AI tools produce different brand recommendation lists more than 99% of the time when given the same prompt. The same list in the same order appears less than 0.1% of the time . That finding reset the industry's understanding of AI visibility measurement. Repeat presence means something. Exact rank does not . The metric that matters is visibility frequency: what percentage of runs for a given prompt category surface your brand?

If you really want to know an AI's set of recommendations, you need to ask over and over again-usually at least 60–100 times-then average the results . Most teams can't do this manually, which is where automated prompt tracking tools justify their cost.

Why AI Visibility Is Probabilistic, Not Positional

This is the hardest mental shift for teams trained on SEO. There is no "position one" in AI search. Unlike keyword rank tracking, which tells you where a specific page ranks, prompt tracking tells you whether your brand is mentioned at all in an AI-generated answer. There is no Page 1 or Page 2 in AI search. A brand is either in the answer or it is not .

The SparkToro/Gumshoe team ran 2,961 prompts across ChatGPT, Claude, and Google Search AI across 12 categories including chef's knives, headphones, and digital marketing consultants. Each prompt was run 60–100 times per platform. Nearly every response was unique in three ways: the list of brands presented, the order of recommendations, and the number of items returned . But here's the nuance that makes tracking worthwhile despite the volatility. Despite phrasing differences, a handful of headphone brands (Bose, Sony, Sennheiser, Apple) appeared in 55–77% of 994 responses to varied, user-written prompts . The individual responses are chaotic. The aggregate pattern is meaningful.

The smaller the market, the more stable the results. In tight spaces-like regional service providers or niche B2B tools-AI answers clustered around a few familiar names. In massive categories-like novels or creative agencies-results scattered into chaos. More options create more randomness . The implication for prompt volume research: category specificity matters more than raw volume. A niche B2B prompt with modest volume but tight competitive sets will yield more predictable visibility than a high-volume consumer query where the AI cycles through dozens of brands.

How to Structure Content That Answers the Prompts You Discover

Discovering prompts is research. Making your brand appear in the answers is execution. The two require different skills.

Match Content Architecture to AI Retrieval Patterns

LLMs seek salient, distinctive, non-generic content to surface in answers. Redundant or boilerplate content is more likely to be filtered out. LLMs often prioritize precise data points and statements of fact in their synthesis. Content that clearly embeds verifiable data is more likely to be selected and cited . Structure your content so individual passages can stand alone. AI engines extract passages, not pages. Each section under a heading should contain a self-contained answer that an LLM could quote without needing surrounding context.

Turn frequent prompts into H2s or FAQ blocks to mirror user language exactly. Match the tone of real user prompts-use "how" and "why" phrasing . If your prompt research reveals that customers ask "How do I migrate from [Competitor] to [Your Product]?", that exact question should be a heading on your site, followed by a specific, data-rich answer.

Build Topical Clusters Around Prompt Families

Prompt research expands the scope of content strategy beyond ranking individual pages to clusters of related questions. For SEO, that means ensuring content covers the full topic landscape. For GEO, it means ensuring content provides the context generative systems need to synthesize answers . A single blog post won't win AI citations for an entire topic. You need interconnected content: a foundational guide, comparison pages addressing follow-up prompts, and FAQ pages covering the long-tail questions that fan-out generates. The foundational guide captures informational search demand, while supporting and comparison content addresses follow-up prompts users ask as they explore the topic-and over time, the content appears in both traditional search results and AI-generated answers .

Earn Third-Party Credibility Where AI Models Source Trust

Content alone isn't enough. Domains with millions of brand mentions on Quora and Reddit have roughly 4x higher chances of being cited than those with minimal activity. Domains with profiles on platforms like Trustpilot, G2, Capterra, Sitejabber, and Yelp have 3x higher chances to be chosen by ChatGPT as a source . This means your prompt volume research should inform not just content creation but also your broader digital presence strategy. If prompts reveal that customers ask comparison questions, make sure those comparisons exist on review platforms where AI models look for consensus signals.

The Metrics That Matter (and the Ones That Don't)

Track these four metrics to measure whether your prompt volume research is translating into results:

• Visibility frequency: Percentage of runs where your brand appears for a tracked prompt set. Run prompts 60+ times per platform to get statistically meaningful data.
• Citation rate: How often your specific URLs are cited in AI responses-distinct from mere brand mentions.

ChatGPT only cites 15% of the pages it retrieves; 85% of the sources retrieved during a user's search are never cited . - Multi-engine coverage: A brand that monitors only one or two AI engines may believe its visibility is healthy, while in reality it is absent from the majority of AI-powered discovery channels that its customers use . - Sentiment accuracy: Is the AI saying something true and favorable, or is it surfacing outdated pricing, discontinued features, or competitor-favorable framing? What not to track: position rankings in AI responses. Ranking positions are so unstable they're effectively meaningless. Any product selling AI rank movement is selling fiction .

Playing the Long Game: Prompt Intelligence as a Strategic Asset

Prompt volume data is still immature. AI platforms are only driving an average of 1% of overall web traffic across 10 major industries . That number is small. But the trajectory is unmistakable- AI referrals were up 357% year-over-year by June 2025 , and ChatGPT-sourced visitors show double the session duration and 33% higher conversion than the average visitor .

Visibility in AI search is inherently volatile. AirOps research found that only 30% of brands stay visible between consecutive answers. Just 20% remain present across five consecutive runs . The brands that maintain consistent visibility are the ones doing the work to understand prompt patterns, build content that addresses those patterns, and measure outcomes over time-not in single snapshots. Prompt volume research won't replace keyword research. It extends it. The companies building prompt intelligence today are assembling a strategic asset that compounds: every month of data reveals seasonal trends, emerging questions, and competitive shifts that inform content investment. Those that wait for AI search to "mature" before paying attention will find the conversation has already been shaped-by competitors who bothered to listen to what their customers were actually asking.