Every AI visibility programme starts with the same decision: which prompts should we track? Get it wrong and every chart that follows measures the wrong thing. Prompt Research Reports replace the guesswork with evidence drawn from the largest dataset of real AI conversations in the industry.

Key takeaways

  • Reports are grounded in 1.5+ billion real user prompts from Flowgen's proprietary dataset, covering the last six months of conversations.
  • A four-stage pipeline retrieves, filters, clusters and selects prompts, turning millions of noisy variations into stable, meaningful topics.
  • Each report shows your input configuration, an intent breakdown (informational, commercial, transactional) and discovered topics ranked by volume with coverage scores.
  • New customers use it to build a first configuration from data; existing customers use it to find the topics they are missing.

The pitfalls of traditional AI prompt tracking methodologies

Most teams assemble their prompt list from assumptions. They borrow keywords from SEO tools, lift phrases from social listening, and add whatever the product team believes customers ask. Each of those is a proxy for how people talk to AI, and none of them is the real thing. A keyword is not a conversation, and a list built from guesses will miss the questions buyers actually ask while over-weighting the ones marketers imagine.

Flowgen's approach is different because the input is different: more than 1.5 billion real prompts that people typed into ChatGPT, Claude and other engines. Configurations built on that data reflect verified behaviour rather than speculation.

How Flowgen Prompt Research Reports work

Each report runs through four sequential stages:

1. RetrieveReal conversations from the last 6 months across ChatGPT, Claude and other engines
2. FilterCandidates scored for relevance to your category and keywords
3. ClusterPrompts grouped into meaningful, stable topics
4. SelectRepresentative prompts chosen for highest coverage

The result is a set of tracking categories that are coherent and stable over time, instead of millions of near-duplicate phrasings that would be impossible to monitor directly.

1.5B+Real user prompts in the dataset
6 monthsOf conversations analysed per report
4 stagesRetrieve · Filter · Cluster · Select

Features of a Prompt Research Report

A report has three views. The first shows the input configuration: the category, keywords and brand that defined the search. The second breaks the matched conversations down by intent, so you can see how much of the demand is informational, commercial or transactional. The third lists the discovered topics ranked by volume, each with a coverage score.

Coverage is the number to watch. It tells you what share of real conversations in your category a given prompt represents, so you can prioritise tracking the prompts that account for the most demand rather than the ones that merely sound right.

Example: discovered topics ranked by coverage
Share of relevant real conversations each topic represents (illustrative, B2B software category)
  • Best tools for [category]24%
  • [Brand] vs [competitor]17%
  • Pricing and plans13%
  • Integrations with [tool]9%
  • How to set up [task]7%
  • Is [brand] worth it?5%
Illustrative values. Real reports are generated per category.

The work behind your report

Organising millions of varied prompts into topics that are both accurate and useful is a hard modelling problem. General-purpose models tend to over-simplify, producing a few catch-all clusters that lump together prompts with very different intent. Flowgen's clustering, engineered by our data science team, is tuned to find the meaningful distinctions inside a topic, so that "best CRM for a startup" and "best CRM for enterprise sales teams" land in different buckets that reflect different points in the customer journey.

How brands benefit from Prompt Research Reports

Three outcomes come up again and again: a data-backed starting configuration for teams new to Flowgen, a clear view of coverage gaps for teams already tracking, and an understanding of how buyer intent is distributed across the category. Together they guide both which prompts to track and which content to build.

Build your first prompt configuration from real data, not guesswork

For new customers, the report is the factual foundation for the first configuration. Working with their account manager, teams review the keywords and topics the report surfaced and build or refine their initial tracking set from verified conversation data rather than intuition. The first dashboard you see is already measuring the right thing.

Spot the topics you are missing

For existing customers, the report compares what you already track against what people actually ask. Coverage metrics show, topic by topic, where your configuration is strong and where it is silent, which makes finding gaps in the current setup a matter of minutes rather than a quarterly review.

Example: intent breakdown for one category
  • Conversations100%
Informational 52%Commercial 33%Transactional 15%
Illustrative values.

Know whether your buyers are researching or ready to buy

The intent breakdown classifies conversations by whether people are learning, evaluating or ready to purchase. Knowing that distribution tells you where to invest: a category dominated by informational prompts rewards educational content, while one heavy with commercial comparisons rewards comparison and proof pages. Either way, you are putting visibility exactly where your audience makes its decisions.

Choose prompts the way your customers choose words: from the conversations they are actually having.

Get started

The Prompt Research tool is available today for all Enterprise customers, self-serve, in the Prompt Volumes tab in Flowgen. Your account team can help with questions about generating a customised report. Learn more on the Prompt Volumes feature page or get a demo.

Alireza Farhadi & Helen MartinMember of Technical Staff and Product Marketing, Flowgen