Method

How Presspool is built

Presspool keeps three records: niche maps of the companies in new AI markets, a page for every company, and a weekly roster of AI startup funding rounds. Every fact is taken from the public record and linked to its source, and every count of AI answers carries its questions, apps and date.

Niche maps and company records

A niche map covers one new AI market, such as physical AI or legal AI. A company joins a map when its own website describes the market as its main product and at least one independent source confirms it; some maps add a funding window, which the map states. Each map's definition, inclusion rule and exclusions are printed on it and versioned. A company appears only after both sources are checked, and every company on a map or roster has a record page listing its sources, its latest announced financing and its AI answer counts.

How AI answers are sampled

Each map has a fixed set of questions, some phrased the way people list companies and some the way buyers ask for recommendations. Once a week Presspool asks every question on ChatGPT, Claude, Perplexity, Gemini, Copilot, Google AI Mode and Google AI Overviews and records each answer. ChatGPT, Gemini and Copilot answers are collected from the live apps, and Google's from live Google results. Since October 8, 2026, Perplexity's come from Perplexity's own API (its Sonar model, with web search), because automated access to the Perplexity app failed across independent collection services; they can differ from what the app shows, and Perplexity counts from before that week, which came from the app, are never combined with them in one rate. No service collects answers from the Claude app, so Claude's come through Anthropic's own API with Claude's web search turned on, using Anthropic's Claude Haiku 4.5 model with up to two web searches per API request, and at most two continuation requests when an answer pauses mid-search (Claude Sonnet until October 6, 2026); they can differ from what the Claude app shows. A company counts when an answer names it or cites its website. Maps show the counts for the latest weekly sample, per app, with the date; an answer that could not be collected is shown as not sampled, never as zero. Each question counts once per app per week (Monday to Sunday, UTC): when it was asked more than once, the latest answer that came back is the one counted, a question no attempt got an answer to that week is not sampled, and asking again never adds to the number of answers. Presspool publishes only these counts and the sources the answers cited, never the text of an answer. Answers vary from run to run, so a count is a dated sample, not a verdict.

Between weekly samples, Presspool also follows the searches the AI apps themselves run and the related questions they suggest, which is how new questions and new companies come to its attention. Those extra answers inform what Presspool checks next; the published counts come only from the fixed questions, so week-to-week changes compare like with like.

Where the rounds come from

Candidate rounds come from a funding-announcement feed built on the reporting of TechCrunch, Tech.eu and EU-Startups, and from daily funding reports and company newsrooms. A candidate becomes an entry only after the amount, the round and the lead investors are confirmed at the company's own announcement or at the outlet that reported it, and both are linked on the entry. Amounts are written as announced, in the currency announced. Fund closes, standalone debt facilities and valuation-only news are not counted as startup rounds.

What counts as an AI company

A company is included when its own announcement describes the company, not one feature, as AI or as built for AI: models, AI infrastructure and compute, AI agents and the security and identity layers around them, or applications whose core work is done by AI. Companies that use AI as one feature among many are left out. The one-sentence description on each entry follows the company's own wording.

The AI citation column

The Machine Relations Index puts a fixed panel of buyer questions across 24 industry categories to ChatGPT, Claude, Gemini, Perplexity, Google AI Mode and Google AI Overviews on a recurring schedule and records the websites each answer cites. For each company, Presspool counts the recorded answers that cited the company's own domain within the Index window named on the page. The panel does not ask about companies by name, so the column measures whether engines reach for a company when a buyer asks about its market, not whether an engine can describe the company when asked directly.

Corrections

Entries are corrected in place, and the page's updated date changes when they are. To report an error, see Contact.

Who publishes Presspool

Rounds are listed because they were publicly announced, and no company pays to be listed. Who owns Presspool and how it stays independent of its owner's other work is set out on the About page.