The FrontStat method

AI recommendations

When a buyer asks ChatGPT, Claude or Gemini which tool to use, the assistant searches the web, reads a handful of pages and names the products it finds there. Whether your product is one of them is the newest check in launch week, and the one with the fewest people paying attention to it.

What we measure

FrontStat asks each assistant two questions: the buyer question for your category, phrased the way a buyer would ask it without naming you, and "What is <product> (<domain>)?". From the answers it records whether the assistant recommends you, which products it names instead, whether it can describe you correctly, and which pages it cited. The report quotes the answers.

Up to 25 points: 5 for each assistant that describes the product correctly, 3.33 for each that recommends it for the buyer question.

Why it matters

The assistants answer from pages their crawlers can read. OpenAI documents its crawlers and how OAI-SearchBot (used for ChatGPT search results) and GPTBot (used for training) can be allowed or blocked in robots.txt; Google documents Google-Extended, the token that controls whether your content is used for Gemini. A site that blocks these crawlers is choosing not to appear.

The pages an assistant cites for the buyer question are its shortlist. If none of them mentions you, the assistant has no way to name you, however good the product is.

What to do

Say what the product is in the first paragraph of the homepage, in plain words an assistant can quote. Add SoftwareApplication structured data with name, category, operating system and offers. Check robots.txt allows OAI-SearchBot, GPTBot and Google-Extended, or decide deliberately which to block.

Look at the pages the assistants cited in your report. The ones that accept listings, reviews or comparisons are where a fair mention gets you into the answer. Then publish a page that answers the buyer question directly and compares your product with the ones named, on facts.

llms.txt, a proposed file that summarises a site for language models, is cheap to add. Adoption by the assistants is not documented, so treat it as insurance rather than a lever.

Check your own launch. First report within 24 hours.