Insights
What building an AI-visibility product taught me about how AI recommends businesses
When someone asks ChatGPT "who fixes furnaces near me," the AI names a few businesses and skips the rest. Everyone in marketing has a theory about how that selection happens. Running Proof Signal — where we audit and fix real small-business websites every week — means we get to test those theories against outcomes instead of arguing about them.
A few things have held up consistently.
Access comes before everything
The most common problem we find isn't subtle. It's that AI crawlers can't read the site at all — blocked by a robots rule someone set years ago, hidden behind JavaScript that never renders as text, or simply absent because the business has no site. No amount of content strategy matters if the crawler bounces off the front door. This is also the least glamorous fix and the highest-leverage one.
AI rewards plain statements of fact
AI tools are trying to answer a question: what does this business do, where, for whom? Sites that say "we repair furnaces and boilers in Elgin and the northwest suburbs, same-day service" get understood. Sites that lead with "Excellence. Integrity. Solutions." do not. The copywriting instinct that served brand marketing for decades — evocative, abstract, mood-driven — actively hurts here. Specificity is the new persuasion.
Trust signals are boring and cumulative
Reviews, a real address, a phone number that matches everywhere it appears, photos that are obviously yours, direct answers to common questions. None of these is decisive alone. Together they form the pattern AI tools read as "this business is real and safe to recommend." There is no single trick, which is inconvenient for people selling tricks.
It decays
The models change, the crawlers change, and your competitors are working on this too. Sites we fixed a year ago that stopped paying attention have slipped. The businesses that keep getting named are the ones that treat AI visibility like bookkeeping — a steady practice, not a one-time project. That finding shaped Proof Signal's whole business model: audit once, but monitor forever.
The meta-lesson
The strategy consultant's version of this essay would have been written from analyst reports. This one comes from shipping fixes and watching what happens. That difference — advice grounded in operating a real product against a real market — is the whole reason I build alongside advising.