Why AI Search Visibility Starts Off Your Own Website
When a local service business follows the usual checklist for showing up in AI assistants — an FAQ section, schema markup, a few articles written "for AI" — and still doesn't appear when someone asks ChatGPT or Gemini for the best plumber in their city, the website is rarely the main problem. That is the central finding from Veilvex, a studio working on websites, apps, ads, SEO and AI search visibility, in an interview about its work in GEO and AI search.
The real problem is often off-site
Asked about a time a client came to them after the usual advice had not worked, the team described a recurring pattern. Assistants that search the web build their answers largely from pages that already rank for the question: directories, review platforms, local roundups and comparison articles. If a business is missing from those sources, or appears there with an old name, a different phone number or a vague category, the assistant has nothing consistent to repeat.
Sometimes the site itself was also blocking AI crawlers in robots.txt, often through a security plugin setting nobody had checked. In the team's own review of 157 small-business websites from Google's first page, 12 — 8% — blocked crawlers such as GPTBot or Google-Extended.
What ranking pages leave out
The second question asked what keeps coming up for those clients that the pages already ranking for that search leave out. The answer was plain facts a buyer needs to decide. In the same review, 48% of the sites showed no prices and no link to a price page. Many don't say clearly which areas they serve, how quickly they can come or what a typical job includes. About half had no WhatsApp or Telegram link, so the only next step is a call or a form.
As the team put it, an AI assistant is answering something like "who can fix this today and roughly what will it cost". A page that states the service, the area, a price range and how to book in a few plain sentences gives it something to quote. Pages that rank are often written for the keyword, not for that question.
Three things people decide too quickly
The final question asked what people decide too quickly when they first try to fix this themselves, and what they usually have to undo later. The team named three things.
First, they rewrite the whole site into long FAQ blocks that ask the same question in different words. It reads like filler to people and adds little for AI, so it usually gets cut back later.
Second, they judge results from one prompt on one day. Answers change with wording, location and time, so a single "ChatGPT didn't mention us" is not a measurement. A fixed set of buyer questions, checked across several assistants for a few weeks, tells you much more.
Third, they treat AI search as separate from Google. Assistants that search the web still lean on pages that rank, so consistent listings and the basics of local SEO come first, and skipping them means coming back to them later.
Practical takeaways
For businesses trying to improve their visibility in AI answers, the interview points to a sequence: make sure the site is not blocking AI crawlers, keep directory and review listings consistent, and put the facts a buyer needs — service, area, price range and how to book — in plain language on the page. Measurement should be a fixed set of buyer questions checked across several assistants over a few weeks, not a single prompt. And local SEO fundamentals remain the base that AI search builds on.
The figures cited come from Veilvex's study of small-business websites, published with open data on the Veilvex blog. More about the studio's work is available at veilvex.com.
For readers new to the topic, Google's guidance on AI features and site controls and OpenAI's documentation on GPTBot explain how crawler access works, while Schema.org covers structured data basics.