An AI thought we were a dark-kitchen agency. Fixing that is now part of SEO.

Earlier this year an AI answer engine described GOAT Pixels as a "dark kitchen agency". We build for food businesses, among others, and somewhere in the model's reading of the web the two things fused. Nobody at the company has ever cooked anything professionally.

It is a funny story until you realise that a growing share of buyers now start with a question typed into ChatGPT, Perplexity or Google's AI Overviews — and the answer they get is a summary of what the machine thinks you are. If it is wrong, you never find out. The prospect just shortlists someone else.

Why this happens

Search engines rank pages. Answer engines describe entities — companies, people, products — and they build that description from every mention they can find: your site, directories, LinkedIn, old press, a client's case study, a blog post that used your name in the wrong sentence. When your own site is vague about what you are, the other sources win.

Most agency sites are vague. "We create digital experiences that matter" tells a language model nothing it can quote.

What we changed on our own site

One source of truth for the facts. Name, legal name, founding year, Chamber of Commerce number, city, the services in plain words, the profiles we control. It lives in one file and feeds both the page copy and the structured data, so the two cannot drift apart.

A single entity with a stable identifier. The company is one ProfessionalService in JSON-LD with an @id, and every article, case study and service page points back to that same @id as its publisher. One node, many references — the shape machines are built to read.

Answer-shaped pages. Each service page opens with what it is, who it is for and what it includes, in the first paragraph. Headings are the questions people actually ask. Numbers have sources. We removed every sentence that only existed to sound like an agency.

Crawlable by default. Our homepage is a WebGL galaxy rendered in text characters — and it sits on top of a page that is fully rendered on the server. Every crawler, including the AI ones, gets the complete content without running a line of our JavaScript. If your site only exists after the JavaScript runs, to an answer engine it barely exists.

Nothing blocked that should not be. Our robots.txt allows the AI crawlers. Blocking them feels like control; in practice it means the model learns about you from everyone except you.

A plain-text summary for machines. We added /llms.txt: what we do, who for, where, and the handful of URLs that matter. It is a convention, not a standard, and it costs nothing.

What to check on your site this week

• Ask three assistants "what does your company do and where is it based?" Screenshot the answers. If any of them is wrong, that is your priority list.
• Search your company name and read the first page as a stranger would. Old directory listings with a former address or a dead service line are exactly the noise a model averages in.
• Open your homepage source. If the first paragraph does not say what you do in a sentence a human would say out loud, rewrite it.
• Check for structured data. A LocalBusiness or Organization node with name, url, address, sameAs and a stable @id is a morning's work for a developer and it is the difference between being an entity and being a rumour.
• Give every page a real description tag. Answer engines quote them more than search engines ever did.

Does this replace SEO?

No. It is SEO, with a wider set of readers. The same work that makes Google confident about who you are — clean entity data, pages that answer questions directly, real facts with sources, a fast crawlable site — is what makes an answer engine cite you rather than guess. We fold it into every SEO engagement now, and it is the first thing we audit on a new client's site.

If you want to know what the machines currently say about your company before your next prospect does, ask us. We will send you the screenshots.