We asked ChatGPT who should build a website. Here's how to be the answer.
Your next customer won't scroll Google — they'll ask an AI. We reverse-engineered how generative engines decide whom to recommend, and how we make our clients the name that comes up.

01The search bar moved
For twenty years, visibility meant one thing: rank on Google. That era is ending quietly. A growing share of buyers now asks ChatGPT, Gemini or Perplexity — and acts on the answer. Not on a list of ten blue links. On the three brands the model names.
Here's the uncomfortable part: to those engines, an unstructured website is invisible. No schema, no clean semantics, no consistent facts across the web — no recommendation. Your site can be beautiful, fast and completely absent from the only conversation your next customer is having.
02We ran the test
We asked four AI assistants the questions your customers actually ask: "Who is a reliable web agency for mid-sized businesses?", "Which company can build my shop in German and Spanish?", "Best WordPress expert since 2018?" — dozens of phrasings, several sessions.
The pattern was brutal in its clarity. The engines recommended brands whose facts they could verify in structured, consistent sources. They quoted pages with clean headings, definition-style answers and schema markup. They skipped — silently — sites that were only pretty. Charm is not a data format.
03What the engines actually quote
Generative engines don't read your homepage like a human; they parse it like a database. Schema.org JSON-LD tells them what you are — organization, service, address, opening hours. llms.txt tells their crawlers what's safe to read and what to quote. Consistent mentions across the web teach them who you are — that's entity building.
And then there's the content itself: definitions, tables, FAQs, direct answers. Text written to be cited, not just scrolled. If a sentence answers a real question in one clean paragraph, an AI can lift it into a recommendation with attribution. That is the new backlink.
04The six signals we build into every site
One: Schema.org structured data in JSON-LD. Two: an llms.txt file in the root directory. Three: Open Graph and social preview images. Four: clean semantic HTML — one H1, real headings. Five: mentions in sources the models train on and cite. Six: fast, mobile-first, crawlable in under two seconds.
Take the test yourself: how many does your current site have? Most sites we audit score one or two — usually by accident. Every missing signal is a customer an AI will never mention you to, and that customer will never write you an email complaining about it. Silence, in GEO, is the sound of losing.
05What GEO takes and what it returns
Every website Mediaberg builds ships with SEO + GEO included — it's part of the AI-accelerated track and the old-school track alike. You'd have to pay extra to have it removed, and we'd argue with you about that invoice line.
Compare that to what one lost customer costs per year — then multiply by every year your site stays unreadable to machines. GEO is not a luxury layer. It's the difference between existing and being recommendable.
06Try it on your own site tonight
Three checks anyone can run: view your page source and search for "application/ld+json"; open yoursite.com/llms.txt; ask ChatGPT "Who is a good [your profession] in [your city]?" and read the answer carefully.
If two of those three checks fail, your site is mute in the room where buying decisions happen now. We fix that as a discipline, not a trick — and we've been doing the equivalent for search engines since 2006. The audit is free; the answer will be specific.
Reading is cheaper than guessing.
Every article above is a pattern we've lived through with real clients. If one of them sounds like your situation — the audit is free and the numbers are real.
Get my free auditKeep reading

Opinion
Page builders are built for the masses — your site is not the masses.
October 28, 2026 · 5 min read
Read article
Use case
Eleven refreshes, one workshop: how DentCar outlasts every design trend
October 14, 2025 · 6 min read
Read article