Buyers increasingly start supplier research in ChatGPT, Claude, Perplexity or an agentic browser: “Who can translate a legal contract into German?” The assistant reads a handful of sources and writes one answer. If it cannot read a translation agency’s services cleanly, it guesses, or leaves the agency out of the answer altogether.
In July 2026 we audited the site the way a growing share of the web now experiences it: as software trying to extract a clean answer. The two disciplines came back very differently.
Before · July 2026 · ChatGPT audit
Same site, same day, out of 10
Overall LLM friendliness: 7.8/10. Traditional SEO had been solved for years. The second discipline had barely started.
Google’s Agentic Browsing audit told the same story. It passed 2 of 2 applicable checks, because without an llms.txt the third check was not even being run.
Together with the client, we set three goals. Discoverable: canonical company and service facts in the files AI agents look for first. Readable: services, FAQs, case studies, team and articles available as clean, structured data. Accurate: explicit rules for what an AI may cite, how it describes pricing and what it must never invent. And no rebuild: the human website stays exactly as it is.