How ChatGPT-style recommendations actually use your website

When someone asks a chatbot for a local electrician, a solicitor, or a restaurant, the answer is assembled from whatever sources the system trusts that day. That might include the open web, licensed data, or prior training. You cannot inspect the whole stack from outside. What you can inspect is your own public website — the same material Repkit reads before you spend on campaigns or rewrites.

Recommendations feel personal because the question was personal. Underneath, many answers still lean on pages that clearly state services, areas served, credentials, and how to enquire. If your homepage is a slogan and a phone number with no scope, an assistant may hesitate or fill gaps from older mentions elsewhere. Repkit lists those gaps as findings so you can fix the public record instead of guessing what a model "remembers." The same applies when your trade spans several towns but the page never names them in prose.

ChatGPT and similar products change prompts, browsing modes, and partner sources without announcing them to your marketing team. That is why Repkit avoids selling a "ChatGPT ranking." The honest product is a reproducible crawl of robots.txt, sitemap, llms.txt, and visible copy. Run the check twice after you publish changes; the diff is yours to see.

Structured data helps when it matches visible text. Repkit notes schema types it finds, but it still expects a human-readable paragraph for visitors. Keyword stuffing works poorly for people and ages badly for machines. Write for a careful customer first; use headings that name the trade and place you serve.

Third-party listings matter, yet your site is the anchor you control. Hours, accreditations, and emergency call-out rules should live where you can update them tonight. Repkit does not scrape every directory; it tells you whether your anchor page makes sense on its own. From there, align Google Business Profile, trade bodies, and review sites so dates and services match.

Do not publish claims you cannot stand behind — "award-winning" without an award, "24/7" when the phone forwards to voicemail overnight. Assistants surface contradictions when the web contains them. Repkit flags thin or conflicting public copy so you can harmonise it.

Privacy and robots rules are part of the story. Blocking crawlers on staging is wise; blocking them on production by mistake is not. Repkit reads robots.txt and reports allow/disallow patterns in plain language. If you intend parts of the site to stay private, say so deliberately rather than inheriting a template rule.

The paid £29 PDF plan turns findings into an ordered checklist you can implement yourself. It does not promise a named slot inside ChatGPT; it improves the public evidence assistants inherit from the open web.

Local questions in assistants often mention "near me" even when the user does not type those words. Your page should still name the places you genuinely serve, in prose, so a summary cannot invent a radius you never agreed to. If you refuse certain job types, say that too. When you add a service or change a phone number, update homepage, footer, and contact together, then re-run the free Repkit check as a regression test.

Start with the free URL check, then read this guide alongside your findings. When the public page reads clearly to a stranger, you have done the part of "AI recommendations" you actually control — the rest is upstream of any single marketing tool.