AI visibility check: can ChatGPT read your website?

ChatGPT search (OAI-SearchBot, ChatGPT-User), Perplexity (PerplexityBot), Claude (Claude-SearchBot, Claude-User), Microsoft Copilot (Bingbot) and Google’s AI Overviews (Googlebot) fetch their sources with their own crawlers. If robots.txt blocks one of them, that service cannot cite your site.

Rating: We read your robots.txt according to the RFC 9309 standard and check, for each of these crawlers, whether the homepage is blocked. Fine: none blocked. Problem: at least one blocked. Blocks by a firewall (e.g. Cloudflare bot protection) are not visible to this check.

With “noindex” the page drops out of the index that AI search also draws on. “nosnippet” or “max-snippet:0” explicitly forbid Google from showing your text in results and AI Overviews.

Rating: We read the “robots” / “googlebot” meta tag and the “X-Robots-Tag” HTTP header of the homepage. Problem: noindex, nosnippet or max-snippet:0 set. Otherwise fine.

Many AI crawlers only load the HTML and do not run JavaScript. If your text is only built in the browser, they see an almost empty page.

Rating: We count the words in the delivered HTML (without scripts, styles, SVG). Fine: 250 words or more. Could be better: 80 to 249. Problem: under 80.

An XML sitemap shows search engines all important pages. The “lastmod” field says when a page changed — Microsoft Bing, whose index Copilot uses, explicitly recommends it so new content is picked up faster.

Rating: We look for the sitemap via the robots.txt entry, otherwise at /sitemap.xml and /sitemap_index.xml. Fine: sitemap with “lastmod”. Could be better: sitemap without “lastmod”. Problem: no sitemap found.

llms.txt is a proposed standard: a short Markdown file in the root directory that tells language models who you are and which pages matter. To be honest: none of the major providers has confirmed that they use it for answers. The effort is small, though, so the item only counts once.

Rating: We fetch /llms.txt. Fine: a text file with a title line (“# Name”) and at least one link, as described in the proposal. Could be better: file missing or not in that format.

GPTBot, ClaudeBot, Google-Extended, Applebot-Extended, CCBot and Meta-ExternalAgent collect text for training language models. Blocking them is a legitimate choice and does not stop AI search from citing you. When blocked, though, the model tends to know your company only from third-party sources.

Rating: Information only, not part of the score. We show which training crawlers your robots.txt blocks.

Structured data following schema.org (JSON-LD) states name, address, phone and logo in machine-readable form. Search engines use it to link your website to your company as an entity — the basis for an AI attributing you correctly.

Rating: We read all JSON-LD blocks on the homepage. Fine: Organization or LocalBusiness (or a subtype) with name and address or phone. Could be better: type present but without address and phone, or broken JSON. Problem: no company schema.

Name, phone number and address should be identical in the schema and in the visible text. Contradictions make it hard for machines to decide which detail is correct.

Rating: We compare the company name from the schema with the title, og:site_name and page text, the phone number with the tel: links, and the postcode with the page text. Fine: everything matches. Could be better: one mismatch. Problem: two or more. Without a company schema it cannot be checked (then it does not count).

AI answers often rely on third-party sources: Google Business Profile, local.ch, search.ch, commercial register (Zefix, Moneyhouse), LinkedIn. If your website points to these entries via “sameAs”, they can be attributed to your company unambiguously.

Rating: We count which known profiles and directories the homepage points to — machine-readable via “sameAs” in the schema and as visible links. Fine: at least two via “sameAs”. Could be better: only visible links or a single “sameAs”. Problem: no references. Whether you are actually listed correctly in those directories is something we only check in the personal review.

People ask AI assistants in full sentences. Pages that state typical customer questions verbatim as headings and answer them directly provide well-fitting passages. Note: since 2023 Google only shows FAQ snippets for government and health sites — for AI answers a clear question-and-answer structure is still useful.

Rating: Fine: FAQPage schema or at least three headings phrased as questions on the homepage. Could be better: fewer. There is no “problem” rating here — FAQs are not mandatory.

AI systems prefer short passages that make sense on their own. A first paragraph under the main heading that says in two or three sentences who you are, what you offer and where is the passage with the best chance.

Rating: We look for the first paragraph (<p>) after the H1 with at least eight words. Fine: 15 to 100 words. Could be better: shorter (usually a slogan) or longer. Problem: no paragraph in the HTML. Whether it actually answers the core question a machine can only judge roughly — we show you the paragraph so you can decide.

The title is the shortest summary of your page. AI search shows it next to source links and uses it to assign the page to a topic.

Rating: Fine: 20 to 65 characters. Could be better: shorter or longer. Problem: no <title>.

The description sums up the page in one sentence. Without it, search engines pick an excerpt themselves — not always the one you would have chosen.

Rating: Fine: 70 to 165 characters. Could be better: shorter or longer. Problem: no description.