Showing up in ChatGPT: what you can control

Assistant makers describe access, not a spot in the answer

For AI Overviews and AI Mode, Google asks that a page be indexed and eligible to appear in Search with a snippet. It adds that there are no additional requirements, and that you don't need AI text files or special markup.

In its crawler documentation, OpenAI recommends allowing OAI-SearchBot so your site can appear in ChatGPT's search results. A site that refuses it isn't shown in those answers, except as a plain navigational link.

The others work the same way. In its page on its crawlers, Anthropic says that blocking Claude-SearchBot may reduce a site's visibility in Claude's search results. In its documentation, Perplexity recommends allowing PerplexityBot to appear in its results.

None of these pages promises a spot in an answer. They describe access conditions, and leave the choice of sources to their systems.

An allowed crawler can still receive an empty page

Access starts in the robots.txt file. AI companies separate the crawlers that collect for training from the ones that search to answer, and you can refuse the first without leaving the second's results. The guide on whether to block or allow AI crawlers covers both families, and the trap of a file that names crawlers.

A firewall or CDN can also turn these crawlers away, whatever robots.txt says.

Then there's what the server sends. Google runs a page's JavaScript to read it. Not every assistant maker documents this, and a page whose text only appears after a script runs may be read as empty. The guide on what Google reads first on a JavaScript site explains the two versions of a page.

Finally, Google's AI features rely on its index: a page marked noindex, or with snippets disallowed, can't be used there. An instruction left over from the staging copy of the site is enough to keep it out.

Pages that answer up front and say where their figures come from

A page is easy to understand when it says early what it's about, then details it under headings that announce what each section holds. Your offer, service area, prices and contact details are better written as text, not only inside an image or a PDF. A comparison table, or a question followed by its answer, can be quoted as is.

Structured data that still pays off describes the page in a format machines can read. Google finds it useful for rich results, not required for its AI features, and asks that it match the visible text.

An llms.txt file, which summarizes the site for language models, is optional: Google Search doesn't use it. The guide on the llms.txt file covers who reads it.

Before trusting a page, a reader looks for landmarks: who publishes the site and how to reach them, who wrote the page and when, where the figures come from. Google asks for them to be visible, and the guide on what Google means by E-E-A-T covers what a page should show.

Contradictions matter just as much. An unsourced figure, a testimonial credited to two clients, a fact that changes from one page to the next all take credibility away from the page that carries them.

Six requests for whoever runs your site

None of them involves writing code. Make them in this order, because the first condition governs the other two:

1. Ask whether the assistants' search crawlers (OAI-SearchBot, Claude-SearchBot, PerplexityBot) are allowed, both in robots.txt and at the firewall.
2. Ask for proof that your important pages are indexed by Google and Bing, and that none carries a noindex or no-snippet instruction left over from staging.
3. Require that the text of the pages that matter is in the server's response, without depending on a script.
4. Reread your main pages: the key point stated up front, headings that announce their content, your key information written as text.
5. Make it visible who you are and what you rely on: publisher, authors, dates, sources for your figures. Fix whatever contradicts itself from page to page.
6. When an offer promises you a spot in AI answers, ask what the promise rests on: no assistant maker guarantees one.

The first two requests are often a matter of a few settings. Getting a JavaScript site to send its text from the server is a project. The rereading and the sources are yours.

Opening with a thesis, a promise or a product grid

The audit reads your robots.txt using the group rule and records which AI crawlers are disallowed. That choice is described, not scored; a block on Bing's crawler is flagged separately, because Microsoft's Copilot assistant relies on that search engine.

On a few pages of each type, it requests the page under the names GPTBot, ClaudeBot and PerplexityBot, and compares the result with what a browser and Googlebot receive. A page served in full to the search engine and empty to these crawlers is reported. On the reference page, the server's response is also compared with the rendered page.

Every page analyzed is then read in full: its structure and headings, how it opens, where its figures come from, who publishes it, authors and dates, structured data checked against the visible text, and facts that change from one page to the next. The audit looks for an llms.txt file and only counts a real one.

The report on a B2B business website we publish as a sample finds access open: no conversational crawler blocked, page text served by the server. Reading the pages qualifies the rest. Recent articles open with their argument, presentation pages with a promise, and the site has neither an FAQ nor a single table. A figure of "more than 70% of B2B buyers" appears in two articles without a source, while another article cites McKinsey, Forrester, Gartner and the Harvard Business Review.

In the one on a sporting goods marketplace, category pages open on the product grid and buying guides on a promise. Yet the main categories carry price-range tables by condition, new, used or refurbished, which the report counts among the passages a machine can quote as they stand.

Whether your site appears in an assistant's answers can't be measured by an audit: it changes from one question to the next and from one day to the next. The audit checks the conditions that depend on you. This site applies all three: its robots.txt allows the assistants' crawlers, its pages are built ahead of time and send their text without scripts, and its llms.txt is generated from the same copy as its pages.