Why ChatGPT Doesn't Mention Your Website, and How to Fix It

Why doesn't ChatGPT mention your website?

ChatGPT skips a website when it cannot reach it, cannot pull a clear answer from it, or cannot tell who is behind it. Google rankings do not carry over on their own. Most AI tools use their own crawlers and their own way of picking sources, so a page can rank well and still never be named.

That gap costs you in a quiet way. People now ask an assistant before they open a search page. If the answer names three of your competitors, the buyer never reaches the results where you rank.

Here is what usually sits behind the silence:

• The door is shut. A crawler is a program that visits pages to read them. AI search tools send their own, such as OAI-SearchBot for ChatGPT and PerplexityBot for Perplexity. A single line in robots.txt, or a firewall rule at your host, can turn them away.
• The answer is buried. AI tools quote short, clear passages. A page that starts with a long story and gives the answer in paragraph nine is hard to quote.
• Nobody knows who you are. An entity is a thing the AI can name: a person, a brand, a business. If your name, what you do and where you work are vague or differ across the web, the AI has nothing solid to point to.
• You never checked. Most site owners have never asked an assistant about their own topic and written down who it names.

On 2026-09-25 I checked the query "yılbaşı ışık süsleme fiyatları 2026" (Christmas light decoration prices 2026) for a lighting-decoration site I worked on. Retail price sites led the results, the AI summary took its answer from a competitor's B2B blog, and the site was not there at all. The same day, a query about municipal street lighting costs pulled four of its URLs and quoted its budget post. Same site, same day, opposite results.

The entity problem showed up in the same check. An old, closed directory listing for the same business made Google's summary describe it as an event company in the wrong district. No code change on the site could fix that.

How to check your AI search visibility by hand, step by step

A manual AI visibility check is five steps: test what the assistants say, check crawler access, check your key pages, check your entity, then rank the fixes. You can do all of it today with free tools. Do it by hand once, even if you later automate it, so you know what a good result looks like.

1. Ask the assistants first. Write down five questions a buyer would ask in your field, such as "best accounting software for freelancers". Ask each one in ChatGPT, Perplexity and Google with AI Overviews showing. Note who gets named and which pages get linked. This is your baseline, the "before" picture.
2. Check crawler access. Open yoursite.com/robots.txt. This is a plain text file that tells bots where they may go. Look for lines that block OAI-SearchBot, PerplexityBot, ClaudeBot or Claude-SearchBot. OpenAI's crawler page explains that OAI-SearchBot finds pages for ChatGPT search, while GPTBot collects training data, and that the two settings work separately. Perplexity's bot page says the same for PerplexityBot. Our free AI crawler checker reads the file for you.
3. Check the layer above robots.txt. Some hosts and CDNs block AI bots by default with a bot filter. A CDN is a network that serves your site from many places to make it faster. If robots.txt looks fine but bots still fail, check your host's bot or firewall settings.
4. Check that Google can index you. Google says its AI Overviews only use pages that are indexed and able to show a snippet. It also says there are no extra technical requirements and no special files or markup needed. So for Google, normal SEO health comes first.
5. Rewrite your key pages answer-first. Pick the three pages that should answer the questions from step 1. Put a two or three sentence answer right under each main heading. Add one real number or date per section. Short, plain passages are easier for an AI to lift and quote.
6. Make your entity clear. Use the same business name, description and links everywhere: your About page, your social profiles, your directory listings. Add Organization or Person schema, a hidden label that tells machines who you are. Link to your real profiles with the sameAs property.
7. Decide on llms.txt honestly. llms.txt is a proposed file that gives AI tools a short map of your site. No major AI engine has confirmed that it uses it to pick sources. It is cheap to add, so add it if you like, but do it last. I wrote up what my own server logs show in Does llms.txt actually work?
8. Rank the fixes and repeat. Put every problem in one list. Fix blocked crawlers first, because nothing else matters until bots can read you. Then fix pages, then entity. Ask your five questions again a few weeks later and compare with step 1.

The 5-minute AI crawler audit goes deeper on step 2, including which bots train models and which ones cite you.

The faster way: run a GEO audit from Claude Code

The faster way is to let Claude Code run these checks with skills that already hold the checklist. GEO means generative engine optimization: making a site easy for AI answer engines to read and quote. A skill is a saved set of instructions Claude Code loads when a task matches it. AI Search Visibility Toolkit is 11 of those skills, built for the AI side of search.

How to use it, step by step

1. Buy and download. Polar or Gumroad sends the download right after payment.
2. Install the skills. Copy the contents of the skills folder into ~/.claude/skills, the folder Claude Code reads skills from, and the agent files into ~/.claude/agents. There is no setup script. If a skill never fires, check where it lives: one of mine once sat for weeks in a folder Claude Code does not load.
3. Open Claude Code in a terminal, inside any folder.
4. Run the audit. Type /geo-audit followed by your site address, for example /geo-audit https://yoursite.com.
5. Read the score. You get a GEO score out of 100 with a breakdown by platform and by area, then a ranked fix list.
6. Fix and run again. Use the llms.txt and schema skills to draft fixes, ship them, then run the audit again to compare.
7. Write the client report. If the site belongs to a client, run /geo-report for a Markdown report you can export to PDF. A /geo-report-pdf skill is also in the folder, but the PDF script it calls is not in the current download, so it will not run yet.

What the output looks like

The audit returns one overall GEO score, then scores for citability, technical setup, schema and brand presence, and a readiness breakdown for ChatGPT, Perplexity, Google AI Overviews, Gemini and Bing Copilot. Those platform scores are estimated from your pages: structure, schema, crawler access and entity signals. The skills do not send questions to the assistants themselves. The last section is the fix list, ordered so blocked access always comes first.

Here is a real summary from a /geo-audit run on that lighting-decoration site on 2026-09-25, translated from Turkish. It ran on the GEO skills installed on my own machine, which share their names and five-agent layout with this toolkit but are not the same files.

Overall GEO score: 56/100. AI citability 66, brand authority 28, content E-E-A-T 49, technical GEO 86, schema 61, platform 54. The technical base is strong: full SSR, every AI bot gets a 200, Content-Signal and markdown negotiation work. Content and entity signals pull the score down. JSON-LD is printed via next/script in 31 files, so Service and LocalBusiness schema never reach the raw HTML.

A score is not a promise. It tells you how ready a site is to be read and quoted. Whether an assistant actually names you also depends on your competitors and on what the rest of the web says about you.

How much time it saves

That /geo-audit run on the lighting-decoration site took 16 minutes 56 seconds on 2026-09-25, with five agents working in parallel. I never timed the manual version of these checks, so I will not claim a ratio.

The real saving shows up on the second audit. You run the same command a month later and get a report in the same shape, so you can see what moved.