Ask ChatGPT about what you sell. Does it name you?

Try this before you keep reading. Open ChatGPT, Claude or Perplexity and type what your ideal customer would type. Not your name: their problem. "I need someone to design the brand for my restaurant in Medellín." "Which company runs emotional wellbeing programs for teams?"

Did you show up?

If you didn't, someone else did, and they're getting conversations you'll never hear about.

A client arrived because ChatGPT recommended her

Stephania López (opens in a new tab), a brand and packaging designer in Medellín with nine years working with hospitality brands, wrote to me recently. I built her site a while ago. A new client had found her by asking ChatGPT who could design the brand for her restaurant, and ChatGPT recommended Stephania.

Nobody paid for an ad and there was no referral involved. An AI read her site, understood what she does and for whom, and named her.

I saw the same thing with Serendipia (opens in a new tab), which sells emotional wellbeing programs to companies. In both cases, 15 to 20 days after launch, inquiries were already coming in from an AI.

Two cases are not a study. But they line up with what research on generative search has been showing, so it's worth understanding the mechanism before copying the recipe.

Your customer doesn't search anymore: they ask an AI

Google showed ten links and left the choice to the customer. ChatGPT answers. Sometimes with three names, sometimes with one, almost never with ten.

This already has an academic name. In a paper presented at KDD 2024 (opens in a new tab), researchers from Princeton and IIT Delhi, among others, called it GEO (generative engine optimization): optimizing for engines that search, synthesize several sources and answer, instead of listing links. Their starting point is uncomfortable for any business: whoever creates the content has little control over when and how it shows up in those answers.

So here's the question back to you: if your ideal customer asks an AI agent today for what you sell, does that agent have any reason to mention you?

What does ChatGPT need to recommend you?

It needs to answer three things about you without guessing. An agent isn't impressed by your intro animation or your color palette; it reads the text and the data on your site.

• Who you are. Name, city and trade, stated plainly instead of hinted at in a slogan.
• What you sell and to whom. "Branding and packaging for restaurants and hospitality brands" is useful to it. "We create experiences that transcend" is useful to nobody.
• Why you. Published work, years of experience, named cases. Something it can cite.

The same study measured what changes a source's visibility inside the answer. Citing sources, adding quotations and adding statistics raised it by up to 40%. Repeating keywords, the classic SEO trick, didn't help. What the AI can verify is what it uses.

If your site doesn't answer those three things in text a machine can read, the agent doesn't know you exist, however beautiful the design is.

How I build a site an AI can recommend

I've been building sites with Astro for a while and putting together a method for this. I won't give you the full recipe, but it rests on four things that are on Stephania's and Serendipia's sites, and each one has evidence you can check.

1. The content is in the HTML when the crawler arrives. In December 2024, Vercel and MERJ analyzed crawler traffic on Vercel's network and found that none of the major AI crawlers execute JavaScript (opens in a new tab): not OpenAI's, not Anthropic's, not those of Meta, ByteDance or Perplexity. The exceptions are Google (Gemini uses Googlebot's rendering) and Apple. If your text only appears after a script runs, the page is empty to them. That's why I build static pages.
2. AI crawlers are let in. OpenAI says so in its documentation (opens in a new tab): sites that block OAI-SearchBot are not shown in ChatGPT search answers, except at most as navigational links. Plenty of sites block it without knowing, because of a template, a security plugin or a setting that came switched on.
3. The site describes the business with structured data (Schema.org): what it is, which services it offers, where and for whom. Precision matters here. Google says no special schema is needed (opens in a new tab) to appear in its AI features. Microsoft, on the other hand, says in its guide to appearing in AI search answers (opens in a new tab) that schema helps search engines and AI systems understand content. I add it because it removes ambiguity, not because it's a shortcut.
4. The copy says what you do. It's the least technical part and the hardest: writing so that anyone, person or model, understands in one line what problem you solve.

What about llms.txt?

It's a 2024 proposal by Jeremy Howard (opens in a new tab) to put a Markdown summary meant for language models on your site, usually at the root. Google Search doesn't use it, and says so in its own guide (opens in a new tab). Google's John Mueller compared it to the old keywords meta tag (opens in a new tab): something the site owner claims about the site, which you'd have to check against the site itself before trusting it.

Server data points the same way. Ahrefs looked at 137,210 of its customers' domains in May 2026. Of the roughly 38,000 that had an llms.txt, 97% got not a single request (opens in a new tab). Among the AI tools that did fetch it, the top two were GPTBot and Claude Code, a coding agent; search bots barely show up.

My take: search engines ignore it, some agents read it. OpenAI's documentation, incidentally, publishes one (opens in a new tab). I publish one because it costs nothing and is generated from the site's own content, but I won't sell it to you as the reason an AI recommends you.

What I can't promise you

Nobody controls what ChatGPT answers. There is no button for ranking first in an AI, and anyone selling you one is selling smoke.

There's also a finding that's uncomfortable for anyone who sells websites, starting with me. A University of Toronto study published in 2026 (opens in a new tab) compared the sources cited by ChatGPT, Claude, Perplexity and Gemini with Google's. When someone is comparing options, the AI engines mostly cite what third parties say about a brand: between 59% and 86% of sources. Only when the person is ready to buy do they give more weight to the brand's own site. Your site is the foundation an AI can read and verify, but at the stage where you're being compared, what others publish about you carries more weight.

What you can do is stop being invisible: make sure that when an agent looks, it finds something clear it can understand and cite. Whether you get chosen after that depends on your work and on who talks about it.

Run the test again

Type the question again, the one your customer would ask, not the one you would. If the AI names your competitors and not you, it isn't because they're better: they made themselves findable.

If you want the next answer to include your name, write to me or look at the sites I've built.

Sources

1. Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K. and Deshpande, A. (2024). GEO: Generative Engine Optimization. Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD '24), 5-16. doi:10.1145/3637528.3671900 (opens in a new tab)
2. Chen, M., Wang, X., Chen, K. and Koudas, N. (2026). Navigating the Shift: A Comparative Analysis of Web Search and Generative AI Response Generation. Proceedings of the Workshops of the EDBT/ICDT 2026 Joint Conference, CEUR Workshop Proceedings, vol. 4192. ceur-ws.org (opens in a new tab)
3. Zecchini, G., Moore, A. A., Ubl, M. and Siddle, R. (December 17, 2024). The rise of the AI crawler. Vercel. vercel.com (opens in a new tab)
4. OpenAI. Overview of OpenAI Crawlers. Official developer documentation. developers.openai.com (opens in a new tab)
5. Google Search Central. AI features and your website. Official documentation, updated December 10, 2025. developers.google.com (opens in a new tab)
6. Google Search Central. Google's Guide to Optimizing for Generative AI Features on Google Search. Official documentation, updated July 10, 2026. developers.google.com (opens in a new tab)
7. Madhavan, K. (October 8, 2025). Optimizing Your Content for Inclusion in AI Search Answers. Microsoft Advertising Blog. about.ads.microsoft.com (opens in a new tab)
8. Montti, R. (April 17, 2025). Google Says LLMs.Txt Comparable To Keywords Meta Tag. Search Engine Journal. searchenginejournal.com (opens in a new tab)
9. Linehan, L., with Guan, X. (June 15, 2026). We Analyzed 137K Sites: 97% of llms.txt Files Never Get Read. Ahrefs. ahrefs.com (opens in a new tab)
10. Howard, J. (September 3, 2024; version 2 updated August 10, 2026). The /llms.txt file. llmstxt.org (opens in a new tab)