How to get cited by ChatGPT: a five-step technical checklist for businesses (2026)
ChatGPT cites pages it can read: server-rendered HTML, open AI-crawler rules, structured data, llms.txt, answer-first content. Five checks that take ten minutes.
You get cited by ChatGPT when the model can read your site and understand it: the content has to be in server-rendered HTML, robots.txt has to let the AI crawlers in explicitly, structured data has to say who you are, an llms.txt file should summarise the business, and the content has to answer the questions people actually ask. None of that is magic — it is five checks you can run yourself, described below.
First, though, why it matters and why a high Google ranking does not mean ChatGPT knows you.
How ChatGPT chooses whom to mention
When a user asks ChatGPT "who does X in Austria", the model runs a web search, reads a few pages and composes one answer with one or two mentions — there is no list of ten links to choose from. If the model does not receive your content, or cannot make sense of it, you are not in that answer.
Two things we learned by running these prompts ourselves through the API in 2026:
- Directories and comparison articles are the sources. When we asked which agencies offer generative engine optimization in Slovenia, four of the five names ChatGPT gave came from a single third-party comparison page, not from the agencies' own sites. Getting into the pages that answer "which company does X" matters as much as your own page.
- Google Business Profile counts. For a Slovenian service prompt in August, the answer was assembled from the companies' websites and their Google Business Profiles. A well-kept profile is a source AI draws on, not just a map pin.
The scale is not small. OpenAI reports more than a billion weekly ChatGPT users, and Google says its AI Overviews reach over 1.5 billion people a month. When we checked seven Slovenian queries in our field in August 2026, an AI Overview appeared on all seven.
The five checks that decide
| Step | What the AI checks | How to check it yourself |
|---|---|---|
| 1. Server-rendered content | Is the text in the HTML without running JavaScript? | View source (Ctrl + U) — is the text there? |
| 2. AI-crawler rules | Does robots.txt allow GPTBot, ClaudeBot, PerplexityBot? | yourdomain.com/robots.txt |
| 3. Structured data | Who you are, where you operate, what you sell (JSON-LD) | Google's rich results test |
| 4. llms.txt | A summary of the business for language models | yourdomain.com/llms.txt |
| 5. Content that answers | Self-contained answers, tables, clear questions | Does any paragraph stand on its own as an answer? |
1. The content has to be in server-rendered HTML. AI crawlers do not run JavaScript. If your page renders only in the browser — common with sites built on modern page builders — the crawler sees an empty shell. This is the most frequent and most fatal problem we find.
2. robots.txt has to let the AI crawlers in explicitly. The crawlers worth naming: GPTBot and OAI-SearchBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot, Google-Extended and Bingbot. Some WordPress security settings block them without the site owner ever knowing.
3. Structured data (JSON-LD) says who you are. Name, address, phone, services, service area — in a form a machine reads without guessing. Without it the model knows the page exists but not whose it is or whom it serves.
4. llms.txt is a summary for language models. A plain-text file with a description of the company, its services and key facts. What it contains and how to write one is covered in What is llms.txt and how to write one.
5. The content has to answer questions. A model quotes paragraphs that stand alone: a question in the heading, a direct answer in the first two sentences, concrete numbers. Content is also the single biggest growth lever we measure: for one client, a single article delivered 40 % of all Google impressions within three months of publication.
What it looked like for us
Two data points from our own work, so you can weigh the advice:
- 3DnaKlik.si, our own 3D-printing service, published a sourced comparison of European online 3D-printing bureaus on 17 August 2026, with answer-first sections, country pages, llms.txt and structured data. In the 28 days that followed, sessions from ChatGPT, Gemini and Perplexity went from 12 to 44, orders from Google organic search from 2 to 13, and ChatGPT with web search began listing the site fourth for "where can I order custom 3D printing online in Europe for a single small part", citing the comparison post as its source.
- Flisko itself. On 6 September 2026 we asked ChatGPT with web search which agencies offer generative engine optimization for businesses in Slovenia, Croatia or Austria. It named us first. The prompt and the full answer are published on our GEO page, so you can repeat the test.
How long it takes, and what can be promised
The honest answer: first movement is a matter of weeks, not days, and depends on how often AI crawlers visit your site. Nobody can guarantee a mention — not even the model providers. What can be guaranteed is a verifiable state: the model can read your site, understand it, and has a reason to cite you.
That is why every engagement starts with a measurement. The free AI visibility checker reviews your domain against twenty-one technical signals — including all five steps above — and emails you a report within minutes. No sign-up; you can check competitors too.
If the report shows something needs fixing, the AI search setup costs 199 € — a final price, files within three working days.
Related reading:
- What is llms.txt and how to write one — step 4 in detail, with a template.
- GEO vs SEO: what your business actually needs — where these steps sit in the bigger picture.
- ChatGPT ads in 2026 — the paid route into the same answer.
Frequently asked questions
How does ChatGPT decide which businesses to cite?
With web search on, ChatGPT runs a search, reads a handful of pages and composes one answer naming one or two sources. It can only cite what it can read: server-rendered text, structured data that says who you are, and paragraphs that answer the question directly. Comparison articles and business directories are frequent sources, because they answer 'which company does X' in one place.
Can I guarantee that ChatGPT will cite my site?
No, and neither can anyone else, including the model providers. What you can guarantee is a verifiable state: the page renders on the server, robots.txt allows the AI crawlers, JSON-LD describes the company, llms.txt summarises it, and the content answers real questions. Our own project went from absent to fourth place in a European buying prompt within three weeks of shipping exactly that.
Do I need llms.txt to be cited by ChatGPT?
It is one signal, not a requirement. Google does not use it for search, but AI crawlers and agents can read it, and it gives the model your facts — names, prices, service area — in a form that is hard to misread. Without it the model assembles your business from fragments, which is where invented prices and wrong services come from.
How long does it take to appear in ChatGPT answers?
Weeks, not days, and it depends on how often AI crawlers visit your site. In our own case a comparison post published on 17 August 2026 was being cited by ChatGPT by early September. Nothing about that timeline is guaranteed; it is one data point.