How to appear in ChatGPT recommendations
Nobody can guarantee a mention in a language model's answer, tool vendors included. What you can do is remove the reasons a model cannot mention you even when it otherwise would. Below are those reasons, in the order in which it makes sense to remove them.
Access first, because without it nothing else matters
Open your-domain/robots.txt and look for six agents: GPTBot and OAI-SearchBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot, Google-Extended and CCBot. Many sites have them disallowed without anyone having chosen that: the rule arrived with a template, a plugin or advice from a few years ago, when blocking was the default answer.
A block is not a mistake in itself. It is a decision: if you do not want your content used to train models, blocking is the right choice and AI visibility is not your goal. The mistake is only the unnoticed block, the one nobody in the company chose.
Say what you do in the words your buyer uses
A model answers the buyer's question, not your slogan. The buyer asks about an industry, a place and a problem. If your site carries only a product name and a phrase about business transformation, the model has nothing to connect to the question.
- State who the product is for, with numbers: company size, industry, market.
- State what the product does not do. A model uses that when it excludes someone from a list, and being excluded from the wrong list is good news.
- Put prices, scope and terms in text, not only in an image or behind an enquiry form.
- Publish the questions you actually get, as questions and answers.
Structured data says the same thing in a form nothing has to decipher
Our measurement checks seven types: Organization, Corporation, LocalBusiness, ProfessionalService, Service, FAQPage and SoftwareApplication. The point is not to have as many as possible. The point is that the ones you publish describe reality: FAQPage only where the questions and answers are actually visible on the page, LocalBusiness only if you have an address a customer can walk into.
Publish a rating in structured data only if it exists somewhere verifiable outside your own site. A made-up rating is grounds for a manual search penalty and is easy to check.
The last and hardest part is outside your own site
Everything above is your own site and everything above you can do yourself in a few days. What remains is what you cannot write yourself: what is said about you elsewhere. Industry directories, comparison articles, customer posts, forum answers, association reports. A model does not rely on your site alone, because it knows the site is yours.
The measurement does not measure that part directly. It does measure its consequence: if your share of mentions rises without anything changing on your own site, the difference came from outside.
Frequently asked questions
Can I pay to be recommended by ChatGPT?
Not with us and not with the model vendor. Places in an answer are not sold the way search ad space is. Anyone promising you a position is promising something they do not control.
How quickly does a change show up in the answers?
It depends on the change. An unblocked crawler and new script-free content can show up at the next measurement, because the effect travels through the live search that models use while answering. Mentions outside your own site take longer. That is why the measurement runs periodically rather than once.
Do I need to publish llms.txt?
In our score it is worth six points out of one hundred. It is not a mandatory standard and it does not replace readable content. It is cheap though: one file that briefly states what the company does and where the main pages are. If it takes half an hour, do it; if it takes a week, you have more important things on the list.
Find out where you stand now
The free measurement shows whether any of the three models mentions you at all and which technical part is holding you back.
Check AI visibility