How do I get my company recommended by ChatGPT?

Key takeaways

The short version of this page, in the order the sections below develop it. Each point is stated here in full so it can be read without the rest of the page, and the section that argues it gives the measurement and the source behind it.

• Because it never saw you.
• Not the ones most teams write for.
• Be crawlable, be indexed and eligible for a snippet, use semantic HTML, give the page a clear structure, and publish something that is not a commodity.
• Measure first, then find the sources, then get into them, then measure again with the same questions.

Why doesn't the model already mention you?

The model does not already mention you because it never saw you. An assistant assembles its answer from what other pages say about your category. Your own site is one input among many, and usually not the one that decides the shortlist. This is also why a company can rank first on Google and be absent from the answer. Ranking moves your page up a list of links. Being named requires other sources to have said your name in a context the model trusts.

Which sources actually carry these answers?

The sources that actually carry these answers are not the ones most teams write for. In the answers we recorded, the most cited domain was YouTube, ahead of Reddit. Vendor blogs barely featured. That has a blunt implication. If your plan for being cited is publishing more posts on your own blog, you are producing the source type these answers used least. The work is in the places that already carry the answer for your category: the directory your buyers check, the review page, the comparison article someone else wrote, the thread where people ask for recommendations.

What does Google say works?

Google says what works is to be crawlable, be indexed and eligible for a snippet, use semantic HTML, give the page a clear structure, and publish something that is not a commodity. Google publishes this directly. It is equally specific about what does not help: llms.txt and other special files, chunking your content into fragments, rewriting pages for machines, and overfocusing on structured data, which it says is not required.

It also names pursuing inauthentic mentions across the web as something to avoid. We agree, and it is the line worth being careful about: a real directory listing, a corrected fact and a genuine comparison entry are not the same as a bought mention, and only the first kind survives.

What is the order of work?

The order of work is to measure first, then find the sources, then get into them, then measure again with the same questions. Doing it in any other order means you cannot tell afterwards whether what you did worked, because you have no reading to compare against. The four steps below take weeks rather than days, and the last one is the one most programmes skip: without a second measurement using the identical question list, a change in the answers is indistinguishable from the models' own drift.

1. 01

   Find the questions your buyers ask. Unbranded, in their words. Your category, your competitors' positioning, the phrasing that turns up in your own sales calls.

2. 02

   Find the domains those answers cite. Every answer comes with sources. Rank them by how often each one carries the answer for your category.

3. 03

   Get into those sources, factually. Directory profiles, review-site pages, comparison articles. Where a source has you wrong, file the correction.

4. 04

   Re-measure with the same questions. Same models, fresh run, dates attached. What moved and what did not.

Questions people ask next

These are the follow-ups we are asked most often once the answer above has landed. The same rule applies to each: where a figure appears, its source and year sit in the same sentence, and where we do not know something we say so rather than estimate it.

Does publishing an llms.txt help?

Not in Google Search. Google's documentation states the file will neither harm nor help your visibility there because Search ignores it. It says nothing about ChatGPT or Perplexity, and neither will we. We audited seven mid-market sites: all seven had the file and none were being cited by the models we checked.

Can I pay for mentions?

Google explicitly lists pursuing inauthentic mentions as something not to do, and the assistants are getting better at discounting sources that look bought. Real listings and corrected facts are slower and they are the part that holds.

How long until it changes?

We do not know for your category, and anyone who gives you a number has not measured yours. What is knowable is what moved week over week, with the model IDs and dates attached, which is enough to decide whether to continue.

Related

Other questions in this library that cover the ground next to this one. Each is answered the same way — the direct answer first, then the measurement behind it — and each carries its own sources, so you can check any of them on its own page without taking this one on trust.

• How do you appear in Google AI Overviews?