How to get your company recommended by ChatGPT (we tested ourselves first)

If you run marketing at a B2B tech company and have typed your own category into ChatGPT to see who comes up, this is for you. To get your company recommended by ChatGPT, you need a page that answers the buyer’s exact question with real proof near the top, and you need to be on the lists and directories the AI assistants read. Most of ChatGPT SEO is those two things.

I know this partly because we tested it on ourselves last week, and it wasn’t flattering. GEO (generative engine optimisation) is getting a company named in AI answers, and we test it on ourselves. When we asked ChatGPT and Claude the questions our own buyers ask, we came up in 10 of 72 answers. On the GEO questions themselves we came up in none.

Below is what we ran and what we found. If you want the basics first, our piece on the difference between GEO and SEO covers them.

What did we test?

We wrote 18 questions a buyer might put to an AI assistant when looking for a company like ours. They fell into four groups: Microsoft partner and channel marketing, B2B tech agencies in the UK, AI tools and agents for marketing teams, and GEO. We asked each one twice on ChatGPT and twice on Claude, both with web search switched on and set to London. Then we recorded whether we were named, where we came in the list, and which pages each answer cited.

The results, in short:

We also asked three questions about us by name. Both assistants got who we are right every time.

Where do ChatGPT and Claude get their recommendations?

They use different sources, and it was the most useful thing we learned.

ChatGPT goes to the provider’s own page. Of the 174 pages it cited, 91% were agencies’ and vendors’ own service pages and articles. It found a page whose title and first paragraph answered the question, then built a shortlist of four or five names. It cited a “best of” list once.

Claude reads other people’s lists. 61% of its 178 citations were “best agencies” and “top 10” articles, plus 21 directory and review pages. It then repeated the names it found. It even pointed out, in its answers, that a lot of these lists are written by agencies that rank themselves first. It used them anyway.

So you need your own page for ChatGPT, and a place on the lists for Claude.

When Claude doesn’t search, it’s different. It answered 6 of its 36 questions from memory, and from memory it doesn’t know us. In a pilot run, asked “What is Fifty Five and Five?” without searching, it said it couldn’t identify us. So we only get named if a live search finds the right page.

What got us named, and what didn’t?

Almost every time we were named on a buyer question, it traced back to a specific page with specific numbers, or to a directory list.

The clearest win was our case study on cleaning SAP’s partner data . Asked for an agency to clean and enrich partner data, Claude put us first in both runs and quoted the figures: around a thousand partner accounts, 93% verified and matched, two weeks from raw data to sales-ready. It’s the only case study of ours that got us named on its own, and it has the firmest numbers at the top.

Our Microsoft MDF guide got us named on partner marketing questions too. Across the six Microsoft partner questions we were the most-named company, in 6 of 24 answers.

Generic positioning got us nowhere. The questions that describe most exactly what we sell, like who can build a partner benchmarking tool or how to prove the return on partner marketing funds, got us named in 0 of 16 answers. We’d done the work but never written a page that answered the question in the buyer’s words. That’s partly why we published how to see what MDF funds produce this week.

We also found old facts. Directory profiles we hadn’t touched in years still listed services we stopped selling, and the assistants repeated them. And Claude spotted that an old “best AI agencies” guide of ours ranked us first, and said so. Fair enough.

How do you get your company recommended by ChatGPT?

Six things, in the order I’d do them.

1. Write the page for the exact question. Use the buyer’s words in the title and answer them in the first two sentences. Owen’s guide to content that AI search engines cite goes into how.
2. Put the proof on the first screen. A named client and a real number. When ours had that, the assistants quoted it.
3. Get onto the lists the assistants read. Look at which “best of” articles and directories they cite for your category, then ask to be included, with evidence.
4. Fix your old facts. Update your directory profiles and anything else that describes what you used to do. The assistants can’t tell it’s out of date.
5. Don’t rank yourself first. A self-published “best agencies” list is easy for an assistant to spot, and it costs you trust.
6. Check it regularly. Put your buyers’ questions to the assistants every quarter, several times each, and read what they actually say about you.

Most of this is normal SEO. The difference is that an assistant gives one answer with a handful of names in it, so being on the page it reads, in words it can quote, matters more than your Google position.

The paper that named GEO found that adding sources, quotations and statistics to a page could raise its visibility in AI answers by up to 40% (Aggarwal and others, KDD 2024 ). Our results agreed: the pages with numbers got quoted.

What does generative engine optimisation mean for a B2B brand?

Generative engine optimisation, or GEO, is the work of getting your company named and described accurately when buyers ask AI assistants like ChatGPT and Claude for recommendations. For a B2B brand that means your own pages and other people’s. On your site, you need a page for each question a buyer asks, which answers it in plain words and backs it up with named clients and real numbers. Elsewhere, you need to be on the “best of” lists, directories and review sites the assistants read and repeat. ChatGPT leans more on the first and Claude more on the second, so a B2B brand needs both. GEO sits alongside SEO, since the pages still have to be found, but a good page looks different when an assistant is going to quote a sentence from it. And it has to be checked by asking the assistants directly, since there’s no ranking report to read.

What are we changing?

We’ve tidied our llms.txt file and the structured data on our site, which helps assistants read who we are. I wouldn’t expect much from that alone. Our llms.txt explainer notes that no AI provider officially uses the file for ranking yet.

The main job is writing pages for the questions we lost, in the words buyers use, with the proof on the first screen. The MDF piece was the first. We’ll run the same 18 questions again next quarter and see what’s moved.

So how do you get your company recommended by ChatGPT?

Answer the buyer’s question on your own page, with a named client and real numbers near the top, because that’s what ChatGPT cites. Get onto the lists and directories Claude reads. Keep what they say about you current, and check it by asking the assistants yourself.

Almost every mention we got came from a page with real numbers on it. If you’d like to see where you stand, I’m happy to run the same test for you.

Frequently asked questions