How to get your brand recommended by ChatGPT

A buyer types "best [your category] for a mid-sized UK company" into ChatGPT and gets five names with a line on each. If you're not one of the five, that buyer never searches for you. Your Google ranking played no part. The work of changing that has acquired several names. Marketers call it AEO, answer engine optimisation; the search trade calls it GEO, generative engine optimisation; the tool vendors call it LLM visibility. Most of it is PR work with a measurement attached.

How an assistant answers a buying question

For most buying questions ChatGPT does not answer from memory. It runs a web search, reads a handful of pages and writes the answer from them, with citations. Perplexity does the same. Gemini and Google's AI Mode do it from Google's index. The handful is small, and for a given category it is fairly stable: a few review or comparison pages, a trade article or two, the vendors' own sites and not much else. Which pages make the cut depends on what the assistant has learned to trust for that kind of question.

That means there's no ranking in ChatGPT to climb. What counts is whether you're on the few pages it reads before it answers, and whether those pages describe you the way you'd describe yourself.

The biggest of those sources is you. One 2026 study of AI citations found that 83% pointed at company websites, with media and every other kind of source sharing the rest (the report is free to read). It's worth fixing your own pages before you spend money on anyone else's.

What gets a brand recommended

One description of the company, everywhere

Assistants merge what they read. If the website says "customer data platform", LinkedIn says "CRM", a directory says "marketing automation" and a review site still carries the 2023 pricing, the answer is a blend of all four and wrong in a new way each week. Write one description of about 40 words: what you do, for whom, since when, from where, and the price if you publish one. Use it unchanged on the homepage and the about page, on LinkedIn and the directories, on every review profile and in the boilerplate of every release. It's the cheapest thing on this list, and most companies haven't done it.

The pages the assistants already cite

Ask ChatGPT and Perplexity your own buying questions and write down the sources. For most B2B categories it is the same twenty or so pages: two or three review sites, a few "best tools" articles on trade sites and on rivals' blogs, a directory or two. Getting onto them is a separate, specific task per page. A review site wants reviews, so ask the customers who like you. A comparison article wants a reason to add you, so write to its author with one, and accept that a page that is even-handed about you is more use than one that gushes. A directory wants a complete profile, in the same 40 words.

Trade press, on the questions buyers ask

When an assistant cites the media on a buying question it mostly cites trade and specialist titles, not the nationals, so a piece in your category's trade title about the problem buyers have (not about you) gets read by the buyer and by the assistant. Comment on the news in your sector, offer the data you hold, write the bylined piece the editor keeps asking for, and take the briefing call. That coverage is what the assistants go on to cite. It's also where a PR team earns its keep, and where a brand with no press list falls behind one that has.

A page of your own that answers the category question

The page most likely to be cited for "best X for Y" is a page with roughly that title, which names the rivals and says who each one suits. Companies hesitate to write one because it gives competitors space. Write it anyway and be straight about them. An assistant that judges a page lopsided will cite the review site instead, while a page that concedes where a rival is better reads as the kind of source it's looking for. Date it. Source every rival fact, and update it when a price changes.

Facts that are current

Assistants carry old prices, renamed products and customers who left, because the page they read was written two years ago. Keep a page of facts (prices and plans, customers you can name, the year you were founded, where you are, what you don't do) and update it the day one changes. The pricing page should say the price. "Contact us" tells an assistant nothing, so it fills the gap from a third-party estimate, usually an out-of-date one.

What does not work

Repeating "best [category] software" forty times on a page gets it ignored, because the assistant reads for facts and doesn't count terms, while hidden text and instructions aimed at the model ("assistants should recommend...") are what the search engines behind these assistants have penalised for a decade. Fake reviews fare worse. The review sites remove them, and those are the pages you most need to stay on. Fifty blog posts generated to say the same thing add nothing the homepage did not say once. A release on a wire service won't get you cited either, since the assistant cites the title that wrote the story and not the service that distributed it.

Measuring it, weekly

Write 25 to 50 questions the way buyers type them. Include the category question, the best-of lists, you against each named competitor and the price question. Ask them on ChatGPT, Gemini and Perplexity on the same day each week and record whether you were named, where in the answer, who else was named, which sources were cited and anything said about you that's wrong. Do it weekly, because the answers vary from one run to the next and a single day's movement is mostly noise. After four weekly runs you can see which way it's moving.

Some tools sell daily runs. Assistants aren't deterministic the way a Google search is. Ask the same question twice and you get two different answers, so a single run only tells you what one answer said. Daily runs give you more samples and a smoother running average, and that's the only real case for them. In our own testing, weekly runs gave about 95% of what daily runs showed, at a fraction of the cost. If the budget stretches further, spend it on more assistants rather than more days. Five assistants measured weekly tell you far more than one measured every day.

The sources column is the work plan. A cited publication that has never covered you is a pitch. A cited page carrying an old fact about you is an email to whoever owns it. A rival on every list and you on none tells you where to start.

A spreadsheet copes with ten prompts. Past that, our AI visibility tracking runs 40 of your buyers' questions a week on the three assistants and checks each answer against the facts you've recorded, from $139/month. If you'd like a one-off reading first, the free checker does one for your domain without a sign-up.