Getting your brand cited by ChatGPT and Google AI

By Joseph Tobin. 17 September 2026. 8 min read.

A growing share of buying research now happens as a conversation with an AI assistant. “Which running shoe for a wide foot?” “Best business bank account for a new limited company.” “Is this university any good for computer science?” The assistant answers with a shortlist and a few links. If the brand is not in the shortlist, the paid search click that used to follow never happens.

This is not SEO with a new name. The models choose sources differently from Google’s ranking. This guide covers what they look for and how to measure it.

How the models pick sources

The main assistants (ChatGPT, Gemini, Perplexity, Google AI Overviews and AI Mode) all now search the web before answering commercial questions. They fetch a handful of pages, read them, and write an answer with citations. The pages they pick tend to share four things.

1. They answer the question directly. A clear statement near the top, not three paragraphs of introduction.
2. They are structured. Headings that match the questions people ask. Lists and tables where the information is a list or a table.
3. They are specific. Prices, sizes, dates, numbers. Vague pages do not get cited because there is nothing to cite.
4. They are corroborated. The brand is mentioned on other sources the model trusts: review sites, trade press, comparison pages, Wikipedia.

The models mostly ignore marketing copy, pages behind JavaScript that does not render, and anything that reads as an advertisement.

Step one: measure where you stand

1. Write down the fifty to two hundred questions your customers ask before they buy. Use your search query data, your sales team and your customer service inbox.
2. Run each question through ChatGPT, Gemini, Perplexity and Google AI Mode. Record which brands are named and which pages are cited.
3. Work out your share of answers for each assistant, and your competitors’ share.
4. Repeat monthly. The answers change often.

This baseline is the number you will report on. Without it, you are guessing.

Step two: fix the pages that should be cited

For each question where you should appear and do not:

• Put the answer at the top. One or two sentences that answer the question as asked.
• Match the heading to the question. If people ask “how much does X cost”, the heading is “How much does X cost”.
• Add the specifics. The price, the specification, the dates, the comparison.
• Add schema. Product, FAQ, Organisation and Article schema help the models understand what the page is.
• Make it fetchable. Check the page renders without JavaScript. Check robots.txt does not block the AI crawlers you want.

Step three: get mentioned elsewhere

The models trust sources that other sources agree with. Get the brand onto the comparison pages, review sites and trade press that already get cited for your category. Find them from your baseline: they are the pages the models are already citing for your questions.

Step four: connect it to paid search

The questions that drive AI answers are the same questions that drive search. Use the AI baseline to find gaps in the paid search account, and use the paid search query data to find questions to add to the baseline. The brands that do well in both treat product data, pricing and content as one problem.

What to expect

Share of answers moves in weeks, not months, because the models re-fetch pages often. But it moves both ways. A competitor with a better page can take the citation back. Treat it as a monthly job.

Getting help

The AI search visibility service runs the baseline, the fix list and the monthly tracking.