How to get cited in ChatGPT and AI answers, and what nobody can promise

The ones we found run a set of prompts through one or more AI systems and record which sources or brands appear in the answers. That is a count of outputs on a chosen sample. It is not a measure of cause.

Examples show the range. Search Engine Land reported data from BrightEdge, a search vendor, comparing how often Google's AI and ChatGPT cite retailers¹. A forum author on Shopify Community, tracking citations for a set of Shopify stores, counted shopping-prompt citations across several AI engines and found ChatGPT almost never cited Reddit². Profound, an AI visibility vendor, published its own breakdown of ChatGPT shopping and names Reddit as the most-cited domain³.

Three things to ask of any such study: who ran it, which prompts went in, and whether the sample matches your market. A vendor that sells tracking has an interest in the topic being worrying.

Look at Reddit: the forum count found ChatGPT almost never cited it for shopping prompts², while Profound's breakdown puts it at the top³. Both are counts of outputs. The gap has ordinary reasons.

• Different prompts. A study built around product comparisons will find different sources from one built around "best of" questions.
• Different systems. Google's AI features and ChatGPT are separate products. The BrightEdge data reported by Search Engine Land compares them precisely because they cite retailers differently¹. Results for one do not carry over to the other.
• Different counting. A citation, a mention and a link are three different events.
• Different dates. The systems change. A count from one month may not describe the next.

None of this makes the studies useless. It makes each of them one data point from one setup.

A digital marketing agency's blog summarises correlational work from vendors, including Ahrefs and AirOps⁴. Its summary of the Ahrefs data says brand mentions across the web correlate more strongly with AI visibility than backlinks do⁴. That is second-hand vendor data and correlation, not cause. A brand that is mentioned widely probably also has other things going for it: reviews, press, products people search for by name.

What that supports, with care:

• Be present where the topic is discussed. Reviews, comparison sites, forums and press are the places people name brands.
• Be clear about what you sell. Pages that state products, prices, delivery and returns plainly give a system something accurate to repeat.
• Be consistent. Use the same brand and product names everywhere.
• Earn mentions you would want anyway. A mention bought through a trick is fragile.

Our view: the useful reading is "be worth mentioning", not "buy mentions".

Google's documentation says there are no additional requirements for appearing in AI Overviews and AI Mode, and that no special schema markup or AI text files are needed⁵. A page needs to be indexed and eligible to show with a snippet⁵.

That covers llms.txt, a proposed file meant to point AI assistants to a site's content. Google's guide to generative AI search says Search ignores such files, so creating one will "neither harm nor help" your visibility in Google Search⁶. The same guide calls optimising for generative AI search "still SEO"⁶. Google's documentation covers Google Search only. It says nothing about ChatGPT, which has its own systems.

So the SEO basics still apply: crawlable, indexable, useful pages with accurate content. They do not guarantee a citation. They remove the avoidable reasons you would not get one.

Keep it simple and honest.

1. Write a short list of real questions. Use the ones customers ask, not keywords.
2. Run each question more than once. The agency summary cites AirOps data showing that brands often drop out between back-to-back answers to the same query⁴, so a single check proves little.
3. Record the date, the system and the wording. Without them the result cannot be compared next month.
4. Note mentions and citations separately. A brand named in text is not the same as a linked source.
5. Check Google separately. Our guide to the Search Console AI Overviews report shows what Google's own reporting gives you for free.

Paid tools automate steps 1 to 3. Their numbers describe their prompts, not the market.

Treat these as red flags.

• "Guaranteed placement in ChatGPT." Nobody outside the system controls its output.
• "A secret file or markup that unlocks AI citations." Google says none is needed for its features⁵, and that Google Search ignores llms.txt⁶.
• A single score with no prompt list. You cannot check what it measures.
• Hidden text aimed at AI tools. Content written to instruct a model is prompt injection: input that changes an AI model's behaviour in unintended ways. Do not put it on your site.
• Proof from one screenshot. A single run shows one run.

Our view: an honest provider tells you what they can and cannot influence, and shows the prompts behind any number.