More and more shoppers skip Google and ask ChatGPT what to buy. They get one answer with a short list of brands, and if yours is not in it, you are not in the running. This year I made that problem a priority for SupplementHub, a premium supplement retailer in the Functional Nutrition Group that sells across Europe. Here is what I changed, what happened and what I would do first on any website.
The results
Comparing 1 January to 23 October 2026 with the same period in 2025:
Visits from AI assistants up 59%, including ChatGPT, Perplexity, Copilot and Gemini.
Orders from AI assistants up 49%.
ChatGPT is now one of SupplementHub's top four referral sources of revenue.
97% of AI-referred revenue comes from ChatGPT, which is why most of the effort went there.
To be clear about scale: AI assistants are still a small channel next to search, email and paid. But the people who click through have already been told you are a good fit, so they arrive ready to buy.
Why most websites get left out
AI tools do not browse a site the way a person does. They read the text, the structured data and what other sites say about you, then decide who to quote. Most websites make that hard:
Key facts sit in images, sliders or PDFs that AI tools cannot read.
Products have no structured data, so price, ingredients and stock are guesswork.
Nobody answers the specific questions buyers ask, so the AI quotes a competitor who does.
The business is described differently on the website, Google profile and directories, so the AI is not confident about any of it.
What I changed on SupplementHub
1. Answer-first content. We wrote blog articles around the questions shoppers actually type into ChatGPT, such as which supplement helps with a specific goal, or how two ingredients compare. Each article opens with a clear two-sentence answer, then the detail. AI tools quote answers, not introductions.
2. Product schema. Every product page got structured data for price, availability, ingredients and reviews. That gives AI tools facts they can repeat accurately instead of guessing or skipping the product.
3. llms.txt. This is a plain-language file at the root of the site that tells AI crawlers what the business does and which pages matter most. It takes an afternoon and costs nothing, and most sites still do not have one.
4. Measuring it properly. Most analytics setups lump AI visits into general referral traffic, so nobody can see whether the work pays. We tracked AI assistants as their own channel in attribution, so every visit and order is counted. Without that you are optimising blind.
Where I would start on your website
If you do nothing else this month, do these three things:
Ask the AI. Open ChatGPT, Perplexity and Gemini and ask the ten questions your customers ask before they buy. Note who gets recommended and what they say about you, if anything.
Fix the facts. Make sure your name, location, offer and proof are written the same way, in plain text, across your website and listings.
Answer one big question per page. Pick your most important buyer question and put a clear answer in the first two sentences of the page that targets it.
A note on promises
No one can guarantee ChatGPT will recommend you. AI answers change and every tool works differently. What you can do is remove the reasons they leave you out, measure where you appear before and after, and track the sales that come through. That is the job.
If you want to know where you stand today, I offer a free AI visibility check: I test the questions your buyers ask and send you a short summary of who gets recommended and why. You can also read more about my AI search optimisation (GEO) work.
