State of AI Product Visibility 2026

What we found

AI assistants build their shopping answers from reviewers, not marketplaces. Of 15,437 citations, 3.5% went to Amazon, Best Buy, Walmart and every other marketplace or big retailer combined. Independent review sites, YouTube and forums such as Reddit took 64%, about 18 times as much. (Manufacturers’ own sites took 7%; the last quarter went to sites we couldn’t put in one group, such as specialist shops.)

Amazon.com was cited 44 times, 62nd of 1,179 sites. Best Buy was the most-cited retailer, 16th. Nearly all of Amazon’s citations came from one assistant, Claude.

Each assistant has its own habits. Perplexity cited YouTube 1,472 times, about one citation in five, and never cited Amazon.com. ChatGPT and Claude never cited YouTube. So “AI” isn’t one shelf: a product can be easy to find in one assistant and missing from another.

Each category has its own gatekeepers. In every one of the ten categories, at least two of the three most-cited sources were independent reviewers or YouTube. The only brand’s own site in a top three was usmile, an electric toothbrush maker.

What it means for a UK brand on Amazon

• Your listing is only part of what an AI assistant reads. Get your product in front of the reviewers and video creators who cover your category, and make sure their facts match your listing.
• Find your category’s gatekeepers. Ask the assistants your shoppers use for the best product in your category, and note the sites they cite. Those are the reviews that shape the answer.
• Keep your facts the same everywhere. Pack size, materials and claims should match across your listing, your website and the reviews. When they disagree, an assistant has to pick one. Our free Profit Leak Finder checks your Amazon listing for facts that disagree.
• Amazon’s own assistant, Rufus, wasn’t part of this study (see below). Ask it about your category and see what it says.

How we did it

• What we asked. 99 products in 10 consumer categories. For each product, 12 shopping questions to each assistant: 11 about the category (for example, the best, the best on a budget, the most reliable, the best for a first-time buyer) and 1 about the product itself. That is 3,564 answers in all, each asked once.
• The assistants. ChatGPT, Claude and Perplexity, each with web search on. ChatGPT cited fewer sources than the other two.
• When. 4 June 2026, in one pass. AI answers change week to week, so read this as a snapshot.
• How we counted. One citation is one link in one answer. Links are grouped by website, so amazon.com is counted separately from amazon.co.uk. We removed wholesale marketplaces such as eBay, Etsy and AliExpress, and film and TV sites, from every count. We didn’t check whether each link still worked.
• Where the sources are from. Most of the cited sites are American (Best Buy, Walmart, Consumer Reports), and the study wasn’t set to the UK. Read it as how the assistants answer in general, not as a UK-only test.
• What we left out. Google’s Gemini isn’t in this study. We also tested a simulated version of Amazon’s shopping assistant, Rufus: Claude answering as if it were Rufus, because Rufus has no public way in. Those answers are not real Rufus answers, so none of them are in any figure here. An earlier version of these figures, from June, included 160 citations from that simulation; we have taken them out.

All figures were recomputed from the raw answers on 19 October 2026. If you’d like the data behind a figure, email info@mrprime.com.

Published 19 October 2026. Data collected 4 June 2026.