Which kinds of websites AI assistants rely on as sources

AI assistants rely on a few kinds of websites. The common ones are vendors' own pages, comparison and roundup articles, community forums, app stores, video platforms and official documentation. Which kinds dominate depends on the category and on the assistant, so the useful question is not "which sites matter to AI" but "which sites matter to which assistant in my category".

What kinds of sites show up most?

Across product categories, citations tend to fall into a handful of groups.

• Vendors' own sites. Product pages, pricing pages and help centres. Assistants cite them when the answer is a fact about one product.
• Comparison and roundup pages. Articles that list tools in a category, often from publishers, agencies or other vendors. They are a natural source for "best" and "alternatives" questions.
• Community and forums. Threads where people describe what they use. These are cited for opinion and for practical detail.
• App stores. For products that live on a phone, store listings are an easy, structured page to cite.
• Video platforms. Tutorials and reviews.
• Official documentation and support pages. Help articles from platforms, device makers and standards bodies, cited for how-to questions.

Each group serves a different job in the answer, and an assistant may choose among them differently.

Does the mix change by category?

Strongly. In a category where products live in app stores, store pages are common. In a category with many rival vendors, comparison pages and vendors' own sites are. In a category built on a few platforms, the platforms' help pages appear.

The Edition 0 reports list the most-cited domains in each of three categories, overall and per assistant. You can see in the editions that the lists differ sharply from one category to another, which is the point: there is no single "AI-favoured website".

Do assistants differ in what they lean on?

Yes, and the differences can be large. In the same category, one assistant might draw heavily from community discussions while another draws mostly from vendors and publishers. The findings guide gives measured examples from Edition 0.

This is one reason assistants end up recommending different brands. A brand that is strong on the pages one assistant prefers may be thin on the pages another prefers. See why AI assistants disagree.

Why does the type of source matter?

Because the source type shapes the content of the recommendation.

• Vendor pages describe a product as its maker would. They are accurate on features and promotional on quality.
• Roundups compare, but may reflect who was easiest to find, who paid, or who the author knew.
• Forums reflect what vocal users say, which can be useful, but unrepresentative.
• Documentation is reliable on how things work and silent on which to buy.

A recommendation built from forum threads reads differently from one built from vendor pages. Knowing which was used helps you judge it. Our guide to reading an AI answer like an analyst takes this further.

How do you find out which sites matter in your category?

You can do a small version of what a lab does.

1. Write 10 to 20 realistic buying questions for your category, without brand names.
2. Ask several assistants, with search on, and more than once.
3. Record every cited address. Group them by domain.
4. Tally the domains, overall and per assistant.
5. Sort the domains into the groups above. What dominates?

Then ask what you can do with that. If community threads dominate, being accurately described in them matters. If roundups do, see how to get onto the roundup and review pages AI relies on. If your own help pages are cited, check they are accurate and crawlable.

What about your own website?

Your own site is one source among many, and often not the main one. Assistants cite vendors' pages for facts about the vendor, such as pricing or features, and turn to third parties for judgement. So make your own pages accurate, specific and easy to read, because they are the best source for facts about you. Then look outward: are the roundups, reviews and community threads that assistants use in your category describing you correctly, and are you present on them at all? An outdated line on a widely cited roundup can do more harm than a gap on your own site. Reviewing those pages regularly is a low-drama way to spend effort.

How should you group domains?

Group by function, not by name. A single domain can play several roles: a company blog may be a vendor page in one answer and a comparison article in another. Decide a rule, such as classing by the page's main purpose, write it down, and apply it to every domain the same way. Note the grey areas as you go. Two people classifying independently will not agree on every case, so check a sample against each other and report how often you differed. It is a small step that makes your tallies far easier to defend.

What are the cautions?

• A citation is not a recommendation. A heavily cited site may be an explainer, not a verdict.
• Counts depend on what the provider reports. Some assistants report few sources, so their lists look shorter without being less thorough.
• Domains hide pages. A site cited often might be cited for one page. Look at pages as well as domains.
• Lists move. What is cited in one edition may not be cited in the next. See how AI recommendations change over time.
• Do not game it. Seeding a forum with promotion is easy to spot and risks your reputation. Be present honestly.

For the broader picture of what gets cited, see the kinds of pages AI assistants cite most, and for the full method behind the domain tallies, the methodology. The Citewake checker gives a quick view of how assistants treat a particular site.