If you ask an AI assistant who the best experts are in a field, it feels like you’re getting something new: an answer that weighs sources, connects the dots, and reads between the lines.
I wanted to know whether that’s actually what happens. So I tested it.
I run a weekly measurement of how AI search tools answer discovery queries in my own field, generative engine optimization (GEO) and AI search visibility. In early August I asked four tools a set of “who should I follow, who should I hire” questions and traced every answer back to its sources.
The short version: the answers looked like synthesis, but the sourcing looked like Google. The AI tools drew heavily on the same content format that has dominated search results for a decade, the “best X” listicle, and passed its rankings along largely as-is. I admit I expected a bit more.
Here’s what I asked and what I found.
1. What I asked, and which tools
On 5 August 2026 I ran the same set of discovery queries across the four tools I track: Google’s AI Overviews, Gemini, ChatGPT and Claude. I use the chatbots’ free tiers, clean sessions and no personalization. The queries came from the tracking list I use in my AI visibility measurement method:
- Best GEO experts to follow
- Best generative engine optimization experts to follow
- Top LinkedIn voices on AI search and SEO
- Best people to follow for practical AI marketing experiments
- Who is working on AI search visibility
- SEO experts to follow in Finland
- AI marketing experts to follow in Finland
- Who to hire for GEO and AI search optimization
- Freelance SEO and website consultant in Finland
For each answer I log two things: the full answer as given, and every source the tool used.
One honest caveat before the results. This observation is now based on four weekly measurement rounds, so it’s a repeated pattern rather than a one-off. It’s still a small sample: free tiers, one field, one language pair. But the pattern doesn’t need a large sample to be visible, and it matches what larger studies have found.
2. What the answers looked like
I’m deliberately not repeating the names the tools gave me. The people on those lists haven’t done anything wrong, and who they are isn’t the point. The point is where the answers came from.
To check how common self-inclusion actually is, I went through 16 of the person listicles that appeared most often as sources in my measurement data. In 11 of the 16 (69%), the named author of the article appears on their own list. The share rises to 12 of 16 (75%) if you also count organizational self-inclusion, where a company’s content team is credited as the author and the company or its founder appears on the list. In four cases the author didn’t seem to be on the list, and one case depends on whether you treat a company and its writing team as the same actor.
A few patterns from the sample, without names: several authors place themselves at number one on their own list. Two practitioners declare themselves the best in their country on their own sites. One consultant includes himself in his own top ten. Others appear on “influencer” lists published on their own domains.
An important limitation: this is a checked sample of the most frequently cited person listicles in my data, not a full audit of all 92 candidate lists my pre-screening surfaced. But the pattern is common enough that a single listicle placement shouldn’t be read as outside recognition without checking who wrote the list and what commercial ties sit behind it.
This deserves a pause, because it’s less scandalous than it sounds. Writing a “best experts in X” roundup and including yourself is a well-established content marketing play. It has worked in Google for years precisely because it serves the searcher reasonably well: someone puts in the effort to compile a useful overview of a field, and takes the visibility as payment. It’s not cheating. It’s how the game has been played for a long time, and everyone doing it is responding rationally to how search engines reward content.
What surprised me was not that these lists exist. It was that the AI tools used them the same way Google does: as ready-made answers, passed along without much visible weighing, combining or second-guessing. The synthesis I expected, cross-referencing multiple sources, noting who compiled each list, flagging the incentives involved, mostly wasn’t there.
One exception worth mentioning: Claude was the only tool that repeatedly added a note on its own, pointing out that the lists it found were likely marketing material because the authors appeared on them. My checked sample says that reservation is well founded. It’s also the kind of reading between the lines I’d assumed all four tools would do. One out of four did, sometimes.
3. The observation: AI search reads the same sources as Google, it just hides them better
This matches what larger studies show. In December 2025, Ahrefs’ Glen Allsopp analyzed 750 “best software, product or agency” prompts in ChatGPT and categorized 26,283 source URLs behind the answers:
- 43.8% of all cited page types were “best X” blog lists, by far the most common content format in the sources.
- Freshness mattered more than authority: 79.1% of cited lists had been updated within the year, and 35% sat on low-authority domains.
So the content that has performed in Google for a decade is now performing in AI answers, too. In one sense that’s reassuring: the rules didn’t change overnight, and SEO fundamentals still carry over. The engines do differ in which sources they favor, as I saw when I tested whether YouTube helps AI search visibility, but the overall pattern holds. In another sense it changes something important for the reader.
When you see a listicle in Google, you see the context around it: the domain, the design, the author, the ads. Those cues help you calibrate how much to trust it. An AI answer strips that context away and presents the same information in a neutral, confident voice. The source is still a listicle. You just can’t see that anymore unless you go looking.
That’s my main takeaway from these four rounds: AI search, at least on free tiers today, isn’t a more critical reader than Google. It’s the same reading with the seams hidden. The thinking still has to happen on our side of the chat window.
If you want to gauge how much weight to put on an AI recommendation, a quick source check goes a long way. Things worth noticing when you open a cited page:
- Is the list published on an independent site, or on the blog of someone who appears on it?
- Is there any stated methodology: criteria, data, how the ranking was made?
- Is it one of several near-identical lists that reference each other?
- Is the “updated” date doing more work than the content?
None of these automatically make a source useless. They just tell you what kind of document you’re reading: an independent evaluation, or a piece of (often perfectly legitimate) marketing.
4. What this means if you’re choosing a partner based on an AI answer
People increasingly ask AI tools the commercial version of my test queries: “who should we hire for GEO and AI search optimization?” If the answer is largely assembled from marketing content, that’s worth knowing before you act on it.
Three practical habits I’d suggest:
- Ask for sources, then open them. “Who compiled this list, and why?” takes a minute and tells you most of what you need to know about how much weight the answer deserves.
- Treat the AI answer as a starting shortlist, not a recommendation. The model isn’t vouching for anyone. It’s summarizing whatever was most citable that week.
- Look for evidence of work. Whatever the list says, a credible partner can show projects, measurements and documented methods. That’s the part no content format can fake.
And if you’re on the visibility-seeking side, as I am, transparently, in my own field, the same finding points to a strategy: since AI tools reward fresh, structured, citable content, the durable move is to publish your own data, methods and results, and let those become the thing that gets cited. That’s the kind of source that holds up when someone does open it.
I’ll keep running this measurement weekly. Claude’s occasional disclaimer suggests the models can get better at reading their sources critically, and if they do, my data will show it. Until then: AI search speeds up the finding. The judging is still yours.
Read more:
How I Track AI Search Visibility Without Paid Tools: A Repeatable Method
Does YouTube Actually Help AI Search Visibility? I Checked Google, ChatGPT, Claude and Gemini.