Which Sources Does AI Use in Its Recommendations?

The whole industry is selling tactics to show up in AI answers right now, and nearly all of them are presented as universal. Since we monitor our clients' AI visibility with Mencoro, I stopped guessing and looked at the data: which sources ChatGPT, Perplexity and AI Overviews actually cite across a sample of 39 projects in very different sectors, over a 30-day window. There is not a single tactic that works for all of them. What there is are five patterns, and several of them are uncomfortable.

A third of brands are not even a source about themselves

The brand's own site is the most cited source when AI talks about it in 13 of 39 projects. It is in the top three in half of them. And here is the figure that made me stop: in 12 of 39, almost a third, the brand does not appear even among its six most cited sources . AI builds what it says about them almost entirely on third parties they do not control.

You are not competing with your competitors, but with intermediaries

Across the 26 projects where the brand does not lead, only 7 have a direct competitor at number one. In the other 19 the one in charge is an intermediary: an official body (ENAC, the Spanish official gazette, usa.gov or the e-signature portal), the manufacturer's site, a giant retailer, a vertical directory, a media outlet, a social network, an aggregator or a services marketplace.

Put differently: in 19 of 26 cases, the work of positioning against your competitor does not touch the problem. In regulated sectors your real rival is the official body, and you do not beat it with content: you win by being the source that explains that framework better than anyone.

There is no such thing as "the five sites you need to work on"

The median number of distinct domains cited per project is 338, with cases above a thousand, and 17 of the 39 exceed 400 different sources. Any strategy promising to cover the citation map by working a handful of sites is ignoring the real scale of the problem.

Instagram and TikTok are no longer just networks: they are cited sources

They appear among the sources in 13 of 39 projects, and in three of them they are in the top 3. The most striking part: Instagram is the number one source in two pharmacy projects, with 86 and 110 citations, ahead of the client's own site and of official medical sources. In vocational training, TikTok racks up 122 citations. AI is treating reels and short videos as citable documentation.

Reddit only shows up where there is purchase comparison

With the whole industry recommending Reddit as the channel of the moment, in our data it appears in 6 of 39 projects (15%) and reaches the top 10 in only 4 (10%). Where it hits hard is adventure travel, where it is the project's third source with 48 citations, and electric cars with 43. It is entirely absent in health, pharmacy, banking, legal, industrial, B2B, education and local services.

My take on the sources AI cites

If I had to keep one idea from all this: the question "how do I show up in ChatGPT" is badly framed, because it assumes the work is about your site. In a third of cases your site is not even in the conversation. Before writing a line of content, what you have to do is look at who is being cited in your specific sector, because the answer changes radically from one to another and is almost never who you expect.

Methodology: a sample of 39 projects with monitored AI visibility, a 30-day window (12 July to 11 August 2026), across ChatGPT, Perplexity, AI Overviews and AI Mode. We analyzed the domains cited in the answers. In four projects we reviewed the top six sources rather than the top fifty, so "does not appear" there means "is not among the first six". No identifiable client data is included: projects are grouped by sector.