AI Assistants Now Write the B2B Shortlist — and 60% of the Lists They Trust Are Competitors' Own Posts

Researchers who asked ChatGPT, Gemini, Perplexity and Google's AI Overviews 120 B2B buying questions found that 60% of the "best of" articles the tools cite — 71 out of 119 — were published by companies selling in the very category they were ranking. In other words, when a procurement lead asks an AI assistant to name the best IT services firms, the "independent" list shaping the answer may well be a competitor's blog post that ranks itself first.

The study, run over three rounds in October 2026 and published alongside the researchers' full AI search dataset, logged roughly 1,050 sources behind 120 answers. The questions covered six B2B categories, including IT services, cybersecurity software and payments. Each question fell into one of five types: broad "best X" queries, industry-specific queries, very specific queries, "X vs Y" comparisons and name checks such as "Is company X any good?" The stakes for IT services firms, MSPs and resellers are straightforward: the shortlist is now drafted by an AI assistant before a buyer ever visits a website or takes a call, and if your firm isn't on the list, you may never make the longlist either.

What decides who gets named in AI answers?

The study's most important finding is that the source mix flips completely depending on how specific the buyer's question is. When someone asks a broad question — "best IT services partner for healthcare" — the AI tools lean on third-party articles. In the latest round, 53% of the sources behind broad "best X" answers were third-party articles: "top 10" lists, comparison pages and roundups. Your own website still matters, but for these questions it is rarely the deciding source. Being named in the articles the AI tools read is what puts you in the answer.

The picture inverts for specific questions. For queries like "Which MSPs offer 24/7 SOC monitoring for mid-market manufacturers?" or "How long does a typical Azure migration take with a managed services partner?", 74% of the sources were the providers' own pages. That's the mechanism worth understanding: the tools want to cite a page that clearly answers the exact question, and most provider sites don't have one.

For name-check questions, the tools leaned on review platforms and directories — about a quarter of the sources in the latest round were review and directory sites. So three different question types pull from three different source pools, and a firm needs presence in all of them.

The takeaway: broad questions are answered by other people's lists, specific questions by your own pages, and name checks by your reviews.

Can you move the needle on your own site?

Yes, and quickly. The researchers tested this on their own firm. After publishing a handful of pages answering specific buyer questions, the AI tools went from recommending them in 3 of 16 answers to 8 of 16 within a week. That is a doubling-plus of share of voice in seven days, from a small amount of publishing work.

There's a catch, though, and it explains why most provider sites get ignored. The AI tools only cite a provider's page when that page clearly answers the exact question. A generic "Our Services" page listing fifteen capabilities rarely gets used. A page that states plainly who you serve, what you do for them, how long it takes and what it costs to start often does.

The rule the study implies is blunt: specificity beats comprehensiveness. For each core service and industry you sell to, the researchers recommend a page that answers the questions buyers ask before they talk to sales — who it's for, what's included, typical timelines, how pricing works and what results look like. And put the answer in the first paragraph, because the tool is looking for a direct match, not an essay.

The takeaway: pages that answer one exact question get cited; pages that list everything don't.

Who writes the lists AI tools trust?

The third finding was, by the researchers' own account, the most surprising. In the latest round, Perplexity and Google cited 119 "best X", "top X" and "X vs Y" articles. Of those, 71 — 60% — were published by a company selling in that same category. The "independent" list an AI tool uses to recommend IT partners may well be a competitor ranking itself first.

Many such lists are useful, the researchers note, but the consequence is that the source material behind AI recommendations is shaped by whoever bothered to publish it. Firms that publish nothing get judged by what others say about them, or are left out entirely. That's the quiet mechanism of this whole shift: AI answers aren't a neutral market survey, they're a reflection of a visible, contestable body of source material — and competitors are already contesting it.

The takeaway: if you're not shaping the inputs, a competitor probably is.

What should IT services firms actually do?

The study lays out five concrete actions. None require new technology; all require consistent work.

• Check what AI tools say about you today. Ask ChatGPT, Gemini, Perplexity and Google the five or six questions your best customers would ask, including one that names you directly. Note who gets recommended and which sources are cited. Repeat monthly — the answers move week to week.
• Earn your place on the lists that get cited. Look at the list and comparison articles the AI tools cite for your category. Many are maintained by industry publications, directories and analysts, and many welcome a well-evidenced suggestion. Being on two or three of the right lists can matter more than a dozen new blog posts.
• Publish pages that answer specific questions. Build the per-service, per-industry pages described above, with the answer in the first paragraph.
• Make your reputation easy to verify. With roughly a quarter of name-check sources being review and directory sites, current reviews and accurate directory profiles are part of how you're judged.
• Get your experts quoted by publications. Coverage on recognised publications is one of the third-party sources AI tools draw on, and it lends weight your own site can't. A practice leader quoted on a topic your buyers care about is evidence an AI tool, and a procurement team, can check.

The takeaway: each of the three question types has a matching input you can work on deliberately — cited lists, specific pages, and verifiable reputation.

Why this matters now

The firms that end up on AI-drafted shortlists aren't necessarily the biggest, the study concludes. They're the ones easiest for AI tools to find, verify and recommend. That reframes "being found" for the entire channel: it's no longer just search rankings and referrals, it's citation inside generated answers.

The good news, as the researchers put it, is that the inputs are visible. You can see what the tools read, and you can work on each input deliberately — a claim few marketing channels of the past twenty years could support.