What AI assistants cite when they name a vendor

Study·11 minutes·26 October 2026

What we collected

In 2026 we recorded the pages that AI assistants cited as sources in answers naming established vendors in two niches. In ITSM software those were Freshworks, SolarWinds, SysAid and InvGate. In proxy servers, Bright Data, IPRoyal and Webshare.

That produced 152 citations: 90 in ITSM and 62 in proxies. Each one is a pair of a vendor and a page that was cited when that vendor was named. We then classified every page by hand – what kind of page it is, who publishes it, and what it would take to appear on it – because the last question is the one a marketing team actually has to answer.

This is a study of sources, not of share of voice. It tells you which pages the assistants lean on when they name a vendor. It does not tell you how often each vendor is named; that needs a fixed prompt set re-run over time, which is what our AI search baseline does.

Rankings are most of what gets cited

In ITSM the pattern is almost uniform. Eighty-one of the ninety cited pages are rankings: “best ITSM tools”, “ServiceNow alternatives”, “best help desk software”. Eight are ordinary blog articles that mention a vendor in passing, and one is a review site.

Across both niches, not one of the 152 pages sits on the named vendor’s own website. The assistants named Freshworks or Bright Data on the strength of what other sites said about them.

Proxies are more mixed, but rankings are still the largest group by a wide margin: 32 of the 61 pages we could classify. The Webshare row in our data contained a single unclassified page, so the proxy percentages below are out of 61.

The practical reading is simple. When a buyer asks an assistant which tool to pick, the assistant is mostly summarising the comparison pages that already exist. If you are not on those pages, there is little for it to summarise about you.

A long tail, not a shortlist

The second surprise is how little the sources repeat. The 152 citations point to 141 different pages on 131 different domains. In ITSM, only seven pages were cited for more than one vendor. One page – a roundup of ITIL change management software on superblocks.com – was cited for all four. TechRadar, the most recognisable publication in the set, appears three times. In proxies, only two pages were shared between Bright Data and IPRoyal.

There is no single directory to get into and no one publication whose editor decides the answer. An assistant composes its answer from dozens of mid-sized pages, and each of them is a separate conversation with a separate owner.

Most sources are other companies’ blogs

We split the publishers into two groups: independent media and review publications on one side, and the websites of companies that sell something on the other. In ITSM the split is lopsided. Fourteen citations, sixteen percent, sit on ten media domains such as TechRadar, The Digital Project Manager, People Managing People and AIMultiple. The other 76 – eighty-four percent – are company blogs: help desk vendors, ITAM vendors, workflow tools and service providers writing “best of” lists for their own category.

In proxies, eighteen of the 61 classified citations, about thirty percent, are blogs of other proxy and web scraping companies listing their competitors. The rest is a mix of affiliate sites, review aggregators, general tech media and unrelated software blogs.

This matters because a company blog is a far more approachable counterpart than TechRadar. Vendors publish comparison lists to rank for category queries, they update them regularly, and many have a stated process for adding a product. Our classification of who counts as media is a judgement call on a few domains; it does not change the picture.

Proxies run on affiliates, ITSM on editorial

The two niches reach the same answer through different kinds of pages. In ITSM, almost everything is an editorial list. In proxies, a third of the sources are commercial by design: eight affiliate programme pages, six review and aggregator sites and three homepages with sponsor blocks, seventeen pages in all.

That follows the money in each market. Proxy providers have run affiliate programmes for years, and a whole layer of sites exists to rank providers for a commission. Assistants read those sites like any other source. ITSM vendors sell through demos and long sales cycles, so the comparison content is written by vendors and media rather than by affiliates.

The route onto each page differs accordingly. In ITSM, 82 of 90 pages call for the same step: contact the publisher and ask what inclusion requires. In proxies, only 32 of 61 do; the rest needs a link inside an existing article (11), an affiliate programme of your own (6), a provider profile or review page (6), a request to be added to a review site’s rating (3) or a place in a sponsor block (3).

Per vendor: two routes to the same answer

Within proxies, the two vendors with enough data took visibly different routes.

Bright Data is cited mostly from rankings and commercial placements: affiliate, review and sponsor pages make up thirteen of its 34 sources. IPRoyal gets more than a third of its citations from ordinary blog articles that mention it in passing, the kind of mention that link insertion produces. Both end up in the same answers. A smaller vendor does not need Bright Data’s budget to be named; it needs to be present on enough pages of whatever kind the assistant reads in that niche.

In ITSM the four vendors look alike: rankings make up between 86 and 94 percent of each vendor’s sources, and Freshworks, with the widest category coverage, simply has the most of them at 31.

What this means if you want to be named

Treat AI visibility as a placement problem first. Rewriting your own pages to be quotable helps, but in both niches the assistants cited third-party pages, not vendor sites. The work that moves the answer is getting onto the pages the assistant already reads.

Build the list before anything else. The sources are a long tail of 60 to 80 pages per niche, almost none of them shared. You cannot guess them; you have to collect them from real answers for your category, competitor by competitor.

Start with competitors’ comparison pages. In ITSM most cited rankings are written by other vendors. Many of them will add a credible product, some for a fee, some for a review, some for nothing more than accurate data.

Check how your niche pays for lists. If, like proxies, it runs on affiliates, an affiliate programme of your own is part of AI visibility rather than a separate channel.

Publish your own comparison page. Nearly every source here is a list with named alternatives. If you sell in a category, a fair comparison of your category on your own site is the page type assistants are built to read. The practical guide to getting named in ChatGPT walks through the steps.

This is also why we treat link building and AI search as one piece of work: the outreach list for a SaaS or B2B product and the list of pages an assistant cites are, in these two niches, largely the same list.

What this data cannot tell you

Five limits are worth stating plainly. The data covers two niches and seven vendors, so it shows a pattern, not a law. It records which pages were cited, not how many prompts were run or how often each vendor was named. It does not break the citations down by assistant, so it cannot say whether ChatGPT and Perplexity lean on different pages. One vendor, Webshare, has a single page in the set and is left out of per-vendor comparisons. And it is a snapshot from 2026: assistants change how they retrieve sources, and a page cited today may not be cited next quarter.

None of these change the main finding. In both niches, the answer an assistant gives about vendors is built on comparison pages written by someone other than the vendor.