Blog

They ranked themselves first. ChatGPT mostly stopped naming them.

In February 2025 ChatGPT named 21 agencies that had ranked themselves first on their own best-of lists. In November 2026, nine of them are never named, and the agency that refused to write one is named in 58 of 90 answers. What the test shows, and what it cannot.

A crumpled violet paper crown lying on its side on a white floor, the self-awarded crown nobody picked up.

TL;DR

Wil Reynolds of Seer Interactive re-ran three ChatGPT prompts from February 2025: of the 21 agencies named then, nine are now never named and the bottom 17 together are named about 5% of the time.

Seer, which never published a self-ranked list, is now named in 58 of 90 answers, but the 2025 baseline was one run and the prompts were chosen to favour Seer.

Larger studies call the self-ranked list a weak tool, not a dead one: Scrunch found the author recommended 11% of the time when its list was cited, and an Italian audit found popular brands cited 1.4 to 2.8 times as often.

What happened

In February 2025, Wil Reynolds of Seer Interactive asked ChatGPT three questions about SEO agencies. ChatGPT named 21 agencies. It never named Seer.

Reynolds looked at the 21. Over and over, he found the same thing: an agency publishes a "best SEO companies" page, puts itself at or near the top, and fills the rest of the list with other agencies. Seer had never written one.

On 12 November 2026, Reynolds published the re-run. Same three prompts, asked 10 times each on three current ChatGPT models with web search on, 90 answers in all. Each of the 21 agencies was scored on the prompt that named it in 2025, which gave each of them 30 chances.

Nine of the 21 were never named, not once in 270 tries. Eight more were named in fewer than one answer in three. The bottom 17 together were named 27 times in all their chances, about 5%, or roughly 1 in 20.

Seer, which never ranked itself on a list, was named in 58 of the 90 answers.

The belief this tests

The belief is simple and widely sold: ChatGPT reads "best of" lists, so publish one with yourself at number one and ChatGPT will name you.

Ahrefs, which sells SEO tools, published a study in December 2025 of 26,283 source URLs that ChatGPT cited across 750 top-of-funnel prompts. "Best X" blog lists made up 43.8% of the page types cited. In Google's own results for 250 "best X software" searches, 67.6% carried a list in which the company writing it ranked itself number one. Ahrefs read the numbers as a reason for agencies and software companies to publish such lists.

That 43.8% is now eleven months old, and it came from prompts that ask for a list, so lists were always going to be cited. It says lists get cited. The harder question is whether the brand that wrote the list gets named.

Reynolds' test asks the second question. For the 21 agencies that had answered yes to it in 2025, the answer in 2026 is mostly no.

What ChatGPT cited instead

In the 58 answers that named Seer, ChatGPT cited Seer's own site in 21 of them. The pages were the homepage and the "work" pages, not research pieces. The four other most cited pages in those answers were listicles on other agencies' sites. Each of them ranks its own publisher first. Two of the four list Seer, at number three and number four.

So the self-ranked list has not vanished from ChatGPT's reading. It has stopped paying the author. Lily Ray of Algorythmic, which sells SEO and AI search consulting, found the same shape in Google's AI Overviews across 100 "best B2B software" queries in September 2026: in 70% of cases, the self-ranked list helped a competitor get recommended while the brand that published it was left out.

Reynolds credits Seer's rise to the things that leave a trace off its own site. In twelve months, ten Seer researchers had their work picked up by 19 publications. He also says plainly that he cannot draw a straight line from any one talk or post to ChatGPT naming the agency.

The other side

This is an agency publishing a test about its own visibility, and Seer sells the kind of work the test flatters. Reynolds lists the problems himself. The 2025 baseline was a single run of each prompt. The 2026 runs went through the API, with no memory and no personalisation, on newer models, so the two setups are not identical. And the three prompts were chosen in 2025 to favour Seer. Two of them now name Seer in 25 and 28 answers out of 30. The third names it in 5.

Larger studies say the self-ranked list is a weak tool, not a dead one. Scrunch, which sells AI visibility software, compared 818 self-ranking pages with 1,033 neutral ones across seven AI engines between May and June 2026, in 5,434 page and response pairs. When a self-ranked list was cited, a competitor was recommended instead of the author 24.3% of the time, and the author brand was recommended 11% of the time, 8.6% on ChatGPT. Writing the list raised the author's odds a little, from about 4% to about 7%.

And size still counts. Three researchers at Politecnico di Milano audited 1,176 brands across OpenAI's GPT, Gemini and Perplexity in four Italian sectors, in a preprint posted on 5 November 2026. Brands in the most popular fifth were mentioned and cited 1.4 to 2.8 times as often as the average brand in their sector. Rewriting a page while keeping its facts moved its rank by about 0.02 positions. Swapping the real brand name for an invented one cost 0.44 positions. The name did the work, not the copy.

The Indian version of the shortcut

The shortcut is on sale here too. A Kolkata agency's guide to getting mentioned by ChatGPT tells brands to "pitch the authors of 'top 10' articles that AI cited" and claims that one placement in a cited listicle "outworks months of blogging". No number backs the claim.

Prashant Puri of AdLift put the counter-argument in afaqs in October: "Any company can call itself a category leader on its own homepage." His advice was to earn Indian third-party coverage instead.

On a client call this morning, a healthcare software brand asked us to reach out to the third-party listing sites in its category on its behalf, and its team is building a list of the editorial sites AI already cites there, with traffic data. Nobody on the call asked for a best-of page of their own. That is the right instinct.

What this means for brands

  1. Ask ChatGPT the three or four "best" questions in your category and read the cited pages. Note who wrote each one and whether the author ranks itself first. If a rival's list is being cited and you are not on it, the job is to get onto that page or the third-party page beside it, not to publish a copy with your name on top.
  2. Spend on the things that leave a trace off your own site. Original research a trade publication will pick up, reviews on sites buyers already use, talks and press. Reynolds cannot show these as citations, and says so, but they are what the agency that won did differently from the 21 that did not.
  3. Keep your home page and your proof pages plain and current. They are what ChatGPT cited when it named Seer. Do not expect a rewrite alone to move you: in the Italian audit, rewriting barely registered. Then run the same buyer prompts every week and count how often you are named. One run proves nothing, and Reynolds' 2025 baseline is the lesson.

Sources

Date

November 13, 2026