Should you pay to be listed in “best X” articles to get recommended by ChatGPT?

List articles are common citation sources. In Glen Allsopp’s December 2025 analysis for Ahrefs, “best X” lists accounted for 43.8% of pages cited in answers to 750 early-stage questions about software, products and agencies. Sellers often quote that figure without its date or sample. It describes those 750 informational prompts, not all AI citations. Our own, smaller sample points the same way with lower numbers: in our page-types study, a classification of 569 links cited across 120 answers by ChatGPT, Perplexity, Google AI Mode and the Gemini app on 14–15 September 2026, 16% of links pointed to directories and comparison sites, 48% to vendors’ and agencies’ own sites, 10% to forums and 4% to media. On questions about choosing a provider the directory share rose to 25%, against 33% for providers’ own pages.

The second fact explains the first. OpenAI’s help page on searching the web says ChatGPT typically rewrites a question into one or more targeted queries, may send further, more specific queries after reading the first results, and ranks what comes back “using multiple factors”, with placement not guaranteed. When a buyer asks which GEO agency to hire, the model is not recalling a ranking. It pulls several pages that already rank for that question, reads them, and compiles its own shortlist. A list article that ranks for the question is exactly the kind of page it pulls.

A placement gives you a chance to appear on a page the model uses as a source. The checks below help establish whether a particular paid page serves that purpose.

Sellers offer three different kinds of placement. The distinction matters when interpreting the 43.8% figure.

A page that assistants already use: an existing article on a domain that ranks for the buyer’s question, where ChatGPT, Perplexity or Google’s AI features can be shown to cite it today. This is the only version where the citation share applies to you. It is also the rarest and the most expensive, because the editor has something to lose.

A page on a ranking domain, written for you: a new “10 best” article on a site that ranks for similar topics. It may get picked up, or it may sit unindexed. You are paying for a probability, and the seller should be able to show earlier articles of the same type that did get cited.

A page on a network: a link vendor’s own sites, built to sell placements. One vendor with “thousands of sites” wrote publicly in June 2026 that clients start at around $3,000 a month and grow to $10,000, and that AI SEO agencies “sell the same link building in the form of listicles”. Those are the vendor’s words, in the same post where he says real prompts cannot be tracked and results differ from one computer to another. That is not a hidden catch. It is the product description.

When four things are true at once. In the audits we run, a placement that fails any one of them has not produced a citation we could attribute to it.

Be cautious when a seller emphasizes the list format but provides no evidence about the actual page. Check for the following warning signs.

The compliance question nobody in this market likes: Google asks that paid placements carry rel="sponsored", and its spam policies list “exchanging money for links, or posts that contain links” as link spam and pages “generated for the primary purpose of manipulating search rankings” as scaled content abuse. Whether ChatGPT, Perplexity or Claude apply similar filters is not documented by any vendor. We do not know, and neither does the seller. What we do know is that a page built only to host paid entries tends to have no other reason to be cited.

One more line you will hear from sellers: “brand mentions are three times more important than links for AI search”. It comes from dividing two correlation coefficients in Ahrefs’ May 2025 study of 75,000 brands with Domain Rating above 40: mentions correlated with AI Overview visibility at 0.664, raw backlinks at 0.218. Every site in that sample already had a strong link profile, so links could not explain much of the difference between them, and the authors themselves write that correlation is not causation. The study does not show that a mention without a link works for a company the model has never seen. Treat the “three times” as folklore.

In twenty minutes, with no tools beyond the assistants themselves. The point is to see whether the seller’s page is on the list of sources the model actually reads for your buyers’ questions.

Assistants disagree with themselves between runs, which is why each prompt is run three times and why a single screenshot from a seller proves nothing. If the page shows up in two or more of your five prompts, a placement is a reasonable test at a per-page price, with a link to the page of yours that answers the same question. If it shows up in none, the seller is asking you to fund their page’s SEO.

We have not bought a placement in anyone’s list, so we cannot show you a result from one. What we can show is how the mechanism looks from the inside. When we asked ChatGPT in late August 2026 to shortlist GEO agencies for English, Spanish and Russian, it composed its own list from pages it searched and named us first for that language combination. The shortlist was the model’s, built from sources it chose, and the pages it chose were narrow guides, not lists. That is the mechanism paid placements try to buy into, and the reason our own comparison page, with competitors named and criteria published, is on our roadmap rather than a placement budget.

The table below is how we weigh the three options when a client asks. If you want to know which pages assistants read when buyers ask about your category, before you pay anyone to be on them, the free audit covers it: ten prompts, one language, the current answers, and the specific list pages that showed up in the citations.