Listicles in the age of ChatGPT: why they get cited and when they backfire

Ask an AI assistant which CRM, camera or web design agency is best, and many of the pages it cites will be list articles. Across 750 prompts of that kind, Ahrefs found that “best X” blog lists made up 43.8% of the page types ChatGPT cited.¹ Lists have become the most copied tactic in AI search as a result, and they have brought a familiar problem with them: companies publishing lists of the best products in their category with their own product at number one.

This post sets out what the studies show about lists in AI answers and why assistants lean on them, what happened to self-ranking lists in 2026, how Google's policies and UK and US consumer law apply, and how to write a list that deserves to be cited or earn a place in someone else's. How assistants find and cite sources in general is covered in the AI search guide.

Three studies with a stated method point the same way. Each defines a list differently, so the shares differ.

The spread, from about 17% to 44%, mostly reflects what each study counted. Ahrefs used only buying-style prompts such as “best crm software for enterprise”, Peec AI split prompts by intent, and Seer used a crude rule on the URL. The two that isolated buying questions both found lists to be the largest single source type for them. The assistants differ around the edges: in Peec AI's data Perplexity drew 17.35% of its citations from discussion pages such as Reddit, against a 7.52% average.²

An assistant answering a buying question has to name several options and give a reason for each. A list article that compares ten options for a stated use supplies most of that in one page, so it is an efficient page to retrieve. That is my reading: none of the provider documentation I have read says how page types are weighed. What is documented points the same way.

The searches an assistant runs often have the same shape as a list's title. Google describes query fan-out as a set of concurrent, related queries that the model generates to fetch more results, and its own example of fan-out queries for a weedy lawn includes “best herbicides for lawns”. Those queries are answered from its Search index, using its core ranking systems.⁴ List articles are already common in the results for “best” searches, as the next section shows.

AI search also leans towards third-party sources. A University of Toronto team (Chen and others, September 2025) ran ranking prompts such as “Top 10 ... brands” in the US and Canada and sorted the sources into brand-owned, social and earned media, meaning independent media, review and comparison sites. For US software prompts, GPT-4o with search drew 72.7% of its sources from earned media, against 45.4% in Google's results; for US consumer electronics the earned share was 92.1%.⁵ A third-party list counts as earned media in that scheme. A vendor's own list is brand-owned.

Freshness helps lists too. Ahrefs (Ryan Law, July 2025) analysed about 17 million URLs from AI citations and Google's organic results, and found AI assistants cited pages 25.7% fresher than those in Google's organic results. ChatGPT cited the newest content, and Google's AI Overviews were the exception, citing pages slightly older than organic results. The average cited page was still 2.9 years old, so freshness is one factor among many.⁶ List articles are updated often: 26% of the dated lists in Ahrefs' ChatGPT study had been updated in the previous two months.¹

Across 250 Google results pages for “best X software” searches, Ahrefs found 169 (67.6%) with a list in which the company writing it ranked itself first. In ChatGPT's software answers, when a list's publisher was among the brands recommended, that publisher's own list appeared in more than a third of the responses.¹

Minuttia, an agency that sells AI search work, ran 18,084 software questions through six platforms in August 2026 with the data firm Convex Works. It counted a page as self-promotional when a software company published a list in a category it sells in and included its own product. On that definition, 43% of the lists AI platforms cited were self-promotional, 87% of those put the company's product first, and only 16% clearly disclosed that the company wrote them.⁷

Peec AI's tracking of 232,000 citations for software review prompts, December 2025 to February 2026, found a lower share. About one citation in ten came from a self-promotional list: 3.6% in ChatGPT, against about 10% in Google AI Mode and Perplexity. It found no evidence of the platforms correcting for them over the 12 weeks.⁸ Minuttia measured a share of cited lists and Peec AI a share of all citations, so the figures do not conflict.

On citations alone, then, a self-ranking list works, and Ahrefs concluded from its data that software companies and agencies should publish them. The same article raised the trust problem, and Ahrefs' own approach is to keep new comparison posts under half a percent of its blog, link to alternatives and not always rank itself first.¹ Being cited and being recommended are different outcomes, and 2026 is when the gap between them started to show in the data.

Lily Ray, who writes about SEO on her Substack, published three sets of data on this during the year.

• February. Seven SaaS and software sites saw sharp falls in their Sistrix US visibility, six of them by 29% to 49%, mostly from mid to late January 2026 and mostly in their blogs. Each had self-ranking lists, from 10 pages on one site to 340 on another, which she counted with site: searches. Google did not confirm an update, and Ray noted other shared traits, including scaled and possibly AI-written content and artificial refreshing, so the lists may be only one part of the story.⁹
• June. She tracked 100 B2B “best [category] software” questions on 15 April, 15 May and 8 June 2026 with Ahrefs Brand Radar; 80 triggered an AI Overview. She found 184 self-ranking list pages from 146 brands. Across those prompts, 224 of the 323 self-ranking lists cited (69%) belonged to brands the AI Overview left out of its recommendation. The brands it did recommend were established category leaders; in one category she compared, they had far more referring domains and more brand mentions in AI Overviews and ChatGPT.¹⁰
• October. Rerunning 100 B2B software questions in September 2026, she reported that AI Overviews cited 38% fewer self-promotional lists than in June and left the publishing brand out of the recommendation 83% of the time. The update does not set out its method in the same detail as the June study.¹¹

ChatGPT has moved too, by one measure. Seer Interactive found ChatGPT's citations of list URLs fell 30% from December 2025 to January 2026, from 160,000 to 111,000, while its citations overall fell 22.7%. Seer read this as AI models beginning to shift away from lists where the publisher names itself first, though it gave no figure for those lists specifically.³

Two cautions apply. Peec AI's software data for December to February showed no sign of a correction,⁸ and Google's spam policies do not mention self-promotional lists.¹² I found no Google statement confirming an update aimed at them. What the data supports is narrower: a self-ranking list can still be retrieved and cited while the answer built on it recommends someone else. In my view a citation that hands the recommendation to a competitor is a poor return on the page.

None of Google's spam policies names the self-ranking list, but several cover the ways a list goes wrong. Google's definition of spam includes attempting to manipulate generative AI responses in Google Search.¹² The rules that apply, as they read in November 2026:

Site reputation abuse comes up often in arguments about self-ranking lists, but it covers third-party content published on a host site mainly because of that host's ranking signals. A company's own list on its own blog is first-party content, and Google says editorial work, and advertorials whose purpose is to share content directly with readers, are not abuse on their own.¹² The policy applies when a brand buys a list on someone else's domain.

Consumer law is stricter than Google on one point: hiding a commercial interest. In the UK, Schedule 20 of the Digital Markets, Competition and Consumers Act 2024 lists commercial practices that are unfair in all circumstances. They include using editorial content to promote a product where a trader has paid for the promotion without making that clear, and falsely creating the impression that a trader is not acting for purposes relating to its business.¹⁷ The provisions apply to practices from 6 April 2025, and the CMA's guidance gives “#Ad” as an example of clear labelling.¹⁸

Affiliate links bring in the advertising code. The ASA says the CAP Code applies in full to affiliate references on an affiliate's own site, that the commercial nature should be clear before readers engage, and that adding “Ad” to the title of a post is one acceptable way to do it. A disclaimer at the bottom of a post is unlikely to be enough, and the ASA found “may earn an affiliate commission” ambiguous.¹⁹

In the US, the FTC's rule on reviews and testimonials, announced in August 2024, prohibits a business from misrepresenting that a website it controls provides independent reviews or opinions about a category that includes its own products, and lets the FTC seek civil penalties against knowing violators.²⁰ A vendor's list presented as an independent roundup is the risk, and a plain disclosure at the top is the simplest protection. This is not legal advice: have named comparisons checked by someone qualified. How to write comparison pages buyers trust covers the UK rules on comparative claims.

The lists that hold up are the ones a careful buyer would trust. Google's guidance on reviews and helpful content, and the citation data, agree on most of what that means.

The local SEO tools comparison on this site follows these rules. It opens with a disclosure that one of the tools is my own product, says nobody paid to be included and there are no affiliate links, judges every tool on the same six questions, gives the date every price was checked, and says where each entry falls short, mine included.

On the evidence so far, third-party lists carry more weight than your own. In the 1,000 most cited URLs Peec AI examined in professional services, third-party lists drew 80.9% of citations and self-promotional lists 19.1%.² In Ray's June data, the brands Google's AI Overviews recommended were the ones talked about and linked to across many third-party pages.¹⁰ Getting into those pages honestly is slower work.

• Be the best option for something specific. List writers sort by use: best for small teams, for a tight budget, for one country. Know which of those you would honestly win, and say it plainly on your own site.
• Make your facts easy to check. Publish prices, plan limits and where you operate, and keep a page with current screenshots and a short factual description. A writer comparing twenty products uses the facts they can find and confirm.
• Offer real access. A trial account, a demo or a sample gives a reviewer something to test. If you give something of value, expect the review to say so; the CMA's own example of an unlabelled paid feature is a gifted holiday.¹⁸
• Earn coverage the way digital PR does. Original data, expert comment and stories a writer can use are covered in the links and digital PR guide, and how mentions feed AI answers in brand signals in AI search.
• Correct errors and leave the ranking alone. When a list gets your price or features wrong, send the author the source. Do not ask to be moved up.
• Treat any payment as advertising. A paid placement needs a clear label under UK law,¹⁷ paid links should carry rel="sponsored",¹⁶ and a bought list on someone else's strong domain can fall under site reputation abuse.¹²

Then measure it. Run the buying questions your customers ask in each assistant, note which lists are cited and whether you are named, and repeat on a schedule; how to measure visibility in AI search sets out a method. The lists that keep appearing are the ones worth earning a place in.

A list is still the right format for a buying question. Write it as if a sceptical buyer will check every line, disclose what you sell, and put as much work into being named in other people's lists as into your own.