Blog · 24 September 2026 · By HeardOf

"Best X" lists as AI sources: what five studies found

Do listicles help AI visibility, by what the five studies measured?

They get cited, and all five studies read for this post agree on that; where they part is on whether a citation of a brand's own list turns into a recommendation of that brand: Lily Ray found the brand left out 69% of the time, and Ahrefs' experiment 43% of the time for a new conference and 11% for an established product. The five are in the table below, each opened and rendered on 24 September 2026: two by Ahrefs, one by Peec, one by the Wix AI Search Lab on Peec's data, and one by Lily Ray. Four were published more than 90 days before the read and are labelled historical; the Ahrefs experiment of 6 July 2026 is not.

The buyer's question, do listicles help AI visibility, splits into three that the studies answer separately: how often lists are cited, how much of that is self-promotion, and what a citation is worth to the brand that wrote the list. Each figure keeps its unit, its engine and its date, because the five count different things — a page type among source links, a citation, an answer, a prompt-and-engine slot — and none can be placed on another's scale.

How often are "best X" lists cited in AI answers?

In Ahrefs' ChatGPT sample, "best X" blog lists were 43.8% of all page types across the source links; in the Wix and Peec study, listicles were 21.9% of 1,056,727 citations across ChatGPT, Google AI Mode and Perplexity. The two are not the same measure. Allsopp's 750 prompts were top-of-the-funnel recommendation queries for software, products and agencies, and his count is of page types among first-party and third-party source links, the 10,000-plus third-party URLs classed through filters and a GPT-5 pass and then reviewed by group. Wells's 75,000 answers span four intents and five industries: listicles reach 40.86% of citations for commercial prompts, 35.37% in SaaS and 25.24% in professional services, against 5.36% for navigational prompts. Allsopp adds that the same lists "were slightly more prominent in Google's AI Overviews", and that the 1,000 most-cited pages on each platform in Ahrefs Brand Radar, across 150M-plus prompts, all included comparison listicles. He prints five example prompts and not the set; Wells describes his by intent and industry.

How much of that is self-promotion?

About one citation in ten, on Peec's count: across 232,000 citations from a fixed set of non-branded software-review prompts on six platforms between December 2025 and February 2026, roughly 11% came from listicles in which the publishing company ranks its own product. The rate differs by engine — a 3.6% average on ChatGPT against 10.3% on Google AI Mode and 10.4% on Perplexity — and over the twelve weeks the post reports December and January at 11 to 14% and a "slight improvement" for Google AI Mode in February, to 9 to 10%, which it reads as "no evidence of algorithmic correction over time". Peec's detection rule is a domain match — a listicle on zapier.com that recommends Zapier is self-promotional, "only if Zapier was promoting itself as a product within that listicle" — and the page does not say how many listicles the rule mislabels. The Wix study cuts the question differently: among the 1,000 most-cited URLs in professional services, which the page calls the industry with the highest listicle citation rate though it prints 35.37% for SaaS against 25.24% for professional services, 19.1% of listicle citations were self-promotional and 80.9% went to third-party lists (the page labels both "of citations"). On the Google results page rather than in the answer, Allsopp found that 169 of 250 "best X software" searches, 67.6%, "featured a list in which the company writing the article ranks itself number one". From her AI Overviews data, Lily Ray pulled 184 self-promoting listicle pages across 146 brands and dated each by the first month it drew organic traffic in Ahrefs; those dates stack up in 2025.

Does a citation mean a recommendation?

Often not, and how often depends on the brand: Lily Ray counted the brand left out 69% of the time, and Ahrefs' experiment 43% for a new conference and 11% for an established product. Lily Ray took 100 B2B "best [category] software" queries, read Google's AI Overview for each on 15 April, 15 May and 8 June 2026 through the Ahrefs Brand Radar MCP, and separated the brand's own list appearing among the sources from the brand being named as a pick in the answer. Eighty of the hundred queries returned an AI Overview; across them, when a brand's own listicle was cited, the brand was left out of the recommendation 69% of the time, 224 of 323 citations, and 74 of the 100 queries returned at least one answer that cited a self-promoter and omitted it. Her examples show the shape — a help-desk vendor cited under a competitor's entry, an LMS vendor cited in-line and in the sidebar and absent from the picks — and her reading of who does get recommended, the brands with more referring domains and more existing AI mentions, is an inference she labels as such.

Ahrefs' experiment measured the same seam from the publisher's side. Makosiewicz published 34 pages promoting a new conference and an established product on five domains, and tracked 9,886 answers from ChatGPT, Gemini, Perplexity and Copilot between 7 February and 31 May 2026. The conference entered 72 prompt-and-engine slots where it had not been mentioned during the 7–12 February baseline, and 82% of those new mentions came in answers citing an Ahrefs page. But among answers that cited a conference page, 43% never mentioned the conference, "recommending a competing event featured on the same page instead"; for the product pages the figure was 11%, and widened to pages the engines retrieved without citing, the conference figure rose to 74%. For the established brand, only 6% of its new mentions came from answers citing one of the 34 published pages.

One figure is ours and sits under the table rather than in it, because it is on none of the five studies' scales: 14 of 14 best-of lists of AI-visibility tools naming two or more tools put the author first, in our read of 24 September 2026.

Do the citations last?

Mostly not, in the study that tracked pages day by day. In Ahrefs' experiment, about a quarter of the page, query and engine combinations were cited once and never again; between a page's first and last citation it was cited on about one in three eligible days; the longest unbroken run was 52 days, and the pages that lasted were nearly all on Copilot. The post relays a figure from a separate Ahrefs study of 43,000-plus keywords in AI Overviews, which it does not date, that about 54% of cited URLs carried over between consecutive checks; we did not open that study. Peec's weekly self-promotion rate, a different measurement, is one its post calls stable over twelve weeks.

What do the vendors say about publishing such lists themselves?

Four pages, and they disagree: Ahrefs' December post says publish them, Ahrefs' July post says do not rank yourself first, Peec says do not use them, and Wix's guide takes neither side. Ahrefs' December post concludes that "Agencies and SaaS companies should publish self-promotional "best" lists", and states its own practice: comparison posts stay under half a percent of its blog, it "won't always put ourselves in the top spot", and it links to the alternatives it names. Ahrefs' July post, by a different author: "The problem starts when you position yourself as "the best" or park yourself at the top of a list you wrote". Peec's post, whose data shows the tactic still cited, says "We still don't recommend using self-promotional listicles", naming reputational and algorithmic risk. Wix's guide, which relays the Wix and Peec figures and carries 30 April 2026 in its structured data and "9/10/26" on the page, closes with "listicles aren't a golden ticket to AI search visibility. But they aren't inherently wrong, either." Ahrefs and Peec both sell AI-visibility tracking: Peec's data underlies both Wells studies, Ahrefs Brand Radar collected the experiment's answers and Lily Ray's answers came through the Ahrefs Brand Radar MCP, while Allsopp's post names Brand Radar only for its cross-platform chart.

What the AI-visibility vendors do, as opposed to say, is on record: in the vendors' own best-of lists, every one of the 14 lists on the Profound, Otterly, AthenaHQ and Geoptie blogs that names two or more tools puts the author's tool first, as rendered on 24 September 2026, and none of Peec's 44 posts matched that post's rule. That post checks the prices those lists print; this one does not repeat it. The vendors' customer case studies are a different page type, and our read of GEO case studies lists what the headline lift figure on each page it read counts.

How was this compiled?

The seven pages — the five studies, Wix's guide, and Lily Ray's earlier post of 3 February 2026 on the January organic demotions, read here and not quoted — were fetched with curl and a browser user-agent between 19:39 and 22:15 UTC on 24 September 2026, and the six that are quoted were rendered in Chrome 153 headless with JavaScript, a nine-second wait and two scrolls, between 22:11 and 22:17 UTC. Every figure printed here was found in both the served text and the rendered text of its page; the counts from Lily Ray's spreadsheet come from the sheet as exported on 24 September 2026. Dates are the page's byline; where structured data disagrees, both are printed. Historical means published before 26 June 2026.

What was not done: no study was re-run and no figure was recomputed from anything but its page. The Seer Interactive and Profound figures that Peec's and Wix's pages relay were not opened, and none appears above. Allsopp's distribution of list positions is a chart with no figure in the text, and this post does not report it; his Domain Rating section prints one figure in a heading, 35% of cited best lists on low-authority domains, with no threshold for low, and this post does not print it. Lily Ray links a public spreadsheet, opened on 24 September 2026: 229 rows, one per query and date, over 99 queries, each row holding the AI Overview's recommended brands and the self-promotional listicles it cited, marked recommended or not; 80 of the queries returned an overview and 74 had a self-promoter's list cited with the brand left out, matching her post; its count columns sum to 496 of 726 cited self-promotional lists not recommended, 68.3%, and we could not find the counting that gives her 224 of 323. Allsopp's supplemental document holds no per-unit data. None of the other pages links a per-unit download; Peec's listicle study is one of the 33 measurement posts scored in our read of the vendor studies, read 24 September 2026, where it states a sample and a month, prints no prompt and links no data. Everything this post says about our own read of the vendor lists comes from us, carries its date in the sentence that says it, and cannot be checked; everything else carries a page, a byline and a date you can follow. The audit this site sells is described in the post on which engines we check, as updated 23 September 2026, and that description is the source of the one sentence about it in the closing.

GEO case studies: the lift figures the vendors publish, and what each one measured