A billion data points confirm AI loves "Best X" lists — until it doesn't
On June 2, Ahrefs CMO Tim Soulo published ten findings from fourteen AI-search studies — a billion data points — and the one that traveled farthest says to write more "Best X" listicles. In the same thread, Soulo quoted Lily Ray reporting that Google was already trimming self-promotional entries from those lists, and added his own note: "there's a good chance this won't work forever."
43.8% is not a quality score — it's a census. Listicles dominate AI citations for the same reason they dominated Google search: they make up most of the web, and that is what the models trained on.
In the same thread, one reply put a deadline on it: 'six months until the chatbots start penalizing the pattern.' SEO's two-year trap, new platform.
Fourteen studies, one billion data points, one tweetable finding
Tim Soulo, CMO of Ahrefs, published ten findings on X on June 2 from the company's six-month AI search research initiative: fourteen studies, over a billion data points covering ChatGPT, Google AI Overviews, AI Mode, Perplexity, and brand visibility signals. The thread drew 3,500 likes and 141 replies within days. The finding that spread fastest: "Best X" blog listicles are the single most prominent content format cited by AI chatbots, making up 43.8% of all page types cited by ChatGPT specifically.
The thread that followed stress-tested each finding, and several cracked under pressure.
What the full dataset shows beyond the 43.8%
The listicle number is not the whole story. Ahrefs' research surfaces several findings that complicate the "write more lists" read:
67% of ChatGPT's 1,000 most-cited sources are unreachable by marketers: Wikipedia (29.7%), home and landing pages (23.8%), app store entries (6.6%). Only 32.3% falls into influenceable territory: educational content, reviews, news, blogs. Most AI SEO advice targets the minority.
28.3% of ChatGPT's top-cited pages have zero organic visibility in Google. They get cited repeatedly despite not ranking in search at all. Soulo called this "a completely separate discovery layer."
ChatGPT retrieves roughly 33 URLs per query and cites about 50% of them. Being fetched and being attributed are two different outcomes. Almost every content optimization strategy treats them as one.
Schema markup: near-zero impact. AI Overviews dipped −4.6% after adding it; AI Mode and ChatGPT moved +2.4% and +2.2%, both within noise. A finding that should retire several agency retainers.
YouTube mentions carry the highest single correlation (0.737) with AI brand visibility, higher than backlinks, domain rating, or page count. This held across both Google-owned products and OpenAI's.
AI Overviews now cut clicks to the top-ranked Google result by 58%, up from 34.5% ten months earlier. And 99.9% of AIOs trigger on informational queries; shopping triggers one only 3.2% of the time.
"The format wins because the web is 80% listicles"
The sharpest response to the 43.8% finding came not from skepticism about methodology but from the simplest explanation. Antoine, a content and marketing strategist in the thread:
"43.8% of AI citations go to listicles and the whole GEO industry is about to spend 2026 rebuilding every client's content as 'best X' roundups. the format wins not because it's good content, it's because LLMs were trained on the internet and the internet is 80% listicles." — marketing consultant and content strategist
Prithvi Jadwani, an SEO specialist in the thread, cut it shorter: "Most 'Best X' lists have zero inherent value, so it's no surprise AI cites them without regard for Google rankings."
The corollary: the moment AI systems retrain on a corpus full of AI-generated "Best X" articles built specifically to game the citation layer, the signal collapses. Several voices in the thread think the collapse is already in motion.
SEO's 2-year trap on a new platform
WillyV3, in the thread:
"'best X' listicles winning AI citations is the SEO trap reborn. you can game it for 6 months until the chatbots start penalizing the pattern, then everyone who built around it has to rebuild. SEO's entire history is this exact loop on a 2 year cycle" — SEO practitioner in the thread
Soulo's own reply confirmed the concern. When Lily Ray, one of the most prominent names in technical SEO, noted that Google was already adjusting how it handles self-promotional listicles, filtering out brands that rank themselves #1 in their own "best X" posts, Soulo quoted her and added: "there's a good chance this won't work forever." The CMO of the company publishing the finding signaled the shelf life in the same thread.
The "separate discovery layer" is probably just the Bing index
The 28.3% number (pages with zero Google visibility that ChatGPT cites heavily) spread as evidence of a new parallel internet that AI has discovered. The thread had a simpler answer.
"ChatGPT doesn't use Google. It uses Bing-like search." — developer in the thread
Multiple people confirmed it independently. ChatGPT draws from Microsoft's Bing index, not Google's. A page can rank clearly in Bing while being invisible in Google. Calling that discrepancy "a separate discovery layer" is technically accurate but frames it as more mysterious than it is.
YouTube correlation of 0.737: cause or proxy?
The YouTube finding is the one most likely to drive action, and Shubham Bansal, an analyst in the thread, flagged the confound before anyone had time to build a Q3 strategy around it:
"0.737 is huge for a single channel. but is youtube driving citations, or is it just that brands big enough to invest in video are also big enough to be everywhere else the models look?" — data analyst in the thread
The correlation could be causal: AI systems scrape YouTube transcripts, find brand mentions, and cite those brands in answers. Or it could be a proxy for brand size and marketing investment. Ahrefs' study establishes the correlation; it does not isolate the mechanism.
Being known beats being optimized
Ravi, a digital strategist, pulled the thread together:
"The calm, less-FOMO way to read all this: AI search doesn't reward optimization, it rewards being known. Stop hacking the algorithm, start earning the mention. The brands that win the next decade aren't the ones with the best schema. They're the ones people already talk about." — digital strategist and marketer
That's consistent with what the full dataset shows: schema markup moves nothing, backlinks correlate less than YouTube presence, and two-thirds of the most-cited sources are entities (Wikipedia entries, brand home pages, app store listings) that no content operation can manufacture.
The listicle finding is real. 43.8% measures what has been cited, not what should be built. "Best X" formats dominate because they're the modal form of content on the web, and AI systems learned from that web. Once those systems retrain on a corpus of AI-generated "Best X" posts built to game the citation layer, the signal degrades. By Lily Ray's account, the degradation has already started.
Two games now run in parallel. Ranking in Google search and being cited by AI use different indexes, respond to different signals, and reward different content forms. For a given query, Google's AI Mode and AI Overviews reach the same conclusion 86% of the time while citing almost entirely different pages (13.7% source overlap). Being good at one does not make you good at the other. The consultant pivoting to listicles for AI visibility is not wrong — just one retraining cycle early.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
The EU's open-source strategy: four objectives, no euros attached yet
Brussels put open source 'at the centre of the EU's technological sovereignty' — alternatives to US cloud, workplace tools and email. The thread's first question was the one the document never answers: where's the money?