Is your 'best of' list helping rivals? Ahrefs' experiment reveals how AI cites your content but recommends competitors. Learn the AI SEO truth!
Publish a "best CRM software" list, rank your own product first, and wait for AI assistants to start recommending you. That's been the playbook for two years. It's also, according to a controlled experiment Ahrefs just ran, a good way to get your own article cited as a source while ChatGPT or Google's AI Overview recommends one of the competitors you listed instead of you.
The experiment that separated citation from recommendation
Ahrefs researcher Mateusz Makosiewicz published 34 self-promotional pages, listicles, reviews, and opinion pieces, across five domains between February and May 2026, promoting two of the company's own products: its new conference brand, Ahrefs Evolve, and its established AI-visibility tool, Ahrefs Brand Radar. Over the tracking period he pulled 9,886 answers from ChatGPT, Gemini, Perplexity, and Copilot to see what actually happened when those pages got picked up.
The results split cleanly by how well-known the brand already was. For Ahrefs Evolve, a genuinely new conference with no existing reputation to lean on, the self-promotional pages worked as a discovery mechanism: 82% of the times Evolve appeared somewhere it hadn't been mentioned before, one of the team's own pages was the cited source. But among the answers that cited one of those Evolve pages, 43% never mentioned Evolve in the actual recommendation. The AI used the page as a research source, then recommended a different conference instead, one that happened to be named in the same article as a comparison.
Brand Radar, the established product, told a different story: only 11% of its citations failed to translate into a mention. When you already have a reputation, the citation and the recommendation tend to travel together. When you don't, they can split apart, and the competitor sitting in your own comparison table is what fills the gap.
Google's version of the same problem, measured separately
This isn't an Ahrefs-only finding. Lily Ray ran a parallel analysis specifically on Google's AI Overviews, tracking 100 B2B "best [category] software" queries across three checkpoints between April and June 2026. Across the 80 prompts that actually triggered an AI Overview, a brand's own self-promotional listicle got cited as a source in 323 instances, and in 69% of those, the brand was left out of the recommendation entirely. Seventy-four of the 100 tracked prompts showed the pattern at least once.
Ray's screenshots make the mechanism concrete. Oasis LMS ranks itself first in its own "best LMS for selling courses" article, gets cited prominently in Google's AI Overview for that exact query, and doesn't get recommended. The brands its article compares itself against, Kajabi, Thinkific, LearnWorlds, and Teachable, are the ones the AI actually names. The same pattern repeats with Kredo Learning (cited, not recommended; TalentLMS and Docebo get the nod), Pylon and Crisp (cited, not recommended; Zendesk, Freshdesk, and Help Scout do), and TMetric (cited, not recommended; Todoist, Asana, Trello, and ClickUp do). In every case, the excluded brand's own article is what supplied the AI with the competitor names that ended up getting picked.
Ray's data points to the same split Ahrefs found: authority decides which side of the split you land on. For "best CRM for small business," the one self-promoting brand that also got recommended was Monday.com, sitting on a Domain Rating in the 90s and 75,000 referring domains. The four brands that got cited but excluded had nowhere near that backlink profile. Checking a domain's referring-domain count against its competitors before publishing a self-ranked list would have told you which side you were likely to land on.
The complication: the data doesn't universally say "don't do this"
Here's where it gets less tidy. A separate Ahrefs study by Glen Allsopp, analyzing 750 ChatGPT prompts across software, agency, and product categories, found that "best X" listicles were the single most-cited page type in AI answers, 43.8% of all sourced pages, including ones where the publisher ranked itself first. Allsopp's own conclusion, based on that data: agencies and SaaS companies should keep publishing these lists, because his research turned up no clear downside and a real correlation between list position and recommendation frequency.
Two researchers at the same company, looking at overlapping questions, land in different places, and the reason is what each one actually measured. Allsopp tracked whether "best" lists get cited and whether list position correlates with getting recommended somewhere in an answer. Makosiewicz and Ray specifically isolated cited-but-not-recommended as its own outcome, the case where your page did the work of getting into the answer's source list but a rival's name is what actually reached the user. That distinction doesn't show up if you're only counting citations. It's the entire finding if you're counting recommendations.
What this actually changes
The tactic isn't dead, and the fix isn't "never publish a comparison page." It's narrower than that: know which side of the authority line you're on before you hand a competitor free real estate in your own content.
If your brand already has real third-party recognition, meaningful referring domains, existing mentions across other people's content, an established market position, a self-ranked list is closer to free upside. You're the Monday.com case: the citation and the recommendation are likely to travel together.
If you're the newer or smaller name in the category, a self-promotional list is a legitimate way to get discovered, the Ahrefs Evolve data shows that clearly, but treat the competitors you name in it as a real risk, not a formality. Every brand you compare yourself against is a candidate the AI can recommend instead of you, using your own page as the source. If you can meaningfully avoid naming a direct, better-known rival in your own comparison table without misleading the reader, that's not a small stylistic choice.
Either way, don't confuse a citation with a win. Makosiewicz's data also found citations are mostly temporary, cited on roughly one in three eligible days between a page's first and last appearance, so even in the better case, a citation isn't a stable asset. What both studies agree actually moves the needle for smaller and mid-sized brands is the same thing that's always mattered: getting named on other people's pages, not just your own. A vetted, high-authority third party recommending you carries the AI-recommendation weight your own self-ranked list can't buy on its own.
