Writing "we're the best" listicles for AI answers works, until it backfires. Ahrefs just tested it.
There is a tactic quietly spreading through AI SEO circles: publish your own "best tools in [your category]" listicle, put yourself at the top, and wait for the AI assistants to cite it. The logic is not crazy. Ahrefs' new experiment write-up opens with the evidence for it: Glen Allsopp analyzed 750 ChatGPT prompts and found that "best [category]" listicles were the most frequently cited source type in AI-generated answers (ahrefs.com).
So AI systems clearly lean on this content format. The question is what happens when the listicle is a costume, a self-promotional page dressed up as an independent roundup. Ahrefs ran an experiment on exactly that, and the title of their write-up gives away the arc: self-promotional content works, until it backfires. We will point you to their post for the full method and results rather than paraphrase data we did not produce (ahrefs.com).
We want to talk about the "backfires" part, because it is the part the tactic-sellers skip.
Machines are learning to smell the costume
An AI assistant citing sources has the same problem a human researcher has: some sources are honest and some are performing honesty. The performing kind has tells. Every comparison mysteriously won by the author. Superlatives without evidence. A "review" of nine competitors that links to one checkout page. Humans learned to discount those pages years ago, and the systems trained on human judgment are catching up.
Here is the structural problem with building your visibility on a costume: it works exactly until the day the model gets better at spotting it, and then it stops working all at once, possibly with your domain carrying the association. You have optimized for a detector's current blind spot. Blind spots close.
What does not backfire
Content that would survive being read aloud to your face. A comparison page that names the case where a competitor is the better choice. Claims specific enough to check. Prices that match across every page. An about page that says who you actually are. This is not moral advice, it is durability advice: honest pages do not have a detection risk, because there is nothing to detect.
We build AuditLamp on that bet. Our own scoring penalizes puffery signals and rewards checkable, structured, consistent facts, because that is what holds up in front of both audiences, the person and the machine.
If you want to know which of those signals your site sends right now, the free scan will tell you. It reads your site the way Google and the AI answer engines actually read it, and tells you in plain language what they see today. No email required to get your diagnosis.
Source: Ahrefs blog, AI SEO experiment on self-promotional content, July 6, 2026 (ahrefs.com), including the cited Glen Allsopp analysis of 750 ChatGPT prompts. We have read Ahrefs' framing and premise; the experiment's data is theirs and we have not reproduced it.