Ranking your own product first in a "best [category]" listicle is one of the most common AEO tactics going.
New research from Scrunch Labs put a real number on what it actually does: when an AI answer cites that listicle, the author brand's odds of being recommended rise from about 4% to roughly 7%. Real, and small.
The lift survives a proper check, but it's modest
A 4-to-7% recommendation rate is easy to dismiss until you see how it was tested. Scrunch analysed roughly 10,000 AI-cited URLs gathered between 17 May and 30 June 2026, then isolated the causal effect with a capped sample of 818 self-ranking pages against 1,033 neutral comparison pages as control - holding the prompt, brand and AI assistant fixed, and comparing answers that cited a brand's self-ranking listicle against answers to the same question that didn't.
To rule out brands that were simply already trending upward in AI search, they ran a falsification test: checking whether a citation that hadn't happened yet could already predict today's result. If it can, the "lift" is a pre-existing trend wearing a citation's clothes. Here, the placebo effect came out to roughly a quarter to a third of the real one, so most of the lift is genuinely down to the citation, not a brand that was rising anyway.
Mentions move far more than recommendations do
A citation and a recommendation are different outcomes, and this study shows exactly how differently they respond to the same tactic. Recommendation rate moved from 4% to 7% when a listicle was cited. Brand mention rate moved from about 19% to 39%, four to six times the size of the recommendation lift.
Most of a self-ranking listicle's value shows up as your name appearing in the answer, not as the AI actively picking you as the winner. Scrunch's own data shows that in 63.9% of the answers that cite one, AI treats the listicle as raw material for a balanced overview and doesn't adopt its number-one ranking outright.
Which platform cites you decides how much that's worth
The same citation is worth more on some AI assistants than others. Looking at how often a self-ranking listicle's author brand gets recommended, across all the answers that cite it, on each platform: Claude 15.6%, Gemini 14.8%, Google AI Overviews 8.9%, ChatGPT 8.6%, Google AI Mode 7.7%, Perplexity 6.3%, Microsoft Copilot 4.8%.
Gemini and Claude are the assistants most willing to pick a winner outright. Copilot and Perplexity mostly summarise what they've read and leave the choosing to the reader. Judge a listicle's performance against the platform it's actually appearing on, not a blended average across all of them.
Naming a rival is usually safe. One scenario isn't
Every listicle that names competitors risks handing them a citation too, and 43.4% of answers that cite a self-ranking listicle also cite a rival's site. Scrunch tested whether that actually helps the named rival with the same falsification approach. For most brands, it doesn't: the raw rival lift (2.8 to 3.4 percentage points) mostly failed the falsification test, so it looks like a pre-existing trend and not the listicle's doing.
The one exception is a lower-visibility brand naming other lower-visibility rivals, where a +7 point rival lift held up against a -1.9 point placebo. Scrunch flags this themselves as one of four scenario cuts without correction for testing multiple comparisons, and a narrower slice of the same data put the figure at +11 points with too few observations to be confident in. Treat it as a real, unsettled risk, not a rule: a smaller brand naming smaller rivals is the one case where that's worth thinking about before publishing.
What this means for the listicle sitting in your content plan
Two practical findings worth keeping. Pages listing four to ten brands outperformed pages listing only two or three, or cramming in eleven to fifteen, so mid-sized lists get the author brand recommended more often than short or overlong ones.
And a self-ranking listicle only works as a citation if the model can find and read it at all, which is a technical and editorial question, not a ranking one. None of this makes a listicle a strategy on its own. It's one page competing for a modest lift, in a system where mentions carry most of the value and recommendations still go, most of the time, to somebody else.
A few common follow-up questions
Should we stop publishing self-ranking listicles?
No. The lift is real and the mention effect in particular is sizeable. The finding is about expectations, not effort: a listicle nudges recommendation odds, it doesn't decide them.
Is it risky to name our biggest competitor in a listicle?
Not based on this data. The rival-lift risk that held up under testing was specifically a smaller brand naming other smaller rivals. Naming a high-visibility competitor didn't produce a reliable rival lift in either direction.
How many brands should a listicle include?
Four to ten performed best in this study. Shorter lists and lists past fifteen both underperformed that range for getting the author brand recommended.