Who Wins When You Rank Yourself #1? What 268,493 AI Answers Reveal About Self-Promotional Listicles
The most requested AEO tactic of the year is publishing a "best in category" list with yourself on top. I traced every citation in 268,493 AI answers to see who actually gets recommended, and what happened to the brands that ran the tactic hardest.
The brands in this study are real, tracked continuously for eleven months, and labeled A through I. I anonymized them because the point is the pattern, not the pillory. Every quoted AI answer is verbatim and dated, with brand names swapped for their labels. Nothing else has been altered.
Every week someone asks me a version of the same question: should we publish a "best tools in our category" article on our own site, with ourselves ranked #1? The reasoning always sounds solid. AI engines cite listicles constantly, the page costs almost nothing to produce, and half the AEO playbooks on LinkedIn recommend it.
For a long time my answer was an instinct, not a number. Then I realized I was sitting on the data to settle it. We track one consumer category, AI English-learning apps, where a dozen vendors ran this exact play at every possible intensity, from zero pages to 241 of them. And because our visibility platform stores every AI answer it collects along with its full citations, I could trace what actually happens after one of these pages gets cited.
of citing Google AI answers name the publisher
of citing Google AI answers name a rival
of citing ChatGPT answers name a rival
Aloha first-party data. Share of answers citing a vendor's own product listicle, non-branded queries only.
The short version: the pages do get cited, but the brands that wrote them rarely get recommended, and the visibility they built this way collapsed within months.
- Citation is not recommendation. On Google, an answer that cites your own "best of" listicle is 6.4× more likely to name a competitor than to name you.
- No engine is a loophole. ChatGPT names the publisher most often, in 67% of citing answers, but it names a rival in 75% of them. Rivals win or tie on all three engines.
- The tactic decayed during the study itself. ChatGPT purged the heaviest publisher's listicles from its citations, a 98% collapse from the December peak across its 241 self-promotional pages, while Google kept citing the same pages but named the brand 69% less.
- The damage reached classic SEO. The heaviest publisher's top-10 presence on tracked queries fell 68%, with the drop timed to Google's official March and May 2026 core updates, while the strongest non-publisher grew 38%.
- The category's biggest winner published none of these pages, and got recommended constantly from claims sourced on five competitors' listicles.
The Cast
Three brands carry most of the story:
The listicle factory
241 self-promotional listicles cited in AI answers, out of 447 "best/top" pages overall, the heaviest publisher of the genre I have ever tracked
The free-rider
The category leader. Publishes essentially no listicles and gets recommended constantly anyway
The control
Publishes none. Grew both AI citations and Google rankings all year on editorial content alone
Citation Is Not Recommendation
The tactic rests on one assumption: if the engines cite your list, they will recommend you. Measured at claim level, that assumption fails on every engine.
ChatGPT looks like the exception until you read the second column. It names the publisher two-thirds of the time, which sounds like the tactic working, but it names a rival three-quarters of the time. Whoever writes the list, the whole list gets recommended.
The reason becomes obvious once you think about what these pages are. A product listicle is a structured directory of the category, with every competitor described, compared, and ranked, hosted on your domain. That is exactly the raw material an answer engine wants. The engine harvests the directory and discounts precisely one claim on the page: your #1 self-placement, the only claim written by an interested party about itself.
At claim level the arithmetic is lopsided:
Brand A paid for the content. Its competitors collected the recommendations, eight to one.
Then the Engines Corrected, in Two Different Ways
For a while the tactic genuinely worked, at least on ChatGPT. At the December 2025 peak, Brand A's own listicles were cited in over 40% of the tracked answers, and ChatGPT named Brand A in most of them. What happened next is the part every AEO playbook is missing.
ChatGPT and Gemini ran the same correction: they stopped citing the pages. ChatGPT's citations of Brand A's listicles collapsed by 98% between December and July, with the drop concentrated in March and April of 2026. The mentions built on those citations fell alongside and stabilized at less than half of peak, resting on whatever earned sources Brand A had left. The timing matters, because this landed right after the December-to-February window in which Peec's 232K-citation study reported seeing no algorithmic correction yet. The correction simply had not happened when they looked. Then it did.
Google corrected differently, and in some ways more brutally. It never stopped citing the listicles. Citations of Brand A's pages grew 17× across the year on our tracked queries, while answers naming Brand A fell from 45 per 1,000 to 14. Google kept the directory and dropped the author.
Two mechanisms, one outcome: visibility built on self-promotional listicles evaporated, either because the engine stopped reading the pages or because it stopped believing them.
The Biggest Winner Published Nothing
Brand G leads the category in AI recommendations while running essentially none of these pages. It does not need to, because its competitors' pages do the work for it.
Conquest pages, meaning content built to capture a competitor's brand searches, backfire the same way but harder. Brand A published "alternatives" and "review" pages targeting Brand H's brand queries, and Google turned them into a marketing channel for Brand H itself:
Query: "[Brand H] review" · Dec 22, 2025 · Google AI Mode
"[Brand H] reviews highlight it as a strong, affordable AI tool for practicing spoken English with instant feedback on pronunciation, grammar, and fluency, praised for its realistic conversations, gamification, and 24/7 availability."
Cited source: Brand A's own review page targeting Brand H's brand queries. Of the 26 branded claims those conquest pages fed, 20 named Brand H.
You cannot write about a rival for the engines without teaching the engines about your rival.
Publishers and Non-Publishers Split, Everywhere I Measured
Group every brand in the category by a single variable, whether it publishes self-promotional listicles, and the trajectories separate cleanly. Note that intensity is not just page counts: two brands farmed hundreds of citations from a couple of programmatic pages each.
By June the publishing group sat at roughly half of its October–December baseline while the non-publishers held around 80% of theirs. Brand by brand, the gradient follows publishing intensity:
Ten months of monthly data supports correlation rather than proof, and the whole category softened over the period. But the same split appears in classic Google rankings, on the same queries, and the declines line up with Google's officially documented core updates, shaded in the charts below.
The numbers behind those lines deserve to be spelled out. For all that publishing, Brand A reached Google's top 3 exactly 7 times in roughly 27,000 tracked search results, and 79% of its ranking appearances sat at position 21 or worse. Its top-10 presence fell 68% across the year, dropping through the March core update and never recovering. Its strongest rankings on these queries are the conquest pages riding Brand H's name. Meanwhile Brand H, which publishes none of the genre, grew its top-10 presence by 38% and finished the year as the panel's top-ranked brand. One heavy publisher has since started quietly deleting its listicle pages, which is its own kind of verdict.
This is the same direction Lily Ray has documented in other verticals, where brands lost 29% to 49% of organic traffic after publishing self-promotional listicles at scale, a pattern Search Engine Land has covered as a likely core-update target.
The Receipts
Claim-level attribution means I can show the exact moment a publisher's page feeds a rival.
Query: "learn english with ai" · Jun 4, 2026 · Google AI Mode
"Brand G: The gold standard for perfecting your accent and pronunciation. It listens to your speech and rates your phonetic accuracy."
Cited source: Brand A's own "top free apps to improve English speaking" listicle. Brand A's page supplies the language crowning its competitor the gold standard.
Query: "free english learning app" · Jul 26, 2026 · Google AI Mode
"The best free English learning apps are Duolingo for daily basics, HelloTalk for human conversation, and Brand I for AI speaking practice."
Cited source: Brand A's own "top free AI tools" listicle, feeding an answer that recommends three other products and not Brand A.
Query: "Which AI English speaking apps have the most accurate speech recognition?" · Jul 23, 2026 · ChatGPT
"3. Brand H: emphasizes free-flowing conversations and provides real-time corrections on pronunciation, grammar, fluency, and vocabulary."
Footnote credit: Brand F's own "best pronunciation apps" listicle, while the wording traces to an independent list cited in the same answer. ChatGPT blends its sources and stamps credit loosely. Either way, the self-promoter is not in the claim.
Pro tipBefore approving budget for this tactic, ask one question: when our listicle gets cited, who gets named in the answer? If the answer is not backed by data from your own category, it is a guess. The engines store nothing for you, so the only way to know is to be tracking the answers before you need them.
What Works Instead
Self-promotional listicles are an asymmetric bet. The upside, being named from your own page, is capped and visibly decaying on every engine. The downside is not: you are hosting a structured competitor directory on your own domain while exposing yourself to a pattern Google's core updates keep hitting. The winners in this category took two routes around the bet:
-
Editorial content in your own lane. In the same corpus, vendors' editorial content (guides, teaching material, lists of things they do not compete with) earned citations at a similar scale to the self-promo genre, with 4% competitor leakage instead of 59%. Same citation power, no competitor directory.
-
Earned placements in the lists engines already trust. The most-cited sources in this category are a Reddit thread and independent reviewers' lists, not any vendor's page. Your citation data tells you exactly which third-party lists matter: rank them by how often the engines cite them on your queries, then earn your way onto the ones you are missing from.
Why This Matters for Our Clients
I could not have answered this question a year ago. Not because the analysis is hard, but because the data has to already exist when the question arrives. AI answers are not archived anywhere you can retrieve later. If you were not capturing them in January, January is gone.
That is the real argument for tracking AI visibility properly rather than spot-checking ChatGPT when a client asks. Our clients did not need to run this study to benefit from it, but the study only exists because we have been storing every answer, every citation, and every source for every tracked prompt since the day we onboarded them. When the next tactic comes along, and it will, the same corpus answers that question too.
AI search tactics now rise and decay within quarters. This one went from working to purged in about four months, and the next one will move just as fast. The brands that navigate that churn will not be the ones with the cleverest tactic of the month. They will be the ones measuring what the engines actually do.
Want to see this running on your brand?
Book a demo and see how our systems turn into compounding organic growth.
