Self-ranking listicles are losing their AI Overview citations. Here is what earns them instead

What Lily Ray found about listicles, and what she did not#

Every marketer with a "best software for X" page has quietly done the same sum. Write the list, put your own brand at number one, let the machine read it. On 5 October that arithmetic took a knock. PPC Land reported that SEO analyst Lily Ray tracked 100 business-software queries and found Google's AI Overviews citing self-promotional listicles 38 per cent less often than they had in June.

The second number stings more. According to the same report, brands that ranked themselves first were left out of 83 per cent of the AI answers that used their pages. Read that again, slowly, as a fox reads a snapped twig: the page was good enough to quote and the brand behind it was not good enough to name.

Before anyone rewrites a content calendar on the strength of that, hold the caveats up to the light. The Bush Letter says plainly that Ray's sample of 100 queries is not large enough to treat the figures as definitive, and that no third party has audited the data. Ray's own essay of 4 October frames the finding as a signal inside a wider prediction about a coming Google update, and does not publish the full query list or method. This is one analyst's careful observation, not a study with a peer behind it.

The June baseline that makes the October number matter#

There is a prequel. In an earlier study published in June, Ray looked at 100 business-software "best" queries on three dates and found 184 self-promoting listicles from 146 brands. Google cited the pages, but left the brand out of the answer about 69 per cent of the time. The October reading of 83 per cent is therefore a worsening of a pattern she had already flagged, not a bolt from a clear sky.

Keep the units straight when you quote this. One row is an index, the other a percentage, and they come from separate samples. What the chart shows is direction, and the direction is the same on both lines: listicles that crown their own author are getting less out of the machine.

Google says nothing about listicles, and plenty about scaled content#

Here is where a sensible fox slows down. Nothing in Google's documentation names listicles. Google's AI features guidance says there are no additional requirements to appear in AI Overviews or AI Mode, and no special optimisations beyond being indexed and eligible to show a snippet. So the idea that Google has a rule against self-ranking lists is, today, Ray's inference and not Google's announcement.

What Google has put in writing is a sharper line on mass-produced pages. Its page on using generative AI content, updated on 1 October, warns that using generative AI to produce many pages without adding value may break its spam policy on scaled content abuse. The spam policies themselves define scaled content abuse as many pages generated to manipulate rankings rather than help users. And Search Engine Journal's weekly roundup reports the same guidance now asks publishers to fact-check AI-generated content manually before they publish it.

Put those side by side and Ray's reading looks plausible without being proven. A thousand near-identical "best of" pages, each with the author ranked first, is exactly the shape of content the scaled-content policy describes. If you are producing listicles by the dozen with a model and a template, you are standing nearest the fence.

There is a subtler point in Google's helpful-content guidance, which asks whether content is made primarily for people rather than to manipulate rankings. A page whose first verdict is that its own author wins is hard to describe as written for the reader. It is written for the scoreboard, and scoreboards are what ranking systems learn to discount.

The counter-evidence: llm citations are not one market#

Any honest read of this story has to include the data that points the other way. Ahrefs studied 26,283 ChatGPT source URLs across 750 prompts and found that self-promotional best-of lists made up 43.8 per cent of the page types ChatGPT cited. Ahrefs still treats them as correlated with visibility. That is a vendor study of a different engine, but it is a large, methodical one, and it does not say what Ray's sample says.

The reconciliation is simple, and slightly annoying. LLM citations are not one market. ChatGPT, Perplexity and Google's AI Overviews each choose sources differently, and a tactic that still works in one can be quietly discounted in another. The pattern we are watching in Google may not generalise, and the pattern in ChatGPT may not hold next quarter. The fox does not bet the den on a single hedgerow.

A second shift is easier to measure. In March, Search Engine Journal reported an Ahrefs study of 863,000 keywords and four million AI Overview URLs. Only 38 per cent of cited pages now rank in the top ten, down from 76 per cent in an earlier Ahrefs analysis. The earlier figure came from Ahrefs' 2025 analysis of 1.9 million citations. The two studies use different datasets, so treat the drop as direction rather than a precise decline. But the message is clean: ranking well and being cited have been drifting apart.

Read the ranked bars as terrain rather than a ladder. Roughly a third of citations came from pages that would never appear on page one. If the machine reaches that far down the undergrowth for its sources, then the quality of the page and the independence of its claims matter more than the old question of whether you hold position three.

What earns citation instead: earned media and brand mentions#

If self-ranking is the weak signal, what is the strong one? The best peer-reviewed hint we found is a controlled study posted to arXiv that compared how AI search engines choose sources. It reports a systematic and overwhelming bias towards earned media over brand-owned and social content. The experiments ran across ChatGPT, Perplexity and Gemini rather than AI Overviews specifically, so the finding is a direction, not a verdict on Google. But it matches the intuition behind Ray's work: a machine asked to recommend prefers a voice that is not selling.

That is the heart of earned media, and it is where brand mentions seo thinking stops being a vanity exercise. Mentions from publications, practitioners and customers carry a signal that no amount of self-description can copy. Our own earlier note on earned citation trails makes the same point from the experimentation side, and our piece on branded keywords in AI Overviews shows why what others say about you now sits at the top of your own brand search.

Three honest things to do with a self-ranking page#

• Rebuild it

  If the topic is core

  Effort

  High

  Risk

  Low

  Payoff

  A page worth quoting

  Best for

  • State your criteria first.
  • Name rivals' strengths and a case where you are not the best fit.

• Label it

  If the list is genuinely useful

  Effort

  Low

  Risk

  Medium

  Payoff

  Honesty on the page

  Best for

  • Say it is written by a vendor.
  • Rank on stated criteria, not on authorship.

• Retire it

  If it is a template

  Effort

  Low

  Risk

  Low

  Payoff

  A cleaner archive

  Best for

  • Fold thin pages into one strong guide.
  • Redirect, do not delete.

Notice what all three share. None of them tells you to stop writing comparison content. Buyers still want comparisons, and a comparison written with its method on show is useful to a human and to a machine alike. The change is who holds the scorecard. When the author is also the winner, a reader has to take it on trust. When the criteria are stated and the rivals get fair treatment, they can check.

We cover this ground for clients through our content marketing service and our SEO and GEO work. The recurring lesson is dull and dependable: pages that name a source, a number and a date are the ones that survive each new fashion in search. For the audit side, see our note on what a 98,020-claim audit found about AI Overview citations.

Five moves for the quarter, before listicles become a liability#

You do not need to panic, and you should not wait for a Google announcement that may never come. The moves below are cheap, reversible and good content in their own right.

That last step is the one most teams skip. A citation is a link in a footnote. A recommendation is your name in the sentence. The gap between them is the gap Ray's 83 per cent describes, and closing it is the quiet quarry of the next year.

• 38%

  fewer citations of self-promotional lists, in one unaudited sample of 100 queries

• 83%

  of AI answers using a self-ranking page left the brand out, as reported

• 43.8%

  of ChatGPT-cited page types were self-promotional best lists in Ahrefs' study

• 38%

  of AI Overview citations came from top-ten pages in Ahrefs' March 2026 analysis

Two of those figures share a number and describe unrelated things. That is the sort of coincidence that embarrasses people on slides. Label every statistic with its source, its sample and its date, and the fox stays out of the trap.

Frequently asked questions#