Google Demoted Listicles in January. The Ones Brands Wrote About Themselves Gained 38%.

If you publish a “best tools for X” page that happens to include your own product, Google’s unannounced January update probably helped you rather than hurt you. And the single thing that best predicts whether that page ranks is the date printed on it.

Both of those come out of two studies Kevin Indig published in the last week, built on 60,000 US-English queries across 15 verticals and 5.32 million organic result rows, collected in a 47-hour window in August using SE Ranking data.

The headline everyone repeated in January was that Google had come for listicles. The data says something more specific and more useful.

What the data says

Listicles did lose ground at the top. Indig’s August 24 analysis puts listicle presence at position one falling from a range of 18.9% to 20.2% in January down to 15.1% to 15.6% in August. Across the top three the decline runs 3.9 to 6.0 percentage points.

They did not disappear. At least one listicle still appears in the top 10 for 55.1% of queries and in the top three for 32.3%. Ask Google for a set of options and it shows 4.5 times more of them. Indig’s own summary is that Google “has systematically and carefully weeded out certain listicles,” which is a narrower claim than a format penalty.

Now the part that reverses the story most people took away.

The vendor pages won

Self-serving listicles, the ones a company writes about a category its own product competes in, did better than everything else. Per PPC Land’s writeup of the same dataset: 62% of vendor-authored listicles gained estimated organic traffic since January, and the median page gained 38%.

Publishers still beat brands in direct competition, winning 54.0% of head-to-head matchups on the same results page. But the direction of travel since January favored the companies with something to sell.

For B2B software this matters more than average. Listicles appear in 62.5% of B2B software results, and brand or vendor listicles show up in 46.2% of them, against 3.3% in food. Writing your own category page is close to table stakes in B2B and almost nonexistent in consumer verticals.

Why the obvious explanation is wrong

The tempting read is that Google got worse at spotting self-interest. Ibrahim Furkan Ozcelik, who works on generative engine optimization and AI search, argued the opposite back in May, before this data existed:

“the signal Google is downweighting is unearned self-elevation, and the format is the scapegoat.”

His framing is that the demotion lands on pages firing several bad signals at once: ranking yourself first, AI-generated body text, swapping the year in the title to fake freshness, no disclosed methodology, programmatic scale, and schema manipulation. “Pages that fire five at once are the ones losing visibility.”

He adds a standard worth writing down: “High quality has acquired a methodological dimension: does the author explain how they arrived at their conclusion?”

Put his May thesis next to Indig’s August numbers and they agree. A vendor listicle built on real testing with its method shown is not what got hit. The version assembled at volume with a swapped date is.

What predicts a top-three spot

Indig’s August 31 follow-up crawled 3,000 currently ranking listicles, 200 per vertical, and manually reviewed 83 pages. The findings are unusually blunt about what doesn’t work.

Freshness is the strongest signal. A date from the previous two years appeared on 66.6% of pages in the top three, against 57.3% of pages in positions 8 to 10. Old or unreadable dates carried 56% lower odds of ranking in the top three.

Counted lists beat editorial roundups. Where both formats competed on the same results page, 56.8% of matchups went to the counted version. Counted listicles averaged position 3.77 against 4.59, and hit the top three 55.9% of the time against 41.4%.

Bigger counts did better. Titles listing 51 to 100 items averaged position 4.11. Titles with 2 to 5 averaged 5.85.

More headings hurt. A one-standard-deviation increase in heading count came with 29% lower top-three odds.

Several things people spend real time on were noise. Schema markup and author bios showed no predictive power. Word count and image count correlated (top-three pages carried 21.5% more words and 38.9% more images at the median) but the effect wasn’t consistent across models. And putting your own product first showed only 4% stronger growth, with a confidence interval running from 51% lower to 120% higher, which is another way of saying nobody can tell.

The AI-search half

None of this happens in a search results page that only contains links anymore. On 92% of the pages carrying a top-10 listicle, Reddit or YouTube also appeared somewhere. Where Reddit showed up at all, it was in the top three 70.7% of the time.

AI Overviews covered 83.7% of these queries on average, and 93.4% in B2B. So the ranking you win is being summarized above itself on nearly every B2B query, which is the same squeeze we covered when Google’s own redirect change raised the cost of rank data and when answer engines started quoting people rather than pages.

A top-three listicle is now an input to a summary as often as it’s a destination.

What to do about it

Go find the dates on your category pages. Highest-return item here, and most teams have never audited it. A visible, recent, machine-readable date on a page that was genuinely updated. Not a year swapped in a title, which is one of the six signals Ozcelik names as the problem.

Count your list and make it longer. Counted beats editorial, and larger counts ranked better. That’s a title and structure change, not a rewrite.

Cut headings rather than adding them. More headings tracked with worse odds. The instinct to break everything into subsections is working against you here.

Stop spending time on schema and author bios for these pages. Both came out as noise. Spend it on the methodology paragraph instead, which is the thing Ozcelik argues separates a surviving vendor listicle from a demoted one.

Then write down how you picked. If your page ranks your own product, say what you compared, on what basis, and when. The data says the vendor page isn’t the problem. The vendor page that can’t show its work is.