We read the year on 32,719 cited best-of pages
The sample is every answer Proofsource collected from October 4 to 25, 2026, for an English buyer question (one that doesn't name the brand) on ChatGPT, Perplexity, Claude and Google AI Overviews: 18,258 answers for 117 brands in 78 workspaces.
We counted a cited page as a best-of list when its address carries "best" or "top". That gave 32,719 citations in 10,018 answers, pointing at 14,722 different pages. Of those citations, 6,703 (20.5%) had a year in the address, and 5,888 of those said 2026.
Most cited lists put the year in the title, not the address
An address is a poor place to look, because most publishers keep the address fixed and change the title each year. So on October 25 we opened 200 cited list pages picked at random and read the title. 166 returned a readable page; the rest were blocked or gone.
Of those 166, 103 had 2026 in the title (62.0%, 95% interval 54.5% to 69.1%). Three had 2027, 11 (6.6%) had an older year, and 49 (29.5%) had no year at all.
So the year is what most cited lists look like. It isn't an edge, and it isn't required either: three in ten lists were cited without one.
Last year's lists still got cited, most often by Claude
Share of each engine's best-of citations whose address carried 2025 or an earlier year:
- Claude: 3.2% (455 of 14,319).
- Google AI Overviews: 2.8% (150 of 5,327).
- Perplexity: 1.5% (153 of 10,271).
- ChatGPT: 0.8% (22 of 2,802).
The year in the engine's search did more than the year on the page
ChatGPT, Perplexity and Claude report the searches they run. ChatGPT put 2026 into its searches on 59.0% of the answers where it searched (2,707 of 4,589), Perplexity on 41.8% (1,699 of 4,069) and Claude on 24.9% (1,094 of 4,388). That is the same order as the stale shares above.
Claude gave us the cleanest comparison. On 300 questions it searched with the year on some days and without it on others. With the year (439 answers, 2,561 list citations), 25.3% of cited lists had 2026 in the address and 1.2% had an older year. Without it (517 answers, 2,180 citations), the figures were 10.2% and 6.9%.
ChatGPT, on the 75 questions where it did both, showed no such gap: 19.4% against 19.6% current-year lists. It seems to prefer recent pages whatever it types.
What to do with your own lists and comparison pages
- Put the current year in the title if the page is genuinely current. You'll look like most cited lists. Don't expect it to lift you above them.
- Change the year only when you change the page. Google's guidance on helpful content lists changing a page's date to make it seem fresh, without substantial changes, as a warning sign. Update the picks, the prices and the tested date first.
- Don't stamp next year early. Google asks that page dates be accurate and not in the future, and a 2027 list in October 2026 tells a buyer the same thing.
- Find the old lists in your category that engines still cite, especially on Claude and AI Overviews. A 2025 list that leaves you out is still being read.
How Proofsource handles it
Proofsource shows the searches each engine ran next to every answer, where the engine reports them, alongside the pages it cited and the brands it named. You can see whether an engine looked for this year's list on your questions and which list it ended up reading.
Where this could be wrong
List pages come from words in the address, so a list without "best" or "top" in it is missed and some pages with those words aren't lists.
We read titles on October 25, not on the day each page was cited. A page retitled since then would count as current here. The title sample is 166 pages, so the 62% has the wide interval shown.
The Claude comparison isn't an experiment. On the days it added the year, its searches may have differed in other ways too, and answers to the same question aren't independent, so the true uncertainty is larger than the counts suggest.
The sample is 22 days, English questions asked from the US, and the categories our users chose to track. We can't see how many old lists existed to be cited, so a low stale share may partly reflect how few remain.
