Ask AI for the Latest and 38% of Answers Cite a Page Over a Year Old

Someone asks an AI assistant for the latest rules on a home office tax deduction, or the newest model of a phone. The answer sounds current and confident. Underneath, one of the sources it cites was published two years ago, before the rule changed or the new model launched.

"Latest" questions are the ones where old information does the most damage. A price, a product version, a law or a deadline can change in months, and an answer built partly on an outdated page can be wrong in ways the reader has no reason to suspect. For publishers, it means an old page can keep feeding outdated facts into AI answers long after it stopped being true.

38% of AI answers to "latest" questions cite at least one page more than a year old, and 24.3% cite a page more than two years old. AI engines do reach for newer sources when asked: across all topics, 57.8% of answers cite a page over a year old. But asking for the latest does not rule old pages out.

We picked the 95 questions in our AI question sets that ask for the latest, newest, new or current version of something, or name 2025 or 2026. On September 23 and 24, 2026 we collected the answers ChatGPT, Gemini and Google AI Mode gave to them and dated each cited page by its published or last-updated date. 111 answers to 74 of those questions cited at least one dated page, 189 dated pages in all.

An answer counts as citing a year-old page when at least one of its dated sources was published or last updated more than a year before the answer. The same test with two years gives the two-year figure. For comparison, we ran the same test on 8,723 AI answers on all topics that cited at least one dated page.

That is 42 of 111 answers. Going further back, 24.3% of answers cite at least one page more than two years old, about one in four. Counting pages rather than answers, 34.4% of the 189 dated pages cited for these questions are more than a year old, so about one source in three predates the last year.

A reader who asks for the newest option can be reading a summary built partly on last year's information. The answer does not usually say which parts came from which source, so the reader has no way to see which facts are current. Old pages often rank and get cited because they have built up links and authority over time, which a newer, more accurate page has not yet earned.

AI engines do respond to the word "latest". Across 8,723 AI answers on all topics, 57.8% cite a page over a year old and 45% cite one over two years old. For "latest" questions those figures fall to 38% and 24.3%.

So the engines look for newer sources when the question asks for them, but they do not filter out older pages. More than a third of "latest" answers still include a source over a year old, which means a date in the question is a preference to AI engines, not a rule.

If you publish "best of 2026", "latest" or "current" content, show a clear, accurate published or updated date on the page and in its structured data. AI engines lean toward newer pages for these questions, and a visible date is how they tell your page is current.

Refresh year-labeled pages when the year changes rather than creating near-copies, and update the facts, not only the date. An old "latest" page that still gets cited spreads outdated information under your name. For pages that go out of date for good, redirect them to the current version so AI engines follow the redirect to the new page.

As an ai seo company, Suff Digital finds which of your pages AI engines cite and keeps time-sensitive pages current.

Embed this research

Cite this study