Updating Your Best Pages Beats Publishing New Ones

Date: 2026-09-24

If the pages on your website haven’t been touched in a couple of years, AI tools are quietly less likely to use them when someone asks for a recommendation in your line of work.

That sounds like an argument for publishing more. It isn’t. The research points somewhere more useful and a good deal cheaper: the pages you already have, kept up to date, do more for you than a stream of new ones. Most businesses have this backwards, which is why it’s worth setting out what the evidence actually shows.

“Publish fresh content” has been the standing advice for two years now. What nobody attached to it was a number. Fresh compared to what — last month, last year, three years ago? Without that, it’s a vague instruction rather than something you can act on. There are now numbers, from three large independent studies, and they tell a more interesting story than the advice does.

AI tools draw on noticeably newer material than Google does

Researchers at the University of Toronto measured how old the pages were that each system actually drew on when answering buying-type questions. They ran the same set of questions through Google and through three AI assistants, then checked the publication date of every page each one used.

The typical age of the material they leaned on:

Read across the bottom row first. In a slower-moving market like cars, the typical page Google surfaced was nearly sixteen months old. The AI assistants were working from material five months old. Across both markets, AI answers drew on content between roughly a third and 70% newer than Google’s.

The gap matters because of what it implies about the competition. Older pages aren’t banned — but they’re up against a pool of material that skews much younger than the one you’re used to competing in on Google. And the slower your sector moves, the wider that gap gets. If your competitors are sitting on pages they last touched in 2023, that’s an opening rather than a threat.

Most content AI draws on is still years old

Here’s the finding that stops this becoming a panic about publishing schedules.

Ahrefs looked at nearly 17 million links cited across seven different AI and search platforms. On average, the pages AI tools drew on were 2.9 years old. Google’s ordinary results averaged 3.9 years. So yes — AI leans newer, by about a year. But the typical page it uses is still the best part of three years past its publication date.

Both findings are true, and the difference between them is worth understanding rather than averaging away. The Toronto study measured the midpoint age in two fast-moving consumer markets where publishers refresh constantly. The Ahrefs figure is an average across every topic, and a handful of very old, very authoritative pages pull an average up in a way they don’t pull a midpoint up.

What survives both readings is the same: recency helps, and it helps more than it used to — but a three-year-old page that genuinely answers the question is still being used every day. Age alone doesn’t disqualify you.

Updating an old page counts as fresh

This is the finding that changes what you should actually do on Monday morning.

Seer Interactive tracked 7,683 pages that AI tools drew on across roughly 47,000 mentions between March and June 2026. Three in four of those pages had been updated within the previous year. Judged by when they were last updated, nearly three-quarters looked current.

Then they looked at the same pages by original publication date. On that measure, fewer than half were recent. More than a quarter of the pages that qualified as “fresh” had first been published more than two years earlier.

In other words: most of the fresh content AI tools are using is not new content. It’s old content that somebody maintained.

That is a much easier instruction to follow than “publish more”. You don’t need a content team. You need to know which of your existing pages matter — your main service pages, your pricing or process explanations, the questions customers actually ask you — and to keep them genuinely current.

The word “genuinely” is doing real work there. Changing a date stamp without changing anything else is not an update, and the author of the Ahrefs study warns explicitly against that. What counts is substance: a price that’s right, a service you no longer offer removed, a new question answered, a case or example from this year rather than three years ago.

How fresh is fresh enough?

There’s no single threshold, and anyone who gives you one is guessing. But the evidence supports a reasonable rule of thumb.

Within the last year is the broad working window. Of the pages AI tools drew on in the Seer study, Google’s Gemini took 78% from the previous twelve months, ChatGPT 73%, and Perplexity 65%. Separately, Muck Rack’s analysis of more than 25 million links across ChatGPT, Claude and Gemini found that more than half of the news coverage cited was published within the past year, with the volume dropping off sharply after the first six months.

So: for the handful of pages that genuinely matter to your business, once a year is the minimum, and twice is better. For a fast-moving sector, or for a page carrying prices, availability or anything else that dates quickly, work to a shorter cycle. That’s it. It doesn’t need a dashboard.

What this means for you

If you have been feeling guilty about not publishing enough, you can stop. The more useful question is whether the pages you already have are still accurate, and when you last checked.

Pick the five or six pages that would matter most if a prospective customer read them tomorrow. Look at when each was last meaningfully changed — not restyled, changed. If any of them describes a price you’ve since altered, a service you’ve stopped offering, or a world that has moved on, that is a page currently working against you in two places at once: with the customers who read it, and with the AI tools deciding whether to put you in front of the next one.

There’s a wider point behind this. Whether AI tools recommend your business isn’t a mystery and it isn’t luck. It can be measured — how often you’re named when someone asks for a business like yours, how that compares with your competitors, and what in your wider online reputation is driving the answer either way. Content age is one input among several, and on its own it will never be the whole story.

Finding out where you currently stand is what we do.

Sources: Mahe Chen, Xiaoxuan Wang, Kaiwen Chen and Nick Koudas (University of Toronto), “Navigating the Shift: A Comparative Analysis of Web Search and Generative AI Response Generation”, EDBT/ICDT 2026 Workshops; Ryan Law, “Do AI Assistants Prefer to Cite Fresh Content?”, Ahrefs, July 2025; Seer Interactive, “Content Recency’s Impact on AI Visibility in 2026”, July 2026; Muck Rack Generative Pulse, “What Is AI Reading?”, May 2026.