Does Content Freshness Matter for AI Search? The Evidence and the Folklore

Old pages get cited every day

Start with the number that deflates the hype. In an Ahrefs study of about 17 million citations (July 2025), the URLs cited by AI assistants averaged 1,064 days old. That is almost three years. The URLs in Google's organic results averaged 1,432 days. So AI answers do lean newer, by about 25.7%, but they are hardly a news ticker. A careful page from 2023 is still a citable page.

The same study found the preference is uneven. ChatGPT showed the strongest pull toward newer content. Google's AI Overviews showed almost none, which fits Google's description of AI Overviews as rooted in its core ranking systems rather than a separate engine with its own taste for new pages. The useful conclusion is not 'publish more.' It is 'know which engine you are trying to reach, and what it is being asked.'

AI engines do not reward new dates. They punish wrong facts.

What each engine has said or shown about recency

Freshness is not one signal. Each engine handles it differently, and only some have said anything on the record. This is everything we could verify, as of October 2026.

Notice how little of this is a ranking promise. Microsoft says current content improves the chance of a citation in Copilot. Anthropic exposes a page's age to the model in Claude, which tells you the date is visible but not how much it weighs. Anyone claiming a precise freshness formula for these engines is guessing.

Freshness folklore, sorted by how true it is

Most freshness advice is half-remembered SEO lore with 'AI' pasted on top. Some of it is right. Some of it wastes a week. A little of it can hurt.

The middle ground is plausible but unproven: that a visible 'last updated' date helps an engine choose between two similar pages. It would be consistent with Claude seeing page age and ChatGPT's measured lean toward newer sources. Nobody has isolated it. We treat it as good practice, not a lever, and the content AI can quote matters far more than the date above it.

Stale facts are the real freshness problem

Here is where freshness bites a business. A model reads a services page that still lists the location you closed, or a pricing page from before the last increase, or holiday hours from two years ago. It repeats them with total confidence. The customer drives to the wrong address and blames you, not the machine.

This is a different problem from a page being old. A ten-year-old page about how a water heater works is fine if water heaters have not changed. A six-month-old page with last season's prices is a liability. Freshness work should be organized by facts, not by publish dates: which pages state things that change, and who checks them when they do.

• Hours, including holiday hours. The most common wrong answer, and the one that strands customers in a parking lot.
• Prices and price ranges. Engines quote them as if they were today's numbers.
• Services and service areas. A dropped service that still has a page will keep getting recommended.
• Staff, licenses and certifications. Especially in regulated trades, where an expired license number is worse than none.
• Statistics and year-stamped claims. 'Best of 2023' in a title tells every reader, human or machine, that the page stopped in 2023.

If an engine is already quoting something outdated, the fix guide for AI citing old information walks through correcting the source pages and listings, and how long the correction takes to show up.

How to date a page honestly

Dates are worth getting right because engines and people can both see them. Google publishes specific guidance on publication dates, and following it costs nothing.

On a site we build, these are wired into the code rather than remembered by hand: the modified date changes when the content does, the sitemap follows, and the structured data carries the same value as the page. That is less a feature than a refusal to let three systems disagree about one date.

A quarterly freshness routine for owners

You do not need a content calendar to stay fresh in the way that matters. You need a list of pages that state changeable facts and a habit of checking them. This takes an afternoon a quarter for most local businesses.

1. 1

   List the pages that state facts

   Home, contact, pricing, each service page, locations, team and any FAQ. Skip pure explainers that do not go out of date.

2. 2

   Check each fact against reality

   Hours, holiday hours, prices, services, areas served, license numbers, phone numbers. Mark anything that changed.

3. 3

   Update the page, then the date

   Edit the content first. Change the visible 'last updated' date and dateModified only when the change is real.

4. 4

   Update the listings the same day

   Google Business Profile, Apple Business, Bing Places and your industry directories. Some engines read the listing, not the website: Gemini, for one, answers questions about places from Google Maps data.

5. 5

   Tell the engines

   Submit changed URLs through IndexNow for Bing and Copilot, and keep the sitemap's lastmod accurate for Google.

6. 6

   Ask the engines again

   Run your customers' questions through the AI visibility check and confirm the answers moved. If they did not, the old fact is living somewhere else.

When a refresh is worth paying for

Most freshness work is cheap and owners can do it alone. It is worth outside help when the stale facts live in places you cannot easily reach (dozens of directories, old press pages, an abandoned second website), or when an engine keeps repeating an old fact after you have fixed your own pages. That usually means a source you have not found yet.

DataCram folds this into its GEO programs: tracking what the engines say over time, correcting listings so they agree, and rewriting the pages that state facts so they are current and quotable. Programs start at $625 a month, most local businesses land between $625 and $2,250, and the tier is set after a free GEO audit. The GEO pricing page explains what moves the number. Before any of that, the GEO section front page maps the rest of the work, and prompt tracking explains how to tell whether a refresh changed anything.