Does ChatGPT prefer newer content, and how old is too old?

Most founder-led firms we talk to have a website with a hundred or so pages, half of them written years ago by someone who has since left. The worry is usually one of two things. Either "our content is old, so ChatGPT must be ignoring it", or "we cannot keep rewriting everything, so what is the point". Both worries are reasonable. Both are slightly wrong.

AI assistants do lean towards recent pages, and ChatGPT leans hardest. Ahrefs studied about 17 million citations across seven AI platforms in July 2025 and found that the pages assistants cite were on average 25.7% fresher than the pages in Google's ordinary results. ChatGPT's citations were the newest, at an average of 958 days old. A year later, Seer Interactive found that 75% of the pages AI assistants cited had been updated in the previous 12 months.

Fresh does not mean published last week. Even ChatGPT's average cited page is more than two and a half years old; what assistants favour is a page that is still looked after. This guide explains what "current" means to an assistant, how often to review which pages, how to show dates, and how to keep a library current without rewriting it.

#Do AI assistants prefer newer content?

Yes, on average, and ChatGPT more than the others. They favour pages that are current, which is different from pages that are merely new.

An assistant uses your content in two ways. The first is its memory: what the model learned from a snapshot of the web taken months or years earlier. The second is live search: ChatGPT search, Perplexity, Copilot, Gemini and Google's AI features look pages up when a buyer asks, and judge your page as it stands today. Google says its AI features ground their answers in its search index partly to improve freshness (Google Search Central, updated July 2026).

The Ahrefs study shows how much the platforms differ. Average age of the pages each one cited, in days since publication:

Source: Ahrefs, July 2025, about 17 million citations.

Two 2026 studies point the same way. Seer found that 88% of cited pages had been updated within two years, and that among pages showing both dates, 72% looked fresh by their last update while only 42% had first been published in the past year. Seer covered four consumer and finance industries, not B2B services, so treat it as a direction rather than a rule. An AirOps study of 353,799 pages reported by Search Engine Land in April 2026 found that pages published 30 to 89 days earlier were cited most by ChatGPT, pages newer than 30 days did worse, and pages more than two years old were cited less often.

Subject matters too. A three-year-old page on what an audit involves is still broadly right; a three-year-old page on audit exemption thresholds is wrong. In our experience recency counts for most where the answer moves.

#How old is too old for our content?

For the pages that bring in work, three months without a review is when they start to lose ground. For everything else, a page untouched for more than a year sits outside where most cited pages are, and one untouched for more than two years is in a small minority.

That replaces the old rule of thumb of "twelve to twenty-four months is fine", which was looser than the data. SE Ranking's study of 216,524 pages, reported by Search Engine Journal in November 2025, found that pages updated within three months averaged 6 ChatGPT citations, against 3.6 for outdated pages. It is a correlation, since firms that update often may be doing other things well, but it is the best case for a shorter cycle on your most important pages.

For example, a firm of chartered surveyors has a 2022 guide to dilapidations. The law has barely changed, but the example figures are old and two links are dead. That page needs an hour, not a rewrite: new figures, working links, a paragraph on anything new and an honest "Updated" line.

#Should we update old content or write new articles?

Update first, wherever a good page already exists. In Seer's data, only 42% of cited pages showing both dates had first been published in the past year, yet 72% had been updated in it. An old page also carries history a new one lacks: links from other sites, mentions and a track record at the same address.

Write new pages where a buyer question has no page that answers it. As a working balance, a firm with fifty or more solid pages should mostly keep the best ones current and fill gaps; a firm with fewer than thirty should mostly write new pages that answer buyer questions. For how much new publishing a firm like yours needs, see how often we need to publish for AI assistants to notice us.

Rebuilding existing pages is often where results show first. For ID-Pal, an identity verification firm, the work included "Five industry pages rebuilt: fund administration, wealth management, legal, accountancy and estate agency. Each opens with the question a compliance lead types, names the regulator, and answers inside the first ninety words." The result so far: "Where a page has been rebuilt, ID-Pal appears. Where it has not yet, it does not, which tells us exactly what to build next."

#Should we republish old articles with a new date?

No. Keep the original publication date, add an honest updated date, and say what changed.

The old argument was that a changed date on an unchanged page is easy to spot. The evidence says otherwise. A September 2025 study by Fang and others added artificial publication dates to passages and asked seven language models, including GPT-4o, to rank them. The "fresh" passages were consistently promoted, with individual items moving by as many as 95 places. Larger models reduced the effect but none removed it.

So a fake date might work on a machine for a while. The case against it is honesty: a date is a promise to the reader that someone checked the page. Google's guidance says dates "must describe the publication or update date of the page" (Google Search Central), and Ahrefs reports that Google's John Mueller has warned against changing publish dates without changing the content. There is a practical cost too: Google says it uses the lastmod date in your sitemap only if it is "consistently and verifiably" accurate (Google Search Central, July 2026). Redate pages that have not changed and you risk Google ignoring the dates on the pages that have.

A reviewed date is fine when the review was real: if someone read the page and confirmed it is still right, say "Reviewed September 2026". And never rewrite a strong, well-linked article at a new address just to get a new date. Update it in place and keep the URL.

#How do we show the dates correctly?

Show the dates in three places and make them agree: the visible line on the page, the Article markup and the sitemap. Google recommends a user-visible date labelled "Published" or "Last updated", and says the visible and structured dates should match.

Example: a guide first published on 14 March 2023 and updated on 15 September 2026.

Visible line under the title:
Published 14 March 2023. Updated 15 September 2026: new thresholds added, examples refreshed.

Article markup (JSON-LD in the page's HTML):
<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "What does a dilapidations claim involve?",
  "datePublished": "2023-03-14",
  "dateModified": "2026-09-15",
  "author": { "@type": "Person", "name": "Author name" }
}
</script>

Sitemap entry:
<url>
  <loc>https://www.example.co.uk/insights/dilapidations-guide</loc>
  <lastmod>2026-09-15</lastmod>
</url>

Change dateModified and lastmod only when the content, figures or links changed meaningfully, not for a typo or the copyright year, and check that your content system does not stamp a new date on every save. Write the update note in plain words, because that is the part a reader and a model can both see. For more on what markup does and does not do, see whether we need schema markup for AI assistants.

#Does ChatGPT read the dateModified in our schema markup?

OpenAI has not said that it does. Google reads datePublished and dateModified to decide which date to show, so keep the markup accurate, but do not rely on it to reach ChatGPT.

The date any model can be sure to see is the one written in the page text. The Fang study put dates into the passage text, and the models reacted to them. That is why the visible "Updated" line matters most, and why all three dates should agree.

#How do we get ChatGPT to see an updated page faster?

Make sure the crawlers can reach the page, then tell the indexes it has changed. Each assistant looks in a different place.

IndexNow lets a site tell Bing and several other engines the moment a page is added, updated or deleted; Google does not take part. OpenAI says a site must allow OAI-SearchBot to appear in ChatGPT search (OpenAI Help Center). If you use Cloudflare, check its bot settings: it has blocked AI crawlers by default on new domains since July 2025. Our guide to how AI crawlers read your website covers the robots.txt side.

#How to keep a content library current without rewriting it

Schedule reviews, keep them light and batch them by subject.

#Step 1: List what you have

Export every page with its publication date and its last update, and sort by age. For a firm with a hundred pages this takes an hour.

#Step 2: Mark the twenty that matter

Pick the twenty or so pages that answer the questions buyers ask or already bring in enquiries, and put them on the quarterly cycle. If you are not sure which pages matter, our guide on how to audit our existing content for AI assistants gives a way to score them.

#Step 3: Set a review calendar by subject

Group the remaining pages by subject. January: everything on subject A. April: subject B. July: subject C. October: subject D. This spreads the work and means every page is seen at least once a year.

#Step 4: Do a one-hour light update

Refresh figures and say which year each is from, add a recent example, fix dead links, add a paragraph on anything new, write a short update note and change the three dates together. Only rewrite when a page is wrong or a competitor's page is clearly better, and then aim for the standard in what an article that ChatGPT actually quotes looks like.

#Step 5: Retire what is not worth keeping

A page on a service you no longer offer should be redirected to the nearest current page or removed, because old, wrong pages can be the ones an assistant picks up when it describes you. Age alone is not a reason to delete: some of your oldest pages are your most linked and most cited.

#How do we tell whether it is working?

Keep a fixed list of 20 to 50 buyer questions and ask each several times a month in the main assistants, in a clean session, because answers vary from run to run. Record whether you are named and which page is cited, and count a win only when it repeats.

Where an old page is still being cited, leave it alone. Where a page has stopped being cited, review it next. Google Search Console has also shown performance reports for AI Overviews and AI Mode to all sites since 31 August 2026, which lets you see whether updated pages gain impressions there. Our guide on tracking whether AI tools are recommending your business sets this up as a routine.

#What to do this month

1. Export a list of every page on your site with its publication date and last update, and sort it by age.
2. Mark the twenty pages tied to revenue, put a quarterly review in the calendar and name the person who owns it.
3. Add a visible "Published" and "Updated" line to your article template, make the markup and sitemap dates match it, and stop your content system changing dates on trivial saves.
4. Give the ten oldest pages that still matter an hour each: refresh figures, add a recent example, fix links and write an update note.
5. Redirect or retire any page that is no longer true.
6. After each update, request indexing in Search Console and notify Bing through IndexNow.

#Frequently asked questions

#How often should I update blog posts for ChatGPT and AI Overviews?

Review the twenty or so pages that bring in work every quarter, and every other page at least once a year. SE Ranking found pages updated within three months were cited more by ChatGPT, and Seer found 75% of cited pages had been updated within a year. AI Overviews showed the weakest preference for new pages in the Ahrefs study.

#Does changing the date on a blog post help SEO or AI citations?

Changing the date without changing the page is misleading, and we do not recommend it. A 2025 study showed language models do promote passages with newer dates, but Google's John Mueller has warned against it, readers lose trust when a "new" page quotes old figures, and Google may stop trusting your sitemap dates. Change the date when you have changed the content, and say what changed.

#Should we show the original publication date or hide it?

Show it, alongside the latest update. Hiding the publication date does not make a page look newer; it makes it impossible to place in time, and Google recommends a visible, labelled date. A page maintained since 2022 is a better signal than one that appeared last month with no history.

#Do Google and the AI assistants treat freshness the same way?

No. In the Ahrefs study ChatGPT cited the newest pages and Google AI Overviews cited pages of about the same age as Google's ordinary results. Honest dates and real updates serve both.

#How long after an update does an assistant notice?

Assistants that search the web live, such as ChatGPT search, Perplexity, Copilot and Google's AI features, can reflect an updated page within weeks, once it has been recrawled; Google says crawling can take from a few days to a few weeks. Answers drawn from a model's memory change only when the model is retrained, which takes months. For ID-Pal, the tracked count moved month by month: "UK buyer questions where AI names ID-Pal: none in May, two in August, four in September 2026."

Keeping content current is mostly discipline: a review calendar, a template with honest dates and a clear note when something changes. If you want an outside view of where your site stands, book a free Growth Review: 30 minutes with David Adams, and three fixes in writing afterwards. The rest of our content strategy and authority guides pick up from there.

#Sources