Refreshing Old Content for AI Search: A Practical Workflow

Which pages to refresh first?

Refreshing old content means updating the facts, examples, dates and sources on a page that's still relevant but has gone stale — not just editing the publish date. Do this on your best-performing, underperforming or just-outdated pages first, log what changed, and merge or remove pages that are too thin to be worth refreshing. Freshness most plausibly matters for AI engines that retrieve content live at answer time (ChatGPT's search feature, Perplexity), less so for engines leaning more on trained knowledge — though neither OpenAI nor Google publishes the specific mechanism, so treat this as a sound inference, not a confirmed lever.

Prioritise in this order:

1. Pages with real search demand but falling or flat impressions. In Search Console's Performance report, sort by impressions and look for pages that used to get more traffic or are stuck despite ranking.
2. Pages that answer a question AI engines plausibly retrieve for, but are missing a direct answer near the top. A long, rambling page that technically covers the topic but makes the model work to extract an answer is a weaker candidate for citation than one with a clear answer up front.
3. Pages with outdated facts you already know about — old pricing, a deprecated feature, a statistic that's since been superseded. These carry real risk: an AI engine citing your old pricing page is a brand-accuracy problem, not just an SEO one.
4. Everything else, refreshed on the general schedule below once the first three categories are handled.

Google's own Search Console reporting is the practical tool here: the Performance report (filtered to your content pages) shows which pages are losing impressions or clicks over time, and the newer generative AI performance report shows impressions specifically from AI Overviews and AI Mode — though as of its own documentation, that report is impressions-only, with no clicks, and some sites may not see data at all if they get too few AI-feature impressions to report.

What exactly to update?

A refresh is a real content change, not a cosmetic one. On a page worth refreshing, check and update:

• Facts and numbers. Anything stated as current (pricing, feature lists, statistics) that may have changed since you wrote it.
• Dates and examples. Swap stale examples ("as of 2023...") for current ones; update any date references.
• Sources. Re-check every external link and citation still resolves and still supports the claim you're making with it.
• The direct answer. If the page answers a question, make sure the answer is stated plainly near the top, in a sentence or two — this is what a retrieval-based engine is most likely to lift directly. Answer-first content covers this pattern in more depth.
• Structure. Headings, FAQ sections, and anything that was written as one long block of prose when it should be scannable now.

AI-assisted rewriting is fine for drafting the update, but a human has to fact-check every number and claim before it goes live. Using a model to rewrite a stale paragraph without checking whether the "updated" numbers it generated are actually current is how fabricated facts get published — the model doesn't know your real current pricing unless you tell it. Doing this at scale across hundreds of pages without that check is the same mistake mass-producing AI content makes, just applied to old pages instead of new ones.

How often?

There's no universal interval; it depends on what the page is about:

Google's AI optimization guide doesn't specify a mechanism or cadence for how freshness affects its generative AI features — the guide states general SEO practices remain relevant because its AI features are "rooted in our core Search ranking and quality systems," but it doesn't break out freshness as its own documented signal for those features specifically. Treat "refresh regularly" as reasonable, evidence-adjacent practice, not a documented AI-specific rule. [VERIFY: whether Google or any engine has published a more specific freshness mechanism for AI features since this guide's last update — re-check on the day of publishing.]

How to log changes?

Keep a short, visible change log on the page itself, one line per material update — for example:

- 2026-07-14: updated pricing table to current tiers.
- 2026-10-10: replaced 2024 statistic with the 2026 study; added one FAQ.

Only log changes a reader would actually notice — a new source, a corrected fact, a new section, a fixed broken link. Don't log formatting-only passes or bump a "last updated" date without a real change behind it; an inflated or routinely-bumped freshness signal is the kind of thing that erodes trust with both readers and crawlers over time once it's noticed.

Worked example of a refresh, start to finish:

1. Search Console flags a comparison page with falling impressions over 90 days.
2. You find the pricing table cites a tier that was discontinued four months ago.
3. You update the pricing, add one new competitor that's emerged since, rewrite the direct-answer paragraph at the top to lead with the current recommendation, and re-check all five external links (one is dead; replace it).
4. You add one line to the change log: 2026-10-10: updated pricing table, added [Competitor], fixed one dead source link.
5. You note the page for a re-check at next quarter's refresh pass.

When to prune or merge instead?

Not every old page earns a refresh. Merge or remove when:

• Two or more pages answer the same question and compete with each other rather than each serving a distinct angle — merge into one stronger page and redirect the others.
• A page has never had meaningful search demand or AI-answer relevance, and a refresh wouldn't change that — pruning (removing or redirecting it) is more honest than polishing something nobody needs.
• A page is thin — a few hundred words that don't actually answer the question in full — refreshing thin content into the same thin shape doesn't fix the underlying problem; it needs a real rewrite or a merge into a more complete page.

Common mistakes: refreshing a page's date without a real content change; rewriting with AI and skipping the fact-check pass; refreshing a page that should have been merged or deleted instead; refreshing everything on a fixed calendar regardless of actual performance data.

A refresh checklist:

• Pulled the candidate list from Search Console impressions/clicks data, not guesswork.
• Checked every stated fact, price and date for currency.
• Re-verified every external source link still resolves and still supports the claim.
• Rewrote the direct answer to be extractable in the first few sentences.
• Added one line to the change log.
• Considered whether this page should be merged or pruned instead of refreshed.

Tools to use: Google Search Console (Performance report and the generative AI performance report) to find candidates; a spreadsheet to track the refresh calendar by page type; the Direct-Answer Snippet Checker to confirm the rewritten page actually has an extractable answer near the top.

FAQ

Does updating the "last modified" date alone help AI visibility? No evidence supports that, and it risks the opposite — an inflated date with no real change teaches both readers and automated systems to stop trusting the signal. Only bump it alongside an actual content change.

Can I use AI to rewrite an old page? Yes, for the first draft of the rewrite. A human still has to verify every fact, price and statistic in that draft before publishing, since the model doesn't know what's actually current for your business.

How do I know if a page should be merged instead of refreshed? If another page on your site already answers the same question, or the page has never had real search demand, merging or pruning is usually the better move than polishing it in place.

Does freshness matter equally across all AI engines? Probably not equally, though nobody has published a direct comparison. Engines that retrieve live at answer time (ChatGPT's search feature, Perplexity) plausibly benefit more from current pages than ones relying more on trained knowledge for a given answer — this is a reasonable inference from how each engine works, not a measured, sourced claim.

What to do next

1. Pull your lowest-performing pages with real search demand from Search Console and list your top five refresh candidates.
2. Refresh one page fully — facts, sources, direct answer, change log — rather than lightly touching ten.
3. Check whether any other page on your site answers the same question before you refresh; merge first if so.

ContextOwl sells one-time AI visibility reports. The Direct-Answer Snippet Checker is a free way to check whether a page you've just refreshed has an answer AI engines can actually extract.