Refreshing old blog posts with AI for AI Overviews: a triage-and-rewrite method
Why old posts are invisible to AI engines, how to triage the archive, what to change in each post, honest dates and a measurable cadence.
Refreshing old blog posts for AI Overviews means selecting the 10–20% of the archive that already has authority, restructuring each post so every section opens with a direct answer, replacing stale numbers with dated ones, and updating the page honestly so dateModified reflects real change. AI drafting makes the rewrite fast; the selection and the fact-check still need people. Pages updated within the last three months are cited roughly twice as often as older ones.
- Do not refresh everything. Score the archive and work on the posts that already have traffic, links or conversions but no citations.
- The highest-leverage edit is moving the answer to the top of each section; most old posts bury it in paragraph four.
- Freshness is measured by real change, not by touching the date. A refresh that adds nothing is a refresh the engine ignores.
- AI speeds the rewrite from hours to minutes per post; the protocol fact-check stays at 45–90 minutes because every old number needs re-verifying.
- Set a cadence: top-20 posts quarterly, the next tier twice a year, everything else merged or archived.
01Why old posts are invisible to AI engines
Two mechanisms are at work. Retrieval engines prefer recent pages because recency is a cheap proxy for accuracy. And old posts were written for a different reader: a 2021 article was structured to keep someone scrolling, with the conclusion at the end. An engine that extracts passages needs the conclusion first. The archive is therefore doubly disadvantaged — stale and inverted — and both are fixable.
02Step 1 — Triage the archive
- 1Export the inventoryEvery URL with publish date, last-modified date, 12-month organic sessions, inbound links, conversions attributed.
- 2Score citation readiness (0–4 each)direct answer in first sentence of each section; at least one table or list; numbers sourced and dated; author page present.
- 3Cross with authorityPosts with traffic or links but a low readiness score are the targets — the engine already trusts the URL; it just cannot extract anything from it.
- 4ClusterGroup candidates by topic. Five weak posts on one topic become one strong one.
- 5Pick the top 20Not the top 200.
One B2B client with 340 posts found 28 that drove 60% of organic conversions and scored below 8/16 on readiness. Those 28 became the whole refresh programme — not the library.
(Source: Nav43 case, linked above.)
03Step 2 — What to change in each post
| Edit | Why it matters for AI Overviews | AI-assisted? |
|---|---|---|
| Rewrite first sentence of every H2 to answer the heading | The extracted passage is the first 1–2 sentences | Draft yes, approve no |
| Replace every number older than 12 months | Stale figures lower trust and get the page passed over | Find yes, verify no |
| Add a table where the section compares anything | Tables with headers are lifted whole | ✓ |
| Add or rewrite an FAQ block from Search Console queries | Q&A pairs are the unit AI Overviews extract | Draft yes |
| Remove forward/backward references ("as above") | Passages are read in isolation | ✓ |
| Update author byline to a real, linked person | Entity corroboration | ✕ |
| Consolidate overlapping posts and 301 the losers | Removes competing weak passages | Plan yes, execute no |
The first row is where most of the citation gain comes from. The pattern from the Nav43 audit: a paragraph that read "Research indicates that content freshness is an important factor…" became "Content updated within 3 months is 2× more likely to be cited by ChatGPT (SE Ranking, 2025)." Same fact, now extractable, sourced, dated.
04Step 3 — Where AI fits in the refresh
AI handles the mechanical rewrite well: restructuring sections, drafting direct-answer openers, building a table from a prose comparison, generating FAQ candidates from a query list. It should not be trusted to update facts, because the one thing a model will do with a stale number is replace it with a plausible one from its training data.
The fact-check is the same protocol used for new articles — the human-in-the-loop workflow — and it takes as long on a refresh as on a new piece, because every legacy claim has to be re-verified, not just the new ones.
05Honest dates
Changing dateModified without substantive change is a known pattern and engines discount it. A refresh counts when the diff is real: new numbers, restructured sections, a new table, an added limitations paragraph. Keep the original publish date visible and show the update date alongside it.
Keep a one-line changelog on the page or in your ledger: "Updated October 2026: pricing figures, added comparison table, rewritten FAQ." It is a trust signal for readers and the record you need when an engine quotes an older version back — the fix path for that is in Google AI Overview wrong information about your business.
06Step 4 — Cadence
| Tier | Which posts | Cadence | Work per cycle |
|---|---|---|---|
| 1 | Top 20 by conversions and citations | Quarterly | Full protocol refresh |
| 2 | Next 30–50 with traffic or links | Twice a year | Numbers, openers, FAQ |
| 3 | Everything else | Once | Merge into a Tier 1/2 post or archive with a 301 |
Tier 3 is the step teams skip, and it is the one that most changes how a site reads under the scaled content abuse policy: fewer, stronger pages outperform a long tail of thin ones in both Google and AI answers.
07Measuring whether it worked
Track per refreshed post, from the republish date: AI referral sessions (chatgpt.com, perplexity.ai, Gemini referrers), share of answer on the post's target prompts, and AI Overview appearances for its queries in Search Console. Report at 30, 60 and 90 days — the dashboard structure is in Your AI visibility dashboard is live. Now what?.
Questions people also ask
Should I refresh old blog posts or write new ones?
Refresh first. Old posts with traffic or links already carry authority; they lack extractable structure and current facts. A refreshed post reaches citation faster than a new one that starts with no signals.
How often should blog posts be updated for AI search?
Top-performing posts quarterly, the next tier twice a year. Pages updated within three months are cited about twice as often as older pages, and in some audits most top citations went to pages updated within 30 days.
Can I use AI to update old blog posts?
Yes, for restructuring, drafting direct-answer openers, building tables and generating FAQ candidates. Not for updating facts — every number must be re-verified by a person against a current source.
Does changing the modified date help with AI Overviews?
Only when the change is real. Date changes without substantive edits are discounted. Keep a visible changelog of what was updated.
