The update type that most reliably moves citations is the factual update: correcting a wrong or outdated fact on a page that AI engines already cite. Expansion updates come second. Structural updates come last. That ranking comes from what you can watch in server logs and citation checks, not from a single published study.
A content update is any change to a published page that alters its meaning, facts or structure. Three types matter for citation work, and they behave differently. This piece is a measurement explainer: it covers the signals you can watch, the ones you can't, and how to tell which type of update is worth your time.
Any update is not automatically a helpful update
Lots of teams treat a refresh as one action. They think "updated" is a status, and that search engines or AI engines reward it. That's the myth. No public documentation from Google, Bing, OpenAI or Anthropic says a recent modification date lifts rankings or citations.
What the platforms do say is narrower. Google's documentation on publication dates says Google uses dates to help it understand a page, but it doesn't treat a date as a ranking boost. The sitemaps.org protocol has a lastmod field, and the sitemaps.org protocol says it should reflect the last substantive change, not a cosmetic edit.
So the right question isn't "did I update the page?" It's "did I change something an engine has a reason to read differently?" That's where the three types separate.
Factual updates fix the exact claim an engine is answering
A factual update changes a fact that is wrong, outdated or inconsistent with another page on the same site. It's the smallest edit by word count and usually the largest by citation effect. The reason is simple: AI answers cite pages that are correct on the specific claim being asked about. If the page is wrong, the engine has no reason to keep it in the answer.
Examples: a price that moved, a regulation that changed year, a product spec that was superseded, a job title that is no longer accurate. The edit might be one sentence. It might be one number. The important part is that the page now agrees with the outside world and with your own other pages.
This is where entity consistency matters. If your About page says a service launched in 2021 and your service page says 2022, an engine reading both has a conflict. Fixing the wrong one is a factual update. Schema.org markup that carries the same fact (for example datePublished or price) should change at the same time, or the structured data will disagree with the visible text.
Expansion updates add reach but can dilute a tight page
An expansion update adds new material to an existing page: a new section, a worked example, a comparison table, a definition, a FAQ entry. It doesn't fix a wrong fact. It gives the page more surface area to be cited on a related question.
Expansion works when the page already ranks or gets cited for a head term, and you want it to also cover a related sub-question. For example, a page on invoice software that gets cited for "what is an invoice" can be expanded with a section on invoice retention rules. The page becomes a candidate for a second query.
The risk is dilution. Adding 2,000 words to a tight 800-word page can make the answer harder to extract. AI engines quote passages, not whole pages, so the new section needs to stand on its own. Put the direct answer in the first two sentences of the new section, then the detail.
Structural updates fix defects, not thin content
A structural update changes how the page is organised: heading hierarchy, internal links, schema types, table of contents, canonical tags, the order of sections. The facts and the coverage stay the same. Only the packaging changes.
This is the type most teams over-invest in and the one with the weakest observable link to citations. That's not because structure doesn't matter. It's because structure mostly affects whether a crawler can parse and segment the page, not whether the content deserves a citation. If the facts are wrong or the coverage is thin, better headings won't fix that.
Structural updates do matter in three cases: when a page has no clear H1, when the schema doesn't match the visible content, and when internal links point to the wrong target. Those are fixable defects. Beyond that, restructuring for its own sake is hard to measure.
Side-by-side comparison
The table below ranks the three types by expected effect on citations, effort and risk. The risk column is about the chance of doing damage, not the chance of failure.
| Update type | What changes | Typical effort | Risk to existing rankings | Observable signal |
|---|---|---|---|---|
| Factual | A wrong or outdated fact, plus matching schema | Low (minutes to an hour) | Low, if the new fact is correct | Citation checks on the specific claim; log re-crawl of the URL |
| Expansion | New sections, examples, tables, FAQ entries | Medium (half a day to two days) | Medium, if the page gets diluted | New query coverage; impression growth in Search Console |
| Structural | Headings, internal links, schema types, order | Low to medium | Medium, if URLs or canonicals change | Crawl and index status; rich result eligibility |
The effort estimates are typical ranges, not quotes. Your numbers depend on page length and review process.
What you can actually observe
This is the measurement part, and it's where most refresh advice gets vague. Here's what's checkable and what isn't.
Observable: whether a crawler fetched the URL after the change (server logs); whether the page is indexed (Search Console URL Inspection, Bing Webmaster Tools); whether the page appears in Search Console impressions for a specific query; whether an AI engine's answer cites the URL on a specific prompt, checked manually and logged with date and prompt text.
Not observable from outside: the exact weight an engine gives to freshness; whether an engine re-read the page after your edit but before its next scheduled crawl; the internal ranking of a passage inside a model's retrieval. There is no public figure for these.
That gap is why the practical test is a before-and-after citation check on a fixed prompt set. Pick 5-10 prompts the page should answer. Record which URLs are cited before the update. Make one update type. Re-run the same prompts after two to four weeks. Log the result. That's the only signal you control end to end.
Decision rules and cadence
Use these rules in order. Stop at the first one that applies.
- If a fact on the page is wrong, outdated or conflicts with another page on the site, do a factual update first. Nothing else matters until that's fixed.
- If the page is cited for a head query but not for a related sub-query you want, do an expansion update on that sub-query.
- If the page has no clear H1, broken schema, or internal links pointing to the wrong target, do a structural update.
- If none of the above apply, leave the page alone. An unnecessary edit is a risk with no upside.
The order matters because factual errors undermine everything else. An engine that reads a wrong price on your page has a reason to drop the page from the answer, no matter how well structured it is.
Factual updates should be event-driven, not scheduled. When a price, date, name or spec changes, update the page and its schema in the same edit. Don't wait for a quarterly cycle.
Expansion updates work on a review cycle. A quarterly review of the top 20 cited pages is a reasonable starting point. That's a working default, not a rule from any platform.
Structural updates should be rare. Fix defects when you find them. Don't restructure a page that's already being cited just to try a new layout.
Keep the URL. If you must change it, use a 301 redirect and update internal links in the same release. Google's documentation on redirects treats a 301 as a signal that the old URL has moved, but it doesn't promise the same rankings transfer instantly.
Don't change the page title and the H1 and the URL in the same week. If something drops, you won't know which change caused it. One change type per review window is the cleanest way to keep the signal readable.
Keep a change log with the date, the type of update, the URL and the prompt set used for citation checks. That log is the only way to compare types over time.
Key Facts at a Glance
- Factual updates have the clearest link to citation changes, because the correction targets the exact claim being answered.
- Expansion updates add query coverage but can dilute a tight page if the new section doesn't answer its question in the first two sentences.
- Structural updates fix parse and linking defects; they don't fix wrong facts or thin coverage.
- Server logs show whether a crawler fetched the page; they don't show whether the engine re-weighted it.
- A fixed prompt set checked before and after is the most reliable in-house measure of citation movement.
FAQ
What is the difference between a factual update and an expansion update?
A factual update corrects a wrong or outdated fact on an existing page. An expansion update adds new material to cover a related question. Factual updates fix accuracy; expansion updates add reach.
How often should you refresh old content?
Factual updates should happen when a fact changes, not on a schedule. Expansion updates fit a quarterly review of your top cited pages. Structural updates should be rare and defect-driven.
How do you know when content needs updating vs rewriting?
If the page is cited for its main query and the facts are correct, an expansion update is usually enough. If the page is cited for nothing and the facts are wrong throughout, a rewrite is the better call.
Can updating content hurt your rankings?
Yes, if you change the URL, the canonical, or the core answer without a redirect or a replacement. Structural changes carry the most risk because they can break parse and linking signals.
How do you measure content refresh success?
Pick a fixed set of prompts the page should answer. Record which URLs are cited before the update. Re-check the same prompts after two to four weeks. Log the result by update type.
