Content Refresh for AI Citation: Which Pages to Fix First

Most organisations discover their AI visibility problem the same way: someone types a buying question into ChatGPT or Perplexity, a competitor is named in the answer, and they aren’t. The instinct is to write more content. The better move is to fix what already exists, because pages that already rank and already carry authority are the cheapest route to citation.

The hard part is deciding which pages to fix first. A refresh programme that treats every URL as equally worthy burns months and produces a marginal lift. This piece sets out how to triage.

Why refreshing beats publishing for AI citation

Answer engines assemble responses from sources they can retrieve, parse and trust. Retrieval favours pages that are already indexed, already linked and already associated with the topic. A new page starts with none of that.

An existing page that ranks on page one for a commercial query is already inside the retrieval set for the questions your buyers ask. It’s often failing on the second half of the problem: the content isn’t shaped in a way a model can lift a defensible sentence from. That is a formatting and specificity problem, not an authority problem, and it’s fixable in an afternoon.

Which pages should you fix first?

Triage on three axes: commercial value, retrieval proximity and citability gap. A page is a priority when it scores well on the first two and badly on the third.

• Commercial value. Does the page sit against a question a buyer asks before they choose a vendor? Comparison pages, pricing explanations, methodology pages and “how do I evaluate X” guides matter more than culture posts.
• Retrieval proximity. Is the page already indexed, already ranking in the top ten or twenty for its query, and already linked internally? Pages with existing traction convert refresh effort into citation faster than orphans.
• Citability gap. Can a model extract a clean, attributable claim from this page without needing to synthesise across five paragraphs? If the answer to the reader’s question is implied rather than stated, the gap is wide.

Run those three filters and a 400-page site typically collapses to a working list of twenty to forty URLs. That list is the programme.

What does a citability gap actually look like?

The most common failure is the buried answer. The page is titled around a question, then spends four paragraphs on context before the answer arrives, hedged and qualified. A model summarising that page has nothing crisp to quote, so it quotes a competitor who led with the answer.

The second failure is unattributable generality. Statements like “results vary depending on your circumstances” are true and useless. Models prefer sources that commit: a defined range, a named condition, a stated mechanism. Specificity is what makes a sentence quotable.

The third is structural. Long unbroken prose, no headings phrased as questions, no lists, no dates, no author. Retrieval systems chunk pages; a page with no internal structure chunks badly and loses meaning at the boundaries.

How do you fix a page once you have selected it?

Work in a fixed sequence so the effort stays bounded. A refresh should take an hour or two per page, not a rewrite cycle.

1. State the answer first. Directly under the heading, answer the question the heading asks in two or three sentences. Everything else becomes supporting detail.
2. Convert headings into questions. Match the phrasing a person would actually type or speak. This aligns the page’s structure with the shape of the query.
3. Add the specifics you were hedging around. Ranges, conditions, exclusions, timeframes, who it doesn’t apply to. Constraints are a trust signal.
4. Break out one list or table. Comparisons, criteria and steps are disproportionately likely to be lifted into an answer.
5. Fix provenance. Named author with credentials, a visible last-reviewed date, and links to the primary sources you relied on.
6. Add an FAQ block. Three or four genuine follow-up questions with short, self-contained answers. Each one is an independent citation opportunity.

How do you sequence the programme?

Run it in waves rather than as a single sprint. Take the top ten pages by commercial value, refresh them, then wait long enough for recrawl and reindexing before judging the result. Measuring after a week produces noise.

Between waves, check whether the refreshed pages are actually being surfaced. Our ARIA citation tracker exists for this: it runs your priority questions across assistants repeatedly, so you can see whether a refresh moved you from absent to mentioned, or mentioned to cited. Without that feedback loop you’re refreshing on faith.

What should you not refresh?

Resist the urge to touch everything. Some categories are better left alone or consolidated.

• Thin duplicates. Four near-identical posts on the same topic compete with each other. Merge into one canonical page and redirect the rest.
• News and announcements. Time-stamped content is rarely the source an answer engine reaches for on an evergreen question. Leave it.
• Pages with no commercial adjacency. Being cited on a topic no buyer cares about is a vanity outcome.

Consolidation is often the highest-leverage move available. Combining several weak pages into one substantive page concentrates internal links and gives the retrieval system a single obvious best source instead of four mediocre candidates.

How do you know the refresh worked?

Traffic is a lagging and increasingly unreliable signal, because a cited answer may satisfy the user without a click. Judge the programme on citation outcomes instead.

Track, per priority question: are you named at all, are you named with a link, and is the claim attributed to you accurate. A mention without a link still shapes the buyer’s shortlist. An inaccurate attribution is worth fixing quickly, because it usually means the source page is ambiguous.

If you want a structural read on a page before you invest in rewriting it, the AI citability scorer gives you a diagnosis of the extraction problems on that specific URL. For the wider framework this sits inside, see our AI visibility practice.

Frequently asked questions

How many pages should be in a first refresh wave?

Ten to fifteen. Small enough to complete in a fortnight, large enough that the result isn’t attributable to a single page’s luck. Complete a wave before starting the next.

Does changing the publish date help?

Only if the content genuinely changed. A visible last-reviewed date on substantively updated content is a legitimate freshness signal. Rotating dates on unchanged pages is a credibility risk with no offsetting benefit.

How long before a refresh shows up in AI answers?

It depends on recrawl frequency and how each assistant sources its content. Some read the live web at query time and can reflect changes quickly; others rely on periodic index refreshes and take considerably longer. Plan on measuring over weeks, not days.

Should we refresh or write new content?

Refresh first. Existing pages carry retrieval advantages a new page has to earn from scratch. Write new content only where there is a genuine buyer question your site doesn’t address at all. Talk to us if you want help drawing that line.