Should you rewrite existing pages or publish new ones to get cited by AI?

Rewrite the existing pages, and publish new ones only where you have no page that answers the question at all. That order holds for almost every B2B site we look at, for a simple reason: most of them already have a page on the subject a buyer is asking about, and the page is written in the company's own vocabulary rather than the buyer's.

A rewrite is also cheaper than it looks and a new page more expensive. The new page needs a brief, a draft, review, internal links and somewhere to live in the navigation. A rewrite needs a decision about what the page should say first, and then an afternoon. The two jobs get budgeted as though they were the same size, which is how content programmes end up with sixty pages and no answers.

Where the instinct to publish does pay is on questions you have never addressed. If a buyer asks something your site has no page for, no amount of editing produces one. The test is not whether the topic feels covered in a meeting. The test is whether a page exists whose title is the question.

One qualification matters before the detail. Rewriting a page changes what it says, and what it says has to be true, so a rewrite programme runs at the speed of whoever approves claims. Teams that skip that step publish faster and get corrected in public.

What does the evidence say about refreshed pages versus brand new ones?

The evidence on refreshed pages against new ones points one way, and the most useful study of it is about update dates rather than publication dates. Seer Interactive's content recency study, published 24 July 2026, took 7,683 cited pages carrying 47,097 citations from ChatGPT, Gemini and Perplexity between March and June 2026, and found 75% had been updated within the last year and 88% within two.

The second half of that study is the part worth reading twice. On the 4,124 pages where both a publication date and an update date could be read, 72% looked fresh by last update and only 42% had actually been published in the past year. Among the highest-volume cited pages, 75% had been updated in the last year and 40% first published then. Old pages, kept current, are the bulk of what gets quoted.

Two cautions belong with those numbers. Seer sells SEO services and used a commercial visibility tracker for the citation data, and the study covers four brands in four verticals, so the vertical splits rest on small slices. Its own figures for retail energy differ too, 80% in the prose against 72% in the table, which rest on different subsets of the pages. The headline finding survives both, because it is a large difference in the same direction across three engines.

What the freshness result does not mean is that a date stamp is a lever. Discovered Labs, in a study of about 2 million prompt and citation observations across 10,000 cited pages, put page age at a standardised coefficient of +0.05 against +0.37 for how closely a page's content matches the prompt. Discovered Labs sells AEO services, which is worth knowing when reading its ranking of what matters. Updating a page helps because the update changes the page.

Which of your existing pages should you rewrite first?

Rewrite the existing pages that already get retrieved, starting with the ones that answer a buying question. Retrieval is where the odds are decided: Chen and Koudas, October 2026, measured 158,099 answers and 468,037 citations and found the engine's own search step explaining about 32% of the variation between pages, against 2.8% for the properties of the pages themselves. A page an engine already reaches is most of the way there.

In practice that means three groups, in order. First, pages that rank somewhere on page one or two of classic search for a question buyers ask, because something already considers them relevant and nobody is choosing them. Second, pages you have seen quoted in an AI answer, which are proof of retrieval. Third, pages that carry your strongest claim about a buying decision, however they currently perform, because those are the ones you want quoted when the answer names three vendors.

What to change once you pick one is narrower than a full rewrite usually implies. Put the buyer's question in the title and the H2s, in the words they use. Answer each question in the first sentence under it, then support the answer. Attach a source to every number. Those are the moves that show up as alignment, and alignment is the one page-level factor that survived in both of the studies above.

Skip pages that cannot be retrieved for a mechanical reason until the mechanical reason is fixed. A page behind a form, a page that renders only in JavaScript, or a page nothing links to will not start getting cited because the wording improved. Our guide to checking whether AI can read your website covers how to tell which of those applies.

When is publishing a new page the right answer?

Publishing a new page is right when no page you own answers the question, when you have original research nobody else can publish, or when an existing page is about a different decision and would be damaged by having a second one bolted on. Those three cases are narrower than most content calendars assume, and all three are defensible in a planning meeting.

The original research case is the strongest of the three, because it is the one thing competitors cannot copy by editing. Our own study of how four AI engines answer pharmaceutical process skid sourcing questions found between 73% and 89% of citations pointing at suppliers' own capability pages rather than directories or marketplaces, across six categories. No rewrite of an existing page could have produced that, and it needed a page of its own.

The case against publishing is the one teams discover late. Two of your pages written for the same question compete with each other, and an engine picking between them has no reason to prefer either, so the usual outcome is that neither becomes the obvious source. If you are unsure whether a planned page overlaps something you already have, search your own domain for the question before you brief it. We work through the decision in whether two pages can compete for the same AI answer.

There is also a volume question underneath this, and it has a boring answer: the number of pages matters much less than whether each page answers something. Our page on how much content you need to get cited by AI sets out what that means for a small library.

How much can a rewrite actually move the result?

A rewrite can move a page's readiness score a long way in a day, and we published the clearest example we have on ourselves. We ran SIGNALS on our own application page and it scored 28 out of 100, with Alignment at 2 out of 10. After nine content changes, made in a single afternoon with no developer involved, the same URL scored 71 with Alignment at 8. The whole teardown, including what we could not fix, is at from 28 to 71, how we fixed our own page.

The limit on that number matters more than the number. A readiness score measures whether a page can be quoted, not how often it will be. The Chen and Koudas study above puts page properties at 2.8% of the variation between pages, so a score going from 28 to 71 is a page becoming eligible rather than a forecast of citations. Anyone selling you a score as a prediction of citation volume is selling past what has been measured.

The useful way to read a rewrite, then, is as the removal of a reason to skip you. A page that answers a buying question in the buyer's words, with sources attached, is one an engine can use; whether it does also depends on your domain being something the engine searches for in the first place. Discovered Labs found what it calls AI-perceived domain authority carrying a mean absolute SHAP value of 0.38 against 0.06 for the strongest page-level feature beyond alignment, which is the same conclusion arriving from a different model.

Expect the first visible movement in weeks rather than days, and expect it unevenly across engines. Our page on how long AEO takes to work covers the timeline question with the numbers we have.

Rewrite or publish: which should you choose in each case?

Choose between a rewrite and a new page by asking what already exists for the question, not by asking which feels more ambitious. The table sets out the five situations we meet most often, what we do in each, and the reason.

Your situation	What to do	Why
A page covers the question but is written in your vocabulary	Rewrite in place, keep the URL	Alignment is the page-level factor with the largest measured coefficient, and the page is already retrievable
A page ranks on page two of search and gets no clicks	Rewrite it first, before anything new	Retrieval explains about 32% of the variation between pages; this page is already being reached
Nothing you own answers the question	Publish a new page whose title is the question	No edit can produce a page that does not exist
Two of your pages already answer it	Merge into one and redirect the loser	Two pages for one answer split your own retrieval chances and neither becomes the obvious source
You have data nobody else has	Publish it as research on its own URL	Original data is the one thing a competitor cannot match by editing an existing page

One sequencing note sits behind the whole table. Rewrites compound and new pages dilute while they mature, so a quarter that starts with twenty rewrites and ends with four new pages generally reports better than the reverse. If you want the list of which of your pages to rewrite first, scored rather than guessed, our free visibility assessment produces it from your own pages.

What else do people ask about rewriting pages for AI search?

The five questions below come up when a marketing team has an existing library, a fixed budget for the quarter, and has to decide how much of it goes into pages that already exist.