Updating Old Content for AI Search: What to Change and What to Leave

Start with the pages that already perform

Do not refresh everything. Pull a list of pages that get organic traffic or rank for queries that AI engines also answer. Those are the pages where an improvement in structure translates into citations quickly, because the engines already know the URL.

Then ask the engines the question each page answers. If you get cited, note what was quoted. If you do not, note who was, and look at how their page is structured. That comparison tells you what to change.

What to change

The opening. Old posts tend to begin with a warm-up paragraph. Replace it with two or three sentences that state what the page covers and give the main answer. A model reading only the first chunk should get the point.

Headings. Rewrite vague headings as the questions people ask. "Getting started" becomes "How do I set up X?" Keep the heading hierarchy intact so you do not break anchor links.

Buried answers. In each section, move the answer sentence to the top. Keep the explanation beneath it.

Dead facts. Prices, version numbers, feature names, team sizes, and anything with a date attached. A page that says "as of 2022" is telling the engine it may be stale. Update the facts and remove the old date references unless they matter.

Structure. Turn lists hidden in prose into real lists. Turn comparisons into tables. Replace images of text with text.

Schema. Add or update Article schema with a dateModified that reflects the real change. Add FAQPage or HowTo where the content fits.

Internal links. Add links from the refreshed page to newer, related pages and back. Old pages often sit in isolation because they were written before the rest of the site existed.

What to leave alone

The URL. Changing a URL throws away the history that made the page worth refreshing. If you must restructure, use a 301 redirect, but prefer not to.

The core claims that are still true. If the page's central argument holds, keep it. Rewriting for the sake of freshness produces a worse page and gives the engine nothing new.

The parts that get quoted. If the engines already cite a particular passage, do not touch it. Improve the sections around it.

Length. Do not pad. A refresh is not an excuse to double the word count. Longer pages are not favored; clearer ones are.

The author. Keep the original byline if the author page still exists. A real person with a track record is worth more than a fresh "Team" byline.

The date question

Should you change the visible publish date? Only if the changes are substantial. Updating a date on a page where you fixed two typos is a signal engines tend to discount, and it can backfire if the content itself still reads as old. Set dateModified honestly, keep datePublished as it was, and show both if your template allows.

A workable process

One page per week is a sustainable pace for a small team and adds up to fifty pages a year. For each page: ask the engines the question, read what they cite, apply the changes above, republish, and ask again in a few weeks. Keep a log of what was changed and whether the citation shifted.

Do not expect instant results. Engines re-crawl and re-index on their own schedules, and in our testing the lag between a refresh and a change in citations varies from days to months.

How we handle refreshes

At AEO GEO Labs, content refreshes are usually the second phase after an AI visibility audit. The audit identifies which questions you are losing and which existing pages should be winning them. The refresh work follows this checklist, and the weekly report tracks whether share of answer moves for the questions each page targets.

Pick one page now

Choose the old page with the most traffic that is not being cited for its main question. Apply the changes above. AEO GEO Labs offers a free answer-ready checker that scores a page's structure, which makes a useful before-and-after comparison for each refresh.