The AI search content refresh playbook

New content gets all the attention. In AI search, a lot of the quick wins sit in pages you already have. A page was cited for months, then a competitor published something sharper, your pricing changed, or the engine started looking for a different angle. Nobody noticed, because the page still ranks fine in Google and nothing visibly broke.

A refresh program catches that. This is the playbook we run for clients, roughly once a month.

Step 1: pick pages by evidence, not by age

The classic refresh rule is "update anything older than a year". That wastes time on pages nobody needs and misses recent pages that are already slipping. Start from AI visibility data instead and look for four signals.

Lost citations. A page that used to be cited for a tracked prompt and isn't anymore. This is the strongest signal, because you know the page was good enough once.

Crawled but never cited. AI crawlers fetch the page regularly, but it never shows up as a source. Access isn't the problem. The content is.

Cited with stale facts. The page still gets cited, but it carries an old price, a renamed plan, or a feature you've retired. That's a risk, because the engine may repeat the old fact in its answer.

Competitor overtake. On a prompt where you used to be named first, a competitor now leads, and their newly cited page covers ground yours doesn't.

Rank candidates by how many tracked prompts each page touches. A page tied to twelve prompts beats a page tied to one.

Step 2: diagnose before you rewrite

For each candidate, open the current answer for its prompts and the pages now cited instead. Ask what they have that yours lacks. It's usually one of these:

• A direct answer near the top, where yours builds up to it
• A subtopic the engine now searches for, visible in the prompt's fan-outs, that your page doesn't cover
• A comparison table where yours has paragraphs
• Newer facts, or a visible date
• A clearer statement of who the product is for

Write the diagnosis down in one sentence per page. If you can't, you're not ready to edit.

Step 3: refresh for real

A refresh means the page answers the question better. Changing the date and swapping three adjectives doesn't count, and it can backfire if a reader or an engine spots facts that don't match the new date.

What a real refresh usually involves:

1. Correct every fact that changed. Prices, plan names, limits, integrations, screenshots.
2. Move the direct answer to the top of each section.
3. Add the missing subtopics from the fan-outs, each under a heading that matches the question.
4. Add a comparison table if buyers compare on this page.
5. Cut what no longer helps. Shorter and current beats long and padded.
6. Update dateModified in the page and the markup, and only then.

Step 4: keep the URL and help it get recrawled

Keep the same URL. It already has links, history, and in some cases citations. If you must move it, redirect in one hop.

Then nudge the crawlers. In Google Search Console, use URL Inspection to request indexing for the updated page. For Bing and other engines that support it, IndexNow lets you notify them of the change. Update your XML sitemap's lastmod for the page. None of these guarantees a fast recrawl, but they remove delays you control.

Step 5: confirm the bots came back

Before judging the result, make sure the AI crawlers have actually fetched the new version. Judging a refresh no bot has seen is how teams conclude that good work didn't work.

Check your crawler logs for OAI-SearchBot, PerplexityBot, ClaudeBot, and the others hitting the updated URL after the deploy, with 200 responses.

Step 6: re-check the prompts

Once the page has been refetched, run its tracked prompts again over a few weeks. Look for the page returning as a citation, the stale fact disappearing from answers, and your brand's position on those prompts. Then put the page back in the pool. It'll need another look eventually.

Running it as a program

This is where we lean on Promptwatch, and specifically on its Content Agents. Content gap analysis compares tracked prompts against your existing pages and flags the ones that should be cited but aren't, which covers most of step 1. The Content Agent then drafts content aimed at the prompts and fan-outs a page is meant to win, and we use those drafts as raw material for the refresh. Drafts land in a review inbox, so an editor checks facts and voice before anything goes out. Approved changes publish to a connected CMS: Webflow, Framer, or WordPress through the official plugin. One Framer quirk to plan for: there's no per-page publish, so going live redeploys the whole site.

Plan matters here. The automated Content Agent that plans and publishes on a schedule is on the Business plan and Enterprise, not the lower tiers or the agency plans. On other plans we use content gap analysis and Agent Chat to build the refresh list and write the drafts ourselves.

The rest of the playbook stays in the same workspace. Agent Analytics covers step 5 with real-time crawler logs, and prompt trends and citation analytics cover step 6. Most months the refresh list ends up short and specific. That's the point. You fix the pages the data says are slipping, confirm they got picked up, and move on.