Content Refresh for Pages AI Stopped Citing

A page that used to be cited and no longer is gives you something rare in content work: proof that it once worked. That makes the refresh far less speculative than writing from scratch. The risk is the opposite one, rewriting the page before you know why it slipped. This process puts diagnosis first and editing second.

Step 1: confirm the loss is real

Models vary between runs. One missing citation is noise. Before you open the editor, check the trend over a few weeks. In Promptwatch, Page Tracker shows responses, distinct prompts and the last-cited date per URL, and Citation Trends gives you peak rate and decay half-life per page. If the rate has fallen to a fraction of its peak and stayed there, the loss is real.

Also check across engines. If the page lost citations only in one model, the cause may be model-specific. If it fell everywhere, suspect the page or access.

Step 2: rule out access

Never rewrite a page the bots cannot reach. Look at crawler logs for the URL: are search and user-fetch bots such as OAI-SearchBot, ChatGPT-User or PerplexityBot receiving 200s? Agent Analytics shows IP-verified hits with error tracking and failure explanations. A string of 403s or timeouts means the fix is infrastructure, not prose.

Step 3: find who replaced you

Open the prompts that used to cite the page and read the sources now being used. Who is there instead? Add that competing URL to the Page Tracker. You can track any URL, not just your own, so you can see when theirs rose and compare against when yours fell.

Then read it as a customer would. Questions to ask:

• What does it include that yours does not: a table, a step list, current pricing, an example?
• Is it more specific, or just fresher?
• Does it answer the sub-questions the model searched for? Promptwatch lists query fan-outs per model, so you can check them directly.

Step 4: decide the type of refresh

Not every slipped page needs the same treatment. Sort it:

1. Staleness. Dates, prices, versions or screenshots are out of date. Update facts and add a visible last-reviewed date.
2. Coverage gap. The page lacks a section the model now seeks. Add it.
3. Structure. The answer exists but is buried. Move it up, add a direct answer in the first paragraph, and use clear headings.
4. Ownership shift. A forum thread or video now supplies the answer. A refresh may not be enough; consider contributing there too. Reddit, YouTube and LinkedIn citation reports in Promptwatch show which of these sources are being cited for your prompts.

Step 5: use diff-style edits

Large rewrites erase whatever earned the original citations. Prefer targeted changes. Promptwatch's Content Refresh produces diff-style update suggestions for existing pages, which keeps you honest about what you are touching. Review each suggestion against your facts. Do not accept a changed number or claim unless you can source it.

Keep a record of what you changed and the date. You will want it in four weeks when you read the result.

Step 6: publish and mark the date

If you publish through WordPress, Webflow or Framer, remember that Framer has no per-item publish: going live redeploys the whole Framer site, so batch your updates. Promptwatch publishes to all three with AI field mapping, and an official WordPress plugin exists.

Then note the refresh date on the Page Tracker row or in your log. Without a marker, you cannot tell a refresh effect from random drift.

Step 7: wait the right amount of time

Do not judge in three days. Use the ramp-up days from Citation Trends as a guide: if a page typically takes a couple of weeks to climb, give the refreshed version that long. Check crawler logs first to confirm the bots re-fetched the page after your edit. If they did not, no model has seen the new version yet.

A worked scenario

Imagine a comparison page whose citation rate has fallen to about half its peak. Logs show clean 200s from OAI-SearchBot. A competitor's page now holds the citations and includes a current pricing table; yours has a table from last year. Query fan-outs show models searching for "pricing" as a sub-question. Diagnosis: staleness plus a coverage gap. Treatment: update the table with sourced numbers, add a short pricing section near the top, and republish. Then watch crawls, then citations.

This is a hypothetical, not a customer result, but the sequence is the one to follow.

When not to refresh

If a page never earned citations, a refresh is a guess. Treat it as a content gap and write to the prompts instead. Promptwatch's content gap analysis scores coverage per topic which shows where prompts lack a good answer from you.

Recommendation

Make refreshes a weekly queue item, not a quarterly panic. The Action Board in Promptwatch can surface a slipping page as an evidence-backed action, assign it, and track it to done. Promptwatch is the tool we would use because the same workspace holds the decay curve, the crawler proof, the competitor page and the editing and publishing step. Start with your three most-slipped pages, run the seven steps, and write down what you learn.