Find content decay with AI

1. Compare the two periods. The agent calls search_insights with the movers recipe. It returns decliners, risers, new pages and lost pages for the current window against the previous one of the same length, with clicks, impressions and position for both. One data pull. Ask for by: query as well if you want the losing queries (a second pull).
2. Classify each drop. No extra call: the agent reads the position and impression changes. Worse position means a ranking loss; stable position with fewer impressions means demand fell; stable both but fewer clicks means CTR dropped.
3. Chart the shape. For the top decliners, the agent calls search_chart filtered to each page. In Claude this renders an interactive daily chart in the conversation; in other clients it comes back as a compact daily series. A cliff on one date points to a change (an update, a migration, a lost snippet); a slow slope points to staleness or new competition. One data pull per page.

A typical run is 2 to 5 data pulls. If the agent finds no decliners, the run ends after the first call.

This is a real run on mcpsearchconsole.com. Over 28 days (2 to 30 September 2026 against 5 August to 1 September), the movers recipe found no decliners at all: clicks went from 12 to 37 and every page with clicks was rising. That is an honest result for a young site, and it is what you should expect to see when nothing is decaying.

To show what a decliner looks like, the same recipe was run over a 14-day window (16 to 30 September against 2 to 15 September 2026):

The two decliners have different causes. The Grok guide held its position (5.3 to 5.4) while impressions fell by about 90%: people stopped searching for it as much, which fits a burst of interest that faded. The Claude guide lost position (8.1 to 13.8), falling off page one: a ranking loss. The search_chart call for the whole property over 56 days showed the same burst: daily impressions peaked at 200 on 2 September and settled between 25 and 55 a day in the second half of the month.

The verdict for this month: refresh the Claude guide (a real ranking loss on a page with steady demand), and leave the Grok guide alone, because its ranking held. Both drops are small in clicks; on a bigger site the agent would leave them out under the 5-click threshold in the prompt.

Each decliner's cause decides what to do with it.

• Ranking loss (position got worse). Refresh the page: update facts, dates and examples, add what newer results cover that yours does not, and check it is still indexed. These go to the top of the list. Pages slipping from page one into positions 11 to 20 also show up as striking-distance keywords.
• Demand fell (position stable, impressions down). Usually not fixable on the page: the topic is seasonal or interest has faded. Check the same period last year if you have it. Do not rewrite a page that still ranks well just because fewer people search.
• CTR loss (position and impressions stable, clicks down). Something changed on the results page, such as a new AI Overview, or Google rewrote your title. Treat it as a low-CTR page and work on the snippet.
• A sudden cliff on one date. Look for a cause before refreshing: a deploy, a redirect, a robots.txt change, a lost canonical. Run the free check_redirects and check_robots tools on the URL. If you recorded site changes as notes, ask the agent to read them.
• Lost pages (clicks before, none now). Check they still exist and are indexed before anything else.

Refresh no more than a handful of pages a month, in the order the agent gives, and re-run the prompt next month to see which recovered.

• It cannot tell you why demand fell. Search Console shows impressions dropping but not whether the topic is seasonal, a trend ended or AI answers are absorbing the search. Use your own knowledge of the market.
• Short windows are noisy. A 28-day comparison can catch a holiday or a one-off spike. The example's 14-day window was used only to show decliners on a small site; on your own site, prefer 28 days or longer.
• Up to 16 months of history. Search Console keeps about 16 months of data, so year-on-year checks are limited to recent months.
• No competitor data. A ranking loss usually means someone else gained, but Search Console does not show who.
• Anonymized queries. Query-level decliners can miss rare searches that Google withholds, even when the page-level numbers include them.

If you run this every month, pair it with a weekly SEO report to catch cliffs earlier. To connect first, follow the guide to use Search Console in Claude, where search_chart renders as a live chart.