How to Find Decaying Pages in Search Console and Fix Them With AI

Your best page lost 40 percent of its clicks and nothing told you

You published a guide eight months ago. It climbed to page one and brought 60 clicks a week. Now it brings 35, and you have been busy writing new posts, so you have not looked. This is content decay: a page that earned its position slowly loses it as facts age, competitors update and search results change.

Most small site owners spend all their time on new content. A cheaper source of traffic is usually sitting in pages that already rank. This guide gives a weekly routine for finding them in Google Search Console and using an AI model to decide what to fix.

What the documented case claims

One builder documented on this site, who goes by Marvomatic, built an n8n workflow that mines Search Console data for decaying pages and tells him what to fix (builtwithagents.ai/strategy/n8n-search-console-decay-workflow-20x-traffic). He reports his own site growing from roughly 50 to over 800 daily clicks, about twenty times, and credits running this decay-detection loop consistently. That is self-reported and not independently verified, and a result like that also depends on the site, niche and starting point. Read it as evidence that the loop is worth running, not as a number to expect.

You do not need n8n to start. You need a spreadsheet, Search Console and an AI chat. Automate it only after you have done it by hand three or four times and know what you want.

Step by step: the weekly routine

Step 1: Compare two periods by page

In Search Console open Performance, then Search results. Set the date filter to compare the last 28 days with the previous 28 days. Click the Pages tab so the table lists URLs, and sort by the click difference, biggest drop first. Why 28 days: it smooths out weekday effects while staying recent. If your traffic is very small, use 3 months against the previous 3 months.

Step 2: Export and shortlist

Use the export button to send the table to Google Sheets. Keep only pages that lost at least 20 clicks and at least 20 percent of their previous clicks. On a small site lower both thresholds, but ignore pages with a handful of clicks because the swings are noise. Aim for a shortlist of five to ten pages. Mark anything seasonal. A page about winter heating bills will drop in spring and does not need a rewrite.

Step 3: Pull the queries for each page

Click one page in Search Console, then the Queries tab, with the same comparison dates. Note the top five queries, their positions and the change. This tells you which of three things happened. If position fell, you lost the ranking. If position held but clicks fell, your title and snippet are no longer winning the click, or an AI summary or other result is taking attention. If both are flat but impressions fell, demand dropped.

Step 4: Ask an AI to diagnose

Paste the page text, the top queries and their positions into your AI chat and use a prompt like this: "Here is a page and the search queries it ranks for, with positions from the last 28 days and the 28 days before. Position dropped on these queries. List what is likely out of date or missing compared to what a searcher wants, and give me up to five specific edits. Do not rewrite the page. Mark anything you are unsure about as unsure." Asking for edits instead of a rewrite is the important part. Wholesale AI rewrites tend to flatten pages that were working.

Step 5: Check the live results yourself

Search your main query in a private window and read the top three results. Check what changed: newer dates, more complete answers, different formats. The AI cannot see the live results unless you give it a tool that can, and even then you should look. Two minutes here prevents a lot of wasted edits.

Step 6: Make the edits and mark the date

Update stale facts, prices and screenshots. Add the sections a searcher now expects. Improve the title and meta description if clicks fell while position held. Then change the visible updated date only if you really changed the page. Search engines and readers can both tell when a date is cosmetic.

Step 7: Log it and recheck in four weeks

Add a row to a sheet: URL, date edited, clicks in the prior 28 days, what you changed. Recheck in four to six weeks. Keep the log. After a quarter it shows which kinds of edits actually moved clicks on your site.

Automating it with n8n

The documented workflow runs the same logic on a schedule. A few notes from checking the tooling. Search Console exposes a Search Analytics query API, where you POST a start date, end date, dimensions such as page and query, and a row limit of up to 25,000. Google notes that it returns top rows rather than guaranteeing every row, which matters for large sites. I could not confirm a built-in Search Console node in n8n's documentation, and the generic Google OAuth credential page does not list a Search Console scope. So expect to do some credential setup work, or use a community node or a Sheets export as the input instead. If that sounds like too much, run the manual version.

As of November 2026 n8n Cloud Starter is listed at 20 euros a month billed annually for 2,500 workflow executions, and the self-hosted community edition is free. A weekly decay workflow uses a handful of executions a month, so the plan limit is not the issue. The OpenAI or Claude cost of reading ten pages a week is small, typically a few dollars a month or less, but check your model's current rates.

A worked example

A local home services company has a blog with 60 posts and about 1,800 clicks a month. Comparing 28-day periods, a post called "how much does a new water heater cost" dropped from 140 clicks to 80. Its top query fell from position 4 to position 8. Reading the page, the cost figures were two years old and the page did not mention tankless units, which the top three results now cover. The owner pastes the page and the queries into Claude, gets five edits, spends 90 minutes updating prices and adding a tankless section, and logs the change. Four weeks later the page is back around position 5. That is an invented example to show the process, not a documented result, and recoveries are not guaranteed.

What it costs

Search Console is free. Google Sheets is free. An AI chat subscription runs about $20 a month if you do not already have one. n8n adds nothing unless you automate, and then it is free self-hosted or 20 euros a month on Cloud Starter annual billing. Your real cost is about 45 minutes a week.

What goes wrong

The biggest mistake is chasing noise. A page that dropped from 6 clicks to 3 did not decay. Another is rewriting pages whose loss was caused by something you cannot fix, such as a search feature that now answers the query on the results page. In that case you may be better off targeting a different query.

AI diagnosis can sound confident and be wrong. It only sees the text and numbers you gave it. Treat its edits as a checklist to verify, especially anything involving prices, laws or health claims.

Finally, Search Console data has a lag and a sampling limit, so do not read the last two or three days.

When not to do this

If your site has fewer than about 30 pages or under a few hundred clicks a month, your time is probably better spent publishing the pages you do not have yet. If your traffic is mostly branded searches, decay analysis tells you little. And if you do not have the authority to edit the pages, the routine produces a to-do list nobody acts on.

Frequently Asked Questions