Fix low-CTR pages with AI

A low-CTR page ranks on page one but gets far fewer clicks than its position should earn, usually because the title or description does not match what people searched. The prompt below finds those pages in your Search Console data, ranks them by missed clicks and drafts new titles and descriptions from their real queries.

The prompt

Paste it into Claude, ChatGPT or any client connected to the Google Search Console MCP server. Replace [your property] with your site.

Using Search Console data for [your property], find pages on page one of Google (average position 10 or better) whose click-through rate is well below what my site normally gets at that position, over the last 28 days. 1. List them with position, impressions, actual CTR, expected CTR and estimated missed clicks, sorted by missed clicks. 2. For the top 3, pull the queries each page ranks for, with impressions and position. 3. For each of those pages, write 2 new title options (60 characters or fewer) and 1 meta description (under 155 characters). Use the wording of the queries with the most impressions, put the main query near the start of the title, and do not promise anything the page does not deliver. 4. Tell me what the current snippet is probably getting wrong (wrong wording, wrong intent, or too vague). If a page has no query rows, say so and base the rewrite on the page's topic instead.

What the agent does

Tools it calls, in order: search_insights, search_analytics. About 2 to 4 data pulls per run (what counts as a data pull).

  1. Find the underperformers. The agent calls search_insights with the low_ctr recipe, by page. It compares each page-one result's CTR with what your own site gets at that position (it falls back to an industry curve when your data is thin) and keeps rows under half the expected rate, ranked by missed clicks. One data pull.
  2. Get the real queries. For the top pages, the agent calls search_analytics with dimension query and a page filter. These are the words people typed when they saw your result and did not click. One data pull per page.
  3. Draft the rewrite. The agent writes title and description options from those queries. If your client can fetch URLs, ask it to read the current title first so it can say what to change, not just what to write.

Expect about 2 to 4 data pulls per run. You can also run the recipe by query instead of by page, which shows the individual searches with low CTR; that is useful when one page ranks for several unrelated queries.

Example output

This is a real run on mcpsearchconsole.com, for 2 September to 30 September 2026, with the recipe set to pages and a 10-impression floor (at the default 50-impression floor by query, nothing qualified).

PageAvg positionImpressionsCTRExpected CTRMissed clicks
/blog/connect-google-search-console-to-grok-bot5.34061.7%about 6.5%about 19
/blog/connect-google-search-console-to-chatgpt5.8872.3%about 5.0%about 2

The Grok guide is the clear first target: position 5.3 on 406 impressions and only 7 clicks. Step 2 then hit a real limit. The search_analytics call for that page returned no query rows at all: Google had anonymized every query behind those impressions. So the agent could not rewrite from query wording and fell back to the page's topic, as the prompt allows.

The page's title at the time was "Connect Google Search Console to Grok and Grok Bot". With no query data, a reasonable change is to lead with the task and the result, for example "How to Connect Google Search Console to Grok (Grok Bot)", and to make the description say what the reader can do once connected. That is a suggestion to test, not a measured result.

The ChatGPT guide is close to its expected rate and only about 2 clicks short, so it goes to the bottom of the list.

How to act on it

Sort the list by missed clicks and work from the top. For each page, the result tells you which kind of fix it needs.

  • Queries match the page, title uses different words. Rewrite the title to start with the top query's wording. This is the most common case and the easiest win.
  • Queries show a different intent. If people search "how to" and your title reads like a product page, or they search "free" or "pricing" and the snippet does not say, rewrite for the intent, not just the keywords. If the page cannot satisfy that intent, the low CTR may be correct; leave it.
  • Branded or navigational queries dominate. Low CTR on someone else's brand name is expected. Skip these.
  • No query rows (anonymized). Rewrite from the page's topic and test, as in the example. Watch the page's CTR in the next run instead of query-level changes.
  • Missed clicks under about 5. Not worth an edit yet. Note it and re-check next month.

Change one title at a time where you can, record the date (ask the agent to save it as a note, which does not use a data pull), and compare CTR four weeks later. If a page is low in page one rather than low-CTR for its position, it belongs on the striking-distance list instead.

Variations

Try this promptRun the low-CTR check on [your property] by query instead of by page, and for each query tell me which page ranks and whether its title contains the query wording.
Try this promptFor [your property], find low-CTR pages on mobile only and check whether their titles are likely to be cut off on a phone screen.
Try this promptI changed the titles on 3 pages on [your property] four weeks ago. Compare their CTR and clicks for the last 28 days against the 28 days before, at similar positions.

Limits

  • You cannot see your own snippet. Search Console does not report the title Google actually displayed. Google rewrites many titles, so check the live results page before assuming yours is the one people saw.
  • SERP features change CTR. AI Overviews, featured snippets, videos and ads above you can cut clicks at the same position. Search Console does not tell you which features appeared, so a low CTR may not be your snippet's fault.
  • Expected CTR is an estimate. It comes from your own site's averages (or an industry curve when data is thin), not from Google.
  • Anonymized queries. As the example shows, a page can have hundreds of impressions and no query rows. You then rewrite with less evidence.
  • Small samples swing. On fewer than a few hundred impressions, a couple of clicks changes the CTR a lot. Compare over 28 days, not a week.

To try it on your site, follow the guide to use Search Console in Claude.

Questions

Related use cases

Run this on your own site in about a minute.

Connect Search Console free