Step 1: Export a comparison from Search Console
- Open Search Console and pick the property.
- Go to Performance, then Search results.
- Click More next to the date buttons, open the Compare tab and choose Compare last 28 days to previous period. If your traffic is seasonal, choose Compare last 28 days year over year instead.
- Turn on all four metrics at the top: clicks, impressions, CTR and position.
- Click Export (top right) and choose Download CSV, Download Excel or Google Sheets. The export holds each table of the report, including Queries and Pages.
Two limits to know
The Search Console screen and its export show at most 1,000 rows per table, so a big site only gets its top 1,000 pages and queries. And some queries are hidden by Google for privacy, so query clicks never add up to page clicks. Both are Google limits, not a problem with your site.
Pick a range that ended a few days ago. Search Console numbers are not final until about two to three days after each day, so the last few days always look like a drop.
Step 2: Give Claude the files and this prompt
Open a new chat in Claude, attach the Pages and Queries tables (as files, or as one Excel file), and paste the prompt below. It asks for the three causes that the numbers can actually tell apart, and stops Claude from inventing reasons the data cannot show.
I have attached two Google Search Console exports for the same site,
both comparing the last 28 days with the previous 28 days:
1. Pages: clicks, impressions, CTR and position per page, for both periods
2. Queries: the same for each search query
Do this:
1. List the 20 pages that lost the most clicks. For each: clicks before and
after, the change, and the change in impressions, CTR and position.
2. For each page, say which of these explains most of the loss:
a) impressions fell (fewer searches, or the page ranks for fewer queries)
b) position fell (it ranks lower)
c) CTR fell at a similar position (something on the results page changed)
3. List the queries that lost the most clicks and the page each one most
likely belongs to. Mark it "likely" unless the data proves it.
4. Give me the 5 pages to fix first, with one concrete next step each.
Use only the numbers in the files. Do not guess at causes the data cannot
show. Show every table with its date range.The same prompt works in ChatGPT. If your export has more than one comparison column per metric, tell Claude which column is the current period and which is the previous one.
Step 3: Read each drop by its cause
Clicks are impressions times click-through rate, so every drop comes from one of the two, and position usually explains which. This is how to act on each pattern.
- Impressions down, position about the same: fewer people searched, or the page stopped showing for some queries. Check the year before for seasonality, then check which queries disappeared for that page.
- Position down: the page ranks lower. Look at who now ranks above you for its top queries, and update the page so it answers the query better than they do.
- CTR down, position about the same: the results page changed. An AI Overview, ads or other features may now sit above you, or your title stopped matching the query. Rewrite the title and description for the main query.
- One page down, another page up for the same queries: two of your pages compete for one query. Merge them or make each target a different query.
- Clicks to near zero overnight: check the page is still indexed with URL Inspection. A noindex, a redirect or a changed canonical often does this.
To see which queries one page lost, go back to Search Console, add a filter Page equals that URL, and open the Queries tab with the same comparison. Give that export to Claude for the page you are fixing.
Step 4: Find when the drop started
The date matters because it points to the cause: a site change, a Google update, or a competitor. In Search Console, open the Days table from the same export and ask Claude for the first day the daily clicks stayed clearly below the average of the weeks before. Then compare that day with your own change log and with Google's published update dates.
The faster way: let Claude read Search Console directly
Exports work, but you repeat them every time, they stop at 1,000 rows and they do not join with GA4. With a Search Console MCP connector, Claude reads the data itself. See the GA4 and Search Console MCP servers compared for the free and hosted options.
We built Site Insights by KKV Labs for this question. Its traffic drops tool returns the pages and queries that lost the most clicks, the day each page's drop started with its daily average before and after, and the landing pages that lost GA4 sessions, compared with the previous period or the same dates a year earlier. Connecting it to Claude takes about two minutes and it is free for one site. Then ask:
- Which pages on example.com lost the most Google clicks in the last 28 days, and when did each drop start?
- For the top 3, did they lose impressions, position or CTR?
- Is /pricing still indexed? Which canonical did Google pick?
Method. The Search Console steps follow Google's Performance report help and were checked against the live Search Console on 19 November 2026 (date compare presets, the 1,000 row tables and the export options). The Site Insights tool description is from its own code on the same day. The prompt is ours; it works on any Search Console export with the columns described.