Docs / Paid geographies
How the paid region report decides what to pause
Region by source combinations scored for pause or scale, with confidence bands and automatic window expansion when the data is thin. Web report, no MCP tool.
Signature
This instrument has no MCP tool. It runs on the web only, so you drive it from the page rather than from Claude or ChatGPT.
What you ask
Prompts land with the captured session, so that what is printed here is what actually worked rather than what should have.
What runs
Nothing fires over MCP for this one. The pipeline runs server-side when you start it from the web page.
What comes back
No session to show
This instrument has no MCP tool, so there is no chat session to capture. What it produces is the web report itself, described above.
When to reach for this
Use this before you touch a paid budget
It ranks region and paid-source combinations by how badly they underperform, or how strongly they beat, the benchmark for that source. The output is a pause list and a scale list rather than a dashboard, which is the shape a budget decision actually needs.
Not this one for organic
It looks only at paid traffic. A weak organic region will not appear here at all, and traffic mix is the instrument that covers the whole channel picture.
Not this one to explain why a region is bad
It tells you a geo and source combination is losing money at a given confidence. Whether that is the landing page, the offer, shipping, or a mismatched audience is a separate question and a different instrument.
How to read it
The insufficient-data block is the honest part of the report
Most geo and source rows in a normal account never accumulate enough conversions to say anything. Rather than hiding them or scoring them anyway, the report shows them in their own block and labels them insufficient. A tool that returns a confident recommendation for every row is a tool that is guessing on most of them.
The window expands when the data is thin, and it says so
It starts at 28 days and automatically widens to 56 or 90 when a combination does not clear the minimum. That means two rows in the same table can be measured over different windows, and the report prints which. A recommendation drawn from an expanded window is a slower signal: more reliable and less current.
Confidence is a separate column from the recommendation
A pause recommendation at low confidence and one at high confidence are different objects. Read the pair. Acting on a low-confidence pause is how a region that was merely unlucky for a month gets switched off permanently.
Zero scale candidates is a real answer
The scale side is often empty, and that is not a bug. It means no combination beat the benchmark by enough, over enough data, to justify moving money toward it. An empty scale list beside a populated pause list is an instruction to cut, not to reallocate.
Limits
- Paid traffic only. Organic, direct and referral are out of scope by design.
- It needs conversions per region and source combination, which most accounts do not have at region level. Expect the majority of rows to land in the insufficient-data block.
- Web report only: there is no MCP tool for this, so it cannot be driven from Claude or ChatGPT.
- It ranks and recommends. It does not execute: nothing here pauses a campaign for you.
- Recommendations follow your GA4 attribution. If paid conversions are being credited elsewhere, this reads that distortion as underperformance.
Related instruments
Go deeper
Try it on your own data
Every instrument runs on your own GA4, read-only, inside Claude or ChatGPT. One connector covers all of them.
Add the connector →Session captured 2026-08-03 against the ConvRadar demo property. Masked in the image: cropped to the conversation column, so the account sidebar and workspace name are out of frame, no connector URL is in frame, so no token can leak, no account email or avatar is in frame.