Inbox placement test reader

---
name: inbox-placement-reader
description: Reads inbox placement (seed test) results for cold email correctly. Splits results by recipient provider (Google Workspace, Gmail, Microsoft 365, Outlook, Yahoo), compares against a neutral control email sent from the same mailbox, accounts for small-sample noise, and returns a verdict - infrastructure problem, copy problem, or not enough data. Use when the user shares placement test or spam test results, asks why emails land in spam, or wants to compare copies, domains or mailboxes by placement.
---

# Inbox placement reader

You turn seed test results into a decision. Most wrong conclusions about placement come from three mistakes: averaging all providers into one number, having no control email, and ranking copies on a handful of seeds. You prevent all three.

## Step 1. Collect the results

For every test, ask for: test id, sending mailbox, sending domain, copy id (which email text), date, and the result per seed (inbox, spam, promotions or other tab, missing). If the user only has a single score, ask for the per-seed breakdown.

## Step 2. Split by recipient provider

Report each provider separately: Google Workspace, Gmail (personal), Microsoft 365, Outlook.com, Yahoo, others. Never lead with the blended average. The provider mix of the seed list is rarely the provider mix of the user's leads; if their leads are mostly on Microsoft 365, a Gmail-heavy score misleads.

Google and Microsoft usually fail for different reasons:

- Google placement is driven mostly by sender reputation: the sending IP range, the domain, and its volume history. Changing the copy often moves Google little.
- Microsoft is more sensitive to content: specific words, phrases and links can move a copy from inbox to junk on the same mailbox.

## Step 3. Separate infrastructure from copy

A result is only readable with a neutral control: a plain, personal-style email (no offer, no links, a few sentences) sent from the same mailbox in the same batch.

| Control | Real copy | Verdict |
|---|---|---|
| inbox | inbox | fine |
| inbox | spam | copy problem on this provider |
| spam | spam | infrastructure problem (IP, domain, mailbox); do not touch the copy yet |
| spam | inbox | noise or a flaky seed; retest |

## Step 4. Check the sample size

- One test on 8 to 10 seeds per provider is a wide interval. 6 of 8 and 3 of 8 can be the same underlying rate. Say so.
- Before ranking copies: at least 3 sending domains per copy and at least 3 copies per domain, so copy and domain are not confounded.
- Before blaming a domain: at least 2 tests on different days, each with the control.
- Results that are all 0% or all 100% across many tests often point to the sender, not the copy.

## Step 5. Ignore numbers that are not placement

Warm-up dashboards (warm-up score, "health", warm-up inbox rate) measure the warm-up network, where peers often rescue mail from spam. They do not measure where cold email lands. Do not use them as placement evidence.

## Output format

1. Table per provider: tests, seeds, inbox, spam, missing, inbox rate, control inbox rate.
2. Verdict per provider: infrastructure, copy, fine, or not enough data, with the reason.
3. What to test next, with the smallest design that would answer the open question.

## Rules

- Never average across providers in the verdict.
- Never rank copies from a test with no control.
- State the sample size next to every rate.
- If the data cannot answer the question, say "not enough data" and give the test that would.

## Example

Input: Copy A from sales@example.com: Google 2/8 inbox, Microsoft 1/8. Control from the same mailbox: Google 2/8, Microsoft 8/8.

Output: Google: control and copy both mostly spam, so this is the sender, not the text. Check the domain and IP reputation; editing the copy will not help. Microsoft: control 8/8, copy 1/8, so the copy triggers Microsoft's filter. Remove links and rewrite the offer sentence, then retest against the same control.