← Vault Index
Source: business/marketing/campaigns/practice-builders/email/briefs/2026-08-11-transcript-not-summary.md

Brief — the summary can't be checked · email

Piper + Reid. Rebuilt 2026-08-11 after Kathryn: "there is NOT ENOUGH CONTEXT IN THIS BRIEF TO TELL ME THAT YOU GOT IT. THIS IS NOT READY." The first version argued a point. This one shows the email.

Channel: MailerLite broadcast Slot: in-the-wild Send date: 2026-08-11 Audience: Practice Builders — broadcast audience, segment 194625513368585544 (~456 at last read). Whole list minus Consulting Clients, TPC, Start|CIB, Start|LTO-52. The one idea: A summary of your call can't be checked against anything, so the recap you send from it can say something nobody said. The turn: They think the problem with AI summaries is quality. It's citation. A summary has no line you can trace, so an invented commitment and a real one look identical. The material: the six beats of the sent golden, mapped below. The ask: Reply with what you send a client after a call. Do not say: "tape" — her word is transcript; "on the tape" is internal team language and it reached her twice in the last draft · "walk through" · "honest" · "quiet/quietly" · "carry" · "in the room" · any vertical (CPA, tax season, TaxDome, Karbon) · "practice owner" as identity · twinning ("not X. It's Y.") · anything that ends the email on tooling. Sources: 7/27 Blueprint Mapping Session (Otter mo78KuQ7KOAYP93IxPrPNnod1Hc), read in full · aos-client-rc/qc/verify_claims.py, read 2026-08-10 · the sent golden the hour you can't bill (2026-07-06) via the campaign email STATUS.md.


The shape — built to the sent golden, not to a micro magnet

Correction, 2026-08-11. The previous version of this brief laid the email out as a seven-beat micro magnet with an opening exercise. That was wrong. A micro magnet is a multi-page guide that makes the reader do something — write 3 names, calculate the tax, here are the scripts. An email is 158 words. Importing one structure into the other would have produced a guide in an inbox.

The architecture is the sent golden — the hour you can't bill, 2026-07-06, 158 words:

Golden's beatThis send
Anonymized storyA recap that stated something nobody had said. Ours, one or two lines, no client in it.
The line that names the patternNobody caught it, because it read exactly like the lines around it. A summary gives you nothing to check a line against.
"The fix was one thing: …"The four-word rule. Every factual line carries a verbatim span of at least four words from the transcript, or it doesn't ship.
Turn to the readerThe summary from their last call is sitting in their inbox right now, and the recap they sent came out of it.
QuestionOne. What did you promise on that call that isn't in the summary?
Hit reply — I read every one.Kept as written.

158 words is the target, not a ceiling to grow into. One sentence per line, blank line between thought-groups, groups of one or two.

What comes out, versus the last draft: the opening exercise, the three-excuse block, and the sequence beat. All three are micro magnet furniture and none of them survive at this length. What they were carrying — the summary reads complete, so nothing looks missing — collapses into the pattern line.

What the story is. The internal instance stops being an optional strengthener and becomes the opening beat, because the golden opens on a story and this is the only one we have that costs nobody anything. If Kathryn cuts it, the send needs a different opening beat and this brief is short one — flag it rather than substituting a client.

Kathryn's words — the only quote in the send

"always follow up from your meeting with a recap of what you talked about. This is why having a transcript is going to be super important, not the summary that comes out of this AI tool. The summary is crap. You need the transcript, and you absolutely need a way to convert that transcript into a recap that sounds like you."

The mechanic underneath — for accuracy, not for the copy

verify_claims.py, live on every client recap:

Its own line: "an action/date is either traced to the room or disclosed as advisor-proposed. It can never be both un-sourced AND undisclosed."

Sloane does not explain the gate in the copy. Only the four-word rule reaches the reader, in the "The fix was one thing" beat. This section exists so nothing in the send overstates it.

The internal instance — it is now the opening beat

The golden opens on a story. This is the only story available that costs nobody anything, so it stops being an optional strengthener and becomes beat one. One or two lines, first person, ours, no client detail, and the word is transcript:

A recap we produced once stated something nobody had said — and nobody caught it, because it read exactly like the lines around it. That's why the four-word rule exists.

Proof rather than assertion; it puts us on the same side of the problem as the reader; it can't embarrass a client because the failure is entirely ours.

If Kathryn cuts it, the send loses its opening beat and this brief does not have a replacement. Flag it and come back to her — do not substitute a client story.

What would make it wrong: any version that reads as a confession, a war story, or a bid for credit.

The commercial risk — the send must not end on tooling

If it lands on "get a notetaker," the reader buys someone else's subscription and thinks about us less. The transcript is the input; the rule is the point. The turn-to-the-reader beat is what precedes the question for exactly that reason — it puts the reader's own last call in front of them, not a shopping list.

Why this slot

Last send (8/1) was teaching-story, so the rotation alternates to in-the-wild. It also does pipeline work: the live CIB ad sells paste raw material, get structured output to 400 people. This is the same mechanic on a different input, aimed at the biggest group on the list.


py .claude/skills/content-cascade/gate.py --brief <this file>

NOTION: https://app.notion.com/p/3b9a879ece9d812f8578d37414daf513

Confirmed by Kathryn: