Worked examples
Three passes end to end, plus one case where the flagged construction was correct and stayed.
Read these for the method, not the wording. Each one names the tells first, then rewrites, in that order. The diagnosis is the transferable part.
Example 1: LinkedIn post
Before
Most founders think outbound is dead. It isn't. It's just that the old playbook stopped working.
Here's what we're seeing:
- Reply rates are down across the board
- Personalization at scale is harder than ever
- Buyers are more skeptical than they've ever been
The teams winning right now are doing three things differently: they're narrowing their list, they're leading with a real observation, and they're following up with patience.
The bottom line? Outbound still works. You just have to earn the reply.
Diagnosis. Contrast flip in the opening. Two triads, one in the bullets and one in the "three things" sentence. Uniform rhythm throughout. Summary close. Every claim is a quantifier with no number: "down across the board," "more skeptical than ever."
After
Our reply rate on cold outbound went from 4.1% to 1.2% between January and September. Same list source, same sender, same offer.
What changed on our side, in order of impact: we cut the list from 4,000 to 600 accounts, and we stopped sending anything that didn't reference something specific to the account.
Rate recovered to 3.4% over the next six weeks. Volume dropped 85%. Meetings booked went up.
The list was the problem. It usually is.
Structure went first: the flip is gone, the bullets are gone, the two triads are gone. Then specifics replaced every quantifier. The rhythm ends on a short line because the section ran long, not as a formula.
Example 2: Cold email
Before
Hi Sarah,
I noticed that Acme has been scaling rapidly — congratulations on the recent growth! Many companies at your stage struggle with fragmented data across their go-to-market tools.
We help revenue teams unify their data, streamline their workflows, and drive measurable pipeline growth.
Would you be open to a quick 15-minute chat?
Diagnosis. Em dash the sender does not use. "Many companies at your stage" is a quantifier with no number and no source. The triad in the value sentence. Register vocabulary: unify, streamline, drive. Nothing traceable to Acme beyond a generic growth compliment.
After
Hi Sarah,
Saw Acme posted four RevOps roles in the last two months, all reporting into different teams.
That split usually means the pipeline number in the board deck doesn't match the number in the CRM. If that's happening, I can show you how two other teams fixed the reconciliation in about a week.
Worth 15 minutes?
Cold email has the least tolerance for slop of any channel. The observation is traceable, the claim is bounded, and the close is one line. See channel-calibration.md.
Example 3: Blog section
Before
Why attribution breaks
In this section, we'll explore the three main reasons attribution models fail at scale. Attribution is not a technology problem. It is a data problem.
First, tracking is inconsistent across channels. Second, offline touchpoints are invisible. Third, the models themselves are built on assumptions that rarely hold.
Understanding these failures is essential for any team looking to build a robust measurement practice.
Diagnosis. Over-signposting in the first sentence. Contrast flip in the second. Triad, announced and then delivered. Written-out "is not." Closer that restates without adding. No specific anywhere.
After
Why attribution breaks
Attribution fails on data quality long before the model matters.
Tracking drifts first. We audited 12 campaigns last quarter and found UTM parameters missing or malformed on 40% of paid social links, which meant those sessions landed in direct traffic.
Offline touchpoints never show up at all. A conference conversation that starts a deal appears in the model as an organic search visit three weeks later, if it appears.
Then the model assumptions. Most last-touch setups assume a single decision-maker, and enterprise deals average five.
Fix the inputs first. The model choice is a rounding error next to a 40% tagging gap.
The three reasons survived, because there really were three. What changed is that they are no longer announced, no longer parallel in construction, and each one now carries a number traced to the audit.
Example 4: The edge case, where the tell stays
Draft opening
Ever notice how the best reps never sound like they're selling?
The rhetorical question opener is on the inventory. The pass flagged it and the flag was wrong.
The author's fingerprint documents that 9 of their last 15 posts open with a question, and those posts outperform their others on engagement. This is a real habit, visible in genuine samples predating any AI use, and their audience recognizes it.
Resolution: the opener stays. The rest of the piece still gets the full pass.
This is the fingerprint overriding the tell list, and it is the most common way a careless pass damages a draft. The tell list describes the average machine draft. It does not describe this person. When a flagged construction appears repeatedly in the author's genuine, pre-AI work, it is voice, and removing it makes the piece less like them, not more.
How to tell the difference: check whether the construction appears in the author's writing from before they used AI assistance. If yes, it is voice. If it appears only in AI-assisted drafts, it is slop.