channel-calibration.md

Channel calibration

The seven steps of the pass are constant. What changes by channel is which tells do the most damage and where the specificity has to land.

Feed posts

The hook carries everything. Line one must make the subject unmistakable, because confusion equals scroll and the feed truncates early.

AI drafts fail hooks two ways: the rhetorical question that hides the subject, and the clever-but-vague opener that could introduce any topic. Put the subject and the tension in line one, in plain words.

Hashtag piles read as automation and suppress reach past the first few. Use none or almost none.
The bolted-on question appended to force comments, "What do you think?" or "Agree?", now reduces reach rather than raising it. If the post deserves discussion, the discussion prompt is the post's actual stance.
The staircase format became the machine's default post shape and pattern-matches as AI on sight. Write real paragraphs with one earned single-line break.
Distribution follows dwell time, saves, and long comments, and generic content earns none of the three. Saves are worth several likes each. Target one save-worthy specific per post: a number, a template, a named mechanism the reader wants to keep.
Feeds reward accounts that stay in lane. Injecting specifics from the author's actual territory doubles as a topic-consistency pass.
Batch signatures. Posting the same structure at the same time across multiple accounts installs a shared pattern the feed prices in. Vary structure, timing, and format, and watch per-account baselines for correlated movement.

Cold email

The reader's question is always the same: did a human write this to me, specifically? Template smell answers it before the pitch begins.

Fake personalization is worse than none. A scraped first-line compliment, "Loved your recent post!", flags automation harder than a plain opener. Personalize with something that required actually reading: a specific claim they made, a number from their site, a hiring signal interpreted.
Kill the clichés on sight: "Hope this finds you well," "Just following up," "quick question" subject lines, "I know you're busy." Each one is a template fingerprint.
One CTA, low friction. Multiple asks read as a sequence step, because they are one.
Brevity is a human signal. Machines pad. Operators respect the reader's ten seconds. Under 100 words usually beats over 150.
Deliverability compounds the problem. Pattern-matched template copy at volume burns domain reputation alongside reply rates. Varying real specifics per recipient protects both.

Newsletters and long-form

Register mismatch is the dominant tell here. The punchy one-liner style that works in a feed post reads as a failure in an essay. Long-form demands narrative, stakes, and development, and AI defaults produce neither. A long piece assembled from feed-post rhythms is slop wearing a word count.

Earn the length. Every section must advance an argument or a story. The listicle skeleton with headers restating each other is the long-form staircase.
Personal narrative is the moat: real chronology, real decisions, real cost. A model cannot have been there, which is exactly why being there is the differentiator.
Keep the piece pointed at the reader's situation. Newsletter copy lives on "you." Slop drifts into third-person generalities about "businesses" and "teams."
The intro that summarizes and the conclusion that restates are both machine defaults. Open inside the story; end on the sharpest specific.
Signposting is legitimate in documentation. A reader scanning for a procedure benefits from knowing what a section covers. This is the one channel where that relaxes.

Case studies and proof content

The de-AI pass here is an evidence pass, because models fabricate by default when the source is thin.

Every number traces to source material or gets cut. No composite anecdotes, no rounded-up results, no invented timelines.
Run the defensible-claim test on every superlative. Would it survive a hostile reader with a search bar? Weaken claims until they are defensible. The piece gains credibility with every downgrade.
Anchors beat adjectives. One hard number with context outperforms a paragraph of "significant improvements."
Dates attach to every claim. These documents get forwarded and outlive the campaign that produced them. An undated "40% faster" becomes a liability the moment a prospect asks.
In regulated industries, healthcare, finance, and legal, a fabricated claim is a compliance incident, and this pass is non-negotiable.

Sales collateral and decks

Triads are structural to the format. Three columns, three bullets, three phases. Judge the copy inside the structure, not the structure.
Every number needs a date and a source, for the same forwarding reason as proof content.
Superlatives carry legal risk here, not just credibility risk. Cut anything the company could not support if asked.