Write the Opening Line

Outcome: an opening-line column driven by the evidence field, every line passing the interchangeability test, with a working abstain path and an automatic quality gate.

Surface: App
Level: Intermediate
Uses: AI copy column
Credits: ~1 per row
Prerequisite: Lesson 07's evidence column

The test that decides everything

Could this sentence have been sent to a different company?

If yes, it is not personalization. The prospect has received hundreds of merge-field openers and recognizes the shape instantly.

Every rule below exists to pass that test.

What the opening line is for

One job: prove the message was written for this person.

It is not the pitch, not the CTA, not the value proposition. Those are the same for everyone, which means a human should write them once, properly, rather than a model generating 500 slightly different versions.

So: one generated sentence, then human-written body. That is also a fifth of the cost.

The prompt


Write the opening line of a cold email to {{first_name}}, {{title}} at {{company}}.

Use only this evidence: {{evidence}}

Rules:
- One sentence, under 25 words.
- Reference the evidence specifically. Do not generalize it.
- Do not open with "I noticed", "I came across", "I saw that",
  "I was impressed by", or "As a fellow".
- Do not compliment the company.
- Do not mention our product or ask for a meeting.
- If the evidence is empty or vague, return exactly: SKIP

Return the sentence only.

“Do not compliment the company” is the instruction that most improves output. A compliment is the default register a model reaches for, it is what every automated email does, and it signals nothing except that you looked at their homepage. An observation with a consequence beats a compliment every time.

Good and bad, side by side

Evidence	Weak output	Strong output
Posted 3 SDR roles this month	”I noticed you’re growing your sales team!"	"Three SDR openings this month usually means someone is about to own onboarding for all three.”
Raised Series A in June	”Congrats on the funding!"	"Post-Series-A is usually when the spreadsheet the team ran on stops scaling.”
Added a competitor’s tool	”I see you use Acme."	"You added Acme in Q1 — most teams hit its reporting ceiling around month four.”

The pattern in the right column: state the fact, then say what it usually implies. That is the sentence that earns a reply, because it demonstrates you have seen the situation before.

The abstain path

Asked to personalize with nothing, a model does not fail — it invents. A compliment about a project that does not exist, growth inferred from a headcount number, a reference to a launch that never happened.

SKIP prevents it, and it must be checked: if no row returns SKIP, the abstain path is not working.

Rows that skip get one of two treatments:

Held out until a signal fires and evidence exists. Usually the right call.
A separate, lower-volume campaign with an honest generic opener and no fake personalization.

The quality gate

Four free formula columns before export:

Check	Rejects
8–30 words	Truncated output and paragraphs
No banned phrase	Prompt drift
Contains a distinctive token from the evidence	Lines that ignored their evidence
Not `SKIP`	Rows with no usable evidence

Then read 20 by hand. Formulas catch shape; only reading catches tone, and tone is what gets replies.

Do this now

Run on five rows and read aloud

Aloud matters — bad AI copy reads fine on screen and sounds wrong spoken.

Apply the interchangeability test

Rewrite the prompt, not the output, for any failure.

Add the “and what it implies” instruction

If the output is stating facts without consequence.

Build the four gate columns

Confirm SKIP fires

Test deliberately on an evidence-free row.

Widen to gated rows

Read 20 outputs before export

Check your work

Every line references a fact specific to that account
No compliments, no banned openers
SKIP appears in the output
The four gate columns pass on every exported row
You read 20 lines aloud

Where this breaks

Generating a line for every row regardless of evidence produces the worst messages in your campaign, not the most generic ones. Rows with nothing to say get the most confident-sounding invention, because the model fills the gap — and a prospect reading a reference to something that never happened will not reply to anything you send again. Gate on evidence before the copy column, not after.

Further automation

Regenerate when the evidence changes rather than on a schedule. A new funding round or a new job post produces new evidence, a new line, and a fresh reason to re-enter the queue — which is how a campaign becomes a pipeline.

Next lesson

09 — Conditional runs, making expensive columns fire only on rows that qualify.

Reference for this lesson: AI agents, SyncGTM 101 — writing copy, AI-Powered GTM, Outreach integrations.