A cold-email subject line should be judged by whether it accurately frames the email's first decision, because curiosity that the body cannot cash out creates the wrong kind of open. Start with the evidence that makes the route and recipient appropriate, expose the assumptions behind the wording, and stop when the decision cannot be reviewed.
Start with the body’s first decision
Write the body before the subject. Underline the first decision the recipient can make: recognize a workflow, route ownership, inspect bounded proof, or decline. The subject should preview that decision with the smallest accurate phrase. It is not a miniature benefit claim and should not create curiosity the body must later repair.
- Body decision: confirm distributor-routing relevance.
- Candidate: “Distributor routing in Germany?”
- Body decision: identify the onboarding owner.
- Candidate: “Owner for supplier onboarding?”
- Body decision: inspect a fixed three-check artifact.
- Candidate: “Three duplicate account checks.”
- Draft preview text alongside the subject because many inboxes display them together; ensure the preview continues rather than contradicts the stated decision.
- Reject a line that becomes misleading when the recipient sees only the sender name, subject, and first body sentence.
A compression test
If removing a word changes the expected topic or action, keep it. If a word adds status, urgency, or intimacy that the body does not substantiate, remove it. For a compression test, archive the exact line and body, promise classification, cohort assignment, delivered count, reply disposition, and retest trigger before calling a winner.
Keep the promise accurate and contextual
For U.S. commercial email, FTC guidance requires the subject to reflect the message accurately. ICO guidance adds data-planning, fairness, transparency, and objection considerations for relevant United Kingdom contexts. These sources do not say one phrase improves response. They set boundaries around a subject attached to an actual contact decision.
- Do not fake “Re:” or “Fwd:”.
- Do not claim a prior conversation that did not occur.
- Do not promise a result absent from the body.
- Do not use a personal detail merely to prove it was collected.
- Keep the commercial purpose intelligible.
- Apply objections and opt-outs before testing variants.
- Keep an evidence note showing which exact body sentence fulfills the subject promise; missing fulfillment removes the line before testing.
- Review whether punctuation or capitalization adds urgency the sourced event does not support.
Accuracy is recipient-facing
The test is what a reasonable recipient expects after reading the line and whether the opening fulfills that expectation, not whether the sender can defend a creative interpretation. For accuracy is recipient-facing, archive the exact line and body, promise classification, cohort assignment, delivered count, reply disposition, and retest trigger before calling a winner.
Know what the subject cannot fix
A subject cannot repair an irrelevant account, invalid address, missing authentication, complaint pattern, or broken opt-out path. Google distinguishes baseline sender expectations from additional requirements for certain higher-volume senders. Diagnose those systems from their own evidence before treating wording as the constraint.
- Confirm sender identity and applicable requirements.
- Use a validated eligible-recipient cohort.
- Keep list and sender conditions stable.
- Separate provider rejection from human nonresponse.
- Pause after authentication or complaint movement.
- Do not call an open-rate change deliverability proof.
- Verify that OKKI Go or any other drafting interface does not cause a subject edit to bypass recipient, body, or suppression review.
- Record sender-health changes outside the copy experiment so they cannot be misattributed to wording.
The no-copy condition
When the sender or contact fails its gate, there is no subject-line optimization task. The correct output is a stopped send and a named repair owner. For the no-copy condition, archive the exact line and body, promise classification, cohort assignment, delivered count, reply disposition, and retest trigger before calling a winner.
Compare complete subject-and-body pairs
Pair A uses “Distributor routing in Germany?” and immediately cites the verified expansion before asking about routing. Pair B uses “Quick question” with the same body. Pair A makes the topic and decision visible; Pair B withholds them. A second comparison can test “Three duplicate account checks” against “Improve efficiency,” again keeping the body promise fixed.
- Change one subject dimension at a time.
- Keep the body, audience, offer, sender, and timing stable.
- Randomly assign when the eligible cohort supports it.
- Use delivered recipients as the denominator.
- Classify qualified replies, referrals, objections, and opt-outs.
- Review expectation mismatch, not opens alone.
- Include a routing subject and a proof-offer subject only in separate decision cohorts; combining them would test different asks rather than language.
- Ask a blinded reviewer whether each reply actually addresses the subject’s promised topic.
Editorial judgment comes before statistics
An inaccurate or deceptive line is rejected before testing. The experiment chooses among defensible candidates; it does not license every phrase that might attract attention. For editorial judgment comes before statistics, archive the exact line and body, promise classification, cohort assignment, delivered count, reply disposition, and retest trigger before calling a winner.
Promote a winner with a stopping rule
Document the cohort, assignment, exact pairs, dates, delivery counts, reply dispositions, and decision threshold. Promote a line only when it accurately frames the message and improves the chosen downstream outcome without worsening guardrails. Repeat in a new cohort before generalizing.
- Archive the losing line and its result.
- Show raw counts beside every rate.
- Keep opens and clicks as diagnostics.
- Return role or company corrections upstream.
- Use OKKI Go subject variants under human confirmation.
- Retest after audience, offer, sender, or provider changes.
- Keep the decision memo with the complete body so a future editor cannot reuse the winning words for a message that makes a different promise.
- Document the editorial veto separately from the statistical result. A line may lose before testing because it creates false familiarity or an unsupported promise; another may remain acceptable yet lose the cohort comparison. Keeping those reasons separate prevents performance data from laundering an inaccurate subject into future campaigns.
- Require the candidate to win on the selected downstream disposition and remain acceptable on objections, opt-outs, and expectation mismatch.
- Expire the result when the body, offer, audience, sender identity, or provider environment changes materially.
The final subject-line rule
Choose the line whose smallest truthful promise lets the intended recipient understand the message’s first decision. If only open activity improves, the test has not yet shown better cold-email performance. For the final subject-line rule, archive the exact line and body, promise classification, cohort assignment, delivered count, reply disposition, and retest trigger before calling a winner.
Take one planned subject line for cold email message and annotate its recipient evidence, promise, owner, next action, and stop rule. If any field is missing, pause the send and repair the decision before polishing the copy.
Frequently asked questions
Start with the body’s first decision?
Write the body before the subject. Underline the first decision the recipient can make: recognize a workflow, route ownership, inspect bounded proof, or decline. The subject should preview that decision with the smallest accurate phrase. It is not a miniature benefit claim and should not create curiosity the body must later repair.
Keep the promise accurate and contextual?
For U.S. commercial email, FTC guidance requires the subject to reflect the message accurately. ICO guidance adds data-planning, fairness, transparency, and objection considerations for relevant United Kingdom contexts. These sources do not say one phrase improves response. They set boundaries around a subject attached to an actual contact decision.
Compare complete subject-and-body pairs?
Pair A uses “Distributor routing in Germany?” and immediately cites the verified expansion before asking about routing. Pair B uses “Quick question” with the same body. Pair A makes the topic and decision visible; Pair B withholds them. A second comparison can test “Three duplicate account checks” against “Improve efficiency,” again keeping the body promise fixed.
Promote a winner with a stopping rule?
Document the cohort, assignment, exact pairs, dates, delivery counts, reply dispositions, and decision threshold. Promote a line only when it accurately frames the message and improves the chosen downstream outcome without worsening guardrails. Repeat in a new cohort before generalizing.
