Subject lines — the first 35 characters and the open-rate cap.

The premise

A cold outbound campaign produces a sequence of compounding gates: delivered, opened, read, replied, meeting booked, opportunity created. Each gate has a conversion rate, and the cumulative product across gates is what the operator actually cares about. The subject line is the gate that sets the upper bound on every gate downstream of it. A 25% open rate caps reply at 25%, no matter how disciplined the body copy is. A 55% open rate raises that ceiling by a factor of more than two — and in practice the reply-rate lift across that band is closer to 2.5x, because the same recipient population that opens at higher rates also reads at higher rates.

The cost-of-being-wrong asymmetry is severe. The subject line is roughly 35 characters of visible text. The body is 60 to 120 words. The operator who spends an hour iterating on the body and five minutes on the subject has inverted the leverage ratio by an order of magnitude.

The 35-character mobile-preview constraint

The visible subject-line span on a phone-portrait inbox view is approximately 35 to 41 characters on the iOS default Mail client, 28 to 36 characters on the Gmail mobile app depending on device width and accessibility-text setting, and 25 to 30 characters on the Outlook mobile inbox preview. The conservative planning number is 35 characters. Beyond that span, the subject is truncated to an ellipsis and the trailing content is invisible to the recipient at the moment of the open decision.

The implication is front-loading. A subject of the form quick question about the Series B announcement last week reads as quick question about the… in the inbox preview — and the recipient evaluates the message against that fragment, not the full subject. The specific anchor (the Series B announcement) lands invisibly. A subject of the form Series B follow-on question places the anchor in the visible span and produces a higher open rate at a lower total character count.

On desktop clients the visible span is wider — 60 to 90 characters on most layouts — but the open decision on cold outbound is increasingly a mobile decision. The empirical mobile open share on B2B audiences sits between 55 and 70% depending on the segment. Optimizing for the desktop subject-line span optimizes for the minority case.

Subject-line patterns by empirical conversion

The patterns below appear, in observation across a meaningful volume of B2B cold outbound, with sufficient frequency and discriminating performance to be treated as a working taxonomy. Per-pattern open rates are stated as empirical ranges, not promises — the underlying segment, ICP, and infrastructure quality dominate the absolute numbers. The relative ordering, however, is stable across most segments tested.

• Ultra-short — 2 to 5 words, no punctuation. Empirical open range: 45 to 65%. The single best-performing category by absolute rate in most B2B segments.
• Internal-looking — subjects that resemble an internal forward, thread reply, or routing note. Empirical open range: 50 to 65%, with a credibility cost discussed below.
• Specific observation — naming a concrete, recent, recipient-company-specific event. Empirical open range: 40 to 55%.
• Recipient-named — first-name or company-name token in the subject. Empirical open range: 35 to 50%.
• Question — subject ends in a question mark. Empirical open range: 30 to 45%, with a steep failure-mode penalty for generic questions.
• Mutual-connection-named — naming a shared contact. Empirical open range: 50 to 60%, gated on the connection actually being shared.

The question-mark pattern

A question mark in the subject produces a 5 to 15% relative open-rate lift over the same subject phrased as a statement, in approximately 70% of the segments tested. The lift is mediated by the recipient processing the subject as a request for cognitive engagement rather than a broadcast statement.

The failure mode is generic questions. Quick question? produces the lift; Are you the right person? produces a measurable open-rate decrease, because the recipient has seen that exact phrasing in cold outbound for a decade and the pattern now signals automation rather than curiosity. The discriminating feature is whether the question, on its own, contains enough specificity that the recipient cannot answer it without opening the message. How are you handling the 12-day onboarding? contains that specificity; Do you have 15 minutes? does not.

The specific-observation pattern

A subject that names a concrete event in the recipient's company — a hire, a product launch, a funding round, a regulatory action, a public talk — produces an open rate that increases monotonically with the recency and specificity of the event. The empirical lift over a generic subject is 30 to 80%, gated on the event being one the recipient actually recognizes.

The cost is per-recipient research. A specific-observation subject cannot be templated across a list; each recipient requires either manual investigation or an automation pipeline that resolves the company-event to text. The per-recipient research cost is typically 2 to 5 minutes manual or 1 to 4 cents automated. The break-even calculation against generic subjects depends on the downstream reply-rate lift, which is typically 2 to 4x for the open-conditional reply rate — meaningful, but not always sufficient to justify the research cost at low-ACV segments.

The ultra-short pattern

Subjects of two to five words, typically lowercase, no punctuation, no emoji, no recipient token. Examples: quick thought, follow on idea, worth a look, your event last week. The empirical open rate across this category is the highest of any pattern in most segments tested.

The operational rule: if the subject exceeds five words, the operator has not finished editing. The mechanism is plausibly that ultra-short subjects resemble personal-correspondence subjects rather than commercial-correspondence subjects, and the recipient's spam-classification heuristic — the one running inside the recipient's head, not the one in their mail server — is dominated by length and formality cues.

The internal-looking pattern

Subjects that resemble routing notes, internal forwards, or thread continuations: re: pricing, fwd: intro thread, for Sarah, quick handoff. These produce the highest absolute open rates measured — 55 to 65% is routine, 70%+ is reached on some segments.

The credibility question is whether the open-rate lift survives the recipient's discovery, in the first three seconds of the body, that the message is not actually an internal forward. The reply rate conditional on open drops measurably on this pattern — typically to 40 to 60% of the reply rate of a more honest subject — because the recipient processes the bait-and-switch as a breach of the implicit contract. The cumulative reply rate (the product of open rate and reply-conditional-on-open) usually nets positive, but the per-touch goodwill cost compounds across a multi-touch sequence and the second and third touches suffer a steeper degradation than other patterns.

The all-lowercase pattern

A subject written entirely in lowercase — quick thought on your hiring page rather than Quick Thought on Your Hiring Page — produces a small but consistent open-rate lift of 3 to 8% in most segments. The mechanism is authenticity signaling: title-cased subjects are statistically associated with marketing automation, and the lowercase variant signals a human who is sending a casual message rather than a campaign operator pushing a polished template.

The cost is a slight reduction in perceived professionalism, which matters in some senior-enterprise segments and is irrelevant in most operator and IC-level segments. The pattern is dominantly positive below the VP level and is more ambiguous at the C-level.

Anti-patterns

• Question with emoji. The question pattern produces a lift; the emoji removes it and then some. The empirical penalty for any emoji in the subject line is a 15 to 35% open-rate decrease across B2B segments. The emoji signals a marketing tool or, increasingly, an LLM-generated subject — both of which the recipient now reflexively skips.
• All caps. Any subject with two or more all-caps words triggers both human and spam-filter skepticism. Open-rate penalty: 20 to 40%. Spam-folder probability also rises measurably.
• Exclamation points. Each exclamation point in the subject produces a roughly 10% open-rate decrease. Two or more is a near-categorical fail.
• Bracketed-ICP format. Subjects of the form [CTO] - [Company] - Quick Question are pure operator-tool signal. They read as un-customized template residue and produce open rates in the single digits.
• Spam-trigger words. Free, guaranteed, limited time, act now, click here, congratulations, winner, urgent. These words trigger both spam classifiers and the recipient's skim filter. Open-rate penalty: 25 to 50%, plus elevated spam-folder placement that does not appear in the open-rate denominator at all.
• False urgency. Subjects implying time pressure that does not actually exist — last chance, final follow-up, closing tomorrow — produce a small initial open lift and a steep reply-rate collapse once the recipient discovers the urgency was manufactured. The cumulative effect is negative.

The AI-generated-subject detection problem

Recipients have, in 2025 and into 2026, developed a reliable pattern-match for subjects generated by an LLM with no human edit pass. The detected patterns include: a verb-noun construction with a generic-positive adjective (Streamlining your outbound process), a colon-separated value-prop format (Outbound at scale: a quick idea for [Company]), and the formal-cased question with a recipient-token (A question for you about your growth strategy, Sarah).

Open rates for detected-AI subjects have been declining at roughly 2 to 4% per quarter since mid-2024 as recipient detection accuracy improves. The current open-rate penalty for the most obvious AI patterns is 20 to 35% relative to a human-edited equivalent. The operational implication is that LLM-generated subjects are usable, but require a human edit pass that removes the verb-noun-adjective structure and reduces the length.

Subject-body coherence

The subject creates an expectation. The body must satisfy it within the first sentence. A subject of quick question about Tuesday's launch followed by a body that opens with a value-proposition statement and never mentions Tuesday's launch produces a measurable reply-rate collapse — the recipient processes the mismatch as a bait-and-switch and closes the message.

The failure mode is clickbait subjects optimized in isolation for open rate. A subject can be tested in an A/B harness and selected on open rate alone, and the operator can ship a subject-body pair where the subject won the test but the pair's reply rate is below the prior baseline. The selection metric for subject-line testing should be reply rate, not open rate. Open-rate selection is the dominant operator error in subject-line A/B testing.

The reply-touch subject pattern

Subsequent touches in a multi-touch sequence have two options. The first is to continue the existing thread — same subject, no re: prefix change, the message lands inside the existing conversation thread in the recipient's inbox. The second is to open a new conversation with a new subject. Each option produces a different open-rate signature.

Thread continuation produces open rates that are typically 5 to 15% higher on touches 2 through 4 than new-subject variants in the same sequence, because the recipient processes the new message as an active conversation rather than a cold message. The mechanism breaks down on touches 5 and beyond — by that point the thread is itself a signal of automated persistence and produces an opt-out response at elevated rates.

New-subject touches produce a lower per-touch open rate but a higher per-touch reply rate conditional on open, because the recipient who opens a new subject is making a deliberate decision rather than a reflexive inbox sweep. The compounding choice across a 5-touch sequence is typically: thread-continuation on touches 2 and 3, new-subject on touch 4, and a final new-subject break-up touch on touch 5.

A/B testing methodology

Subject-line A/B testing requires sample sizes that the median operator does not respect. The minimum-detectable-effect at typical cold-outbound open rates (25 to 45%) for a 5-percentage-point lift at 95% confidence is approximately 500 sends per variant — that is the threshold below which observed differences are statistically indistinguishable from sampling noise. Operators routinely declare winners off 50 to 100 sends and ship subject changes that are, on the underlying distribution, neutral or negative.

The operational pattern: run two variants concurrently, balanced across the sending pool, for a sample of at least 500 per variant; require a 5-percentage-point reply-rate (not open-rate) lift to declare a winner; treat anything below 500 sends as a directional read, not a decision. A two-variant test on a 100-recipient day will not produce a usable signal; it will produce a number that feels like a signal.

Common operator failures observed in production

• Optimizing the subject in isolation from the body. The subject is tested on open rate, the winner is shipped, and the campaign's reply rate is unchanged or worse. The selection metric must be reply rate.
• Exceeding 35 characters with the anchor at the end. The specific token that would have produced the open is invisible in the mobile preview.
• Title-casing every subject. The default of a marketing tool. A measurable open-rate cost relative to lowercase variants in most segments.
• Re-using one subject across all touches. The recipient sees the same subject five times in two weeks and the pattern itself becomes the signal of automation.
• Declaring A/B winners on under-powered samples. The 80-recipient test produces a 7-point swing that is statistical noise. The operator ships the change. The next test reverses the swing. The operator concludes that subject lines are unpredictable.
• Generating the subject with a model and skipping the human edit. The verb-noun-adjective structure is preserved, the recipient detects it within 1 second of inbox scan, and the open rate drops 20 to 35%.

Pre-send subject-line checklist

• Visible content within the first 35 characters carries the anchor.
• No emoji. No exclamation point. No all-caps word.
• No bracketed-ICP residue, no generic quick question with no specificity, no are you the right person.
• Lowercase variant tested against title-cased variant in segments above the VP level.
• Subject-body coherence verified — the first sentence of the body addresses what the subject promised.
• For touches 2 and 3 of a sequence, thread continuation as default; for touches 4 and 5, new subject.
• For any A/B test, at least 500 sends per variant and reply rate as the selection metric.
• If the subject was generated by a model, a human edit pass has removed the obvious verb-noun-adjective structure and reduced length to under five words where possible.

Where this fits

The subject line determines the open rate, and the open rate caps every downstream metric — reply, meeting booked, opportunity created, revenue. The next chapter handles the opening line, which determines whether the open survives the first 8 to 15 seconds of recipient attention; the chapter after that handles the value proposition, which determines whether the survived attention converts to a reply. The subject is the gate that opens the system. Every downstream optimization is gated on it, and an hour spent iterating on subject-line patterns has, in observation, the highest reply-rate-per-hour-of-operator-effort of any single intervention in the cold-copy stack.