Most advice about cold email subject lines is built on a broken measurement. It says a subject line "lifted open rate 34%" — but open rate is inferred from a tracking pixel that security scanners and Apple Mail Privacy Protection fire on their own. You can win an A/B test on opens and lose on replies. We wrote about that in detail in Cold Email Open Rates Are Lying to You.
So let's reframe the job. A cold email subject line has exactly two jobs: earn two seconds of attention from a real person, and avoid setting up an expectation the body can't pay off. It is not a deliverability lever in the way people think, and it is not where your reply rate is won or lost. But you can absolutely break both with a bad one.
{
h2
}
Do subject lines actually trigger spam filters?
Rarely on their own. Modern filtering at Gmail and Microsoft is dominated by sender reputation, authentication, engagement history, and recipient behavior. A perfect subject line will not rescue a domain with a poor reputation, and a mediocre one will not sink a healthy sender.
That said, content scoring still exists, and subject lines carry a measurable share of it. SpamAssassin — which many corporate gateways still run, alongside newer ML layers — has explicit subject rules. All-caps subjects, long runs of exclamation marks, unusual Unicode substitutions, and certain money/urgency phrasings each add fractions of a point toward the default 5.0 spam threshold. No single one of these gets you filtered. Three of them, plus a young domain and a link-heavy body, does.
The subject patterns that genuinely add risk:
- ALL CAPS, or a capitalized word for emphasis ("FREE audit", "URGENT")
- More than one exclamation or question mark, or mixed "?!"
- Currency symbols and numbers framed as claims: "$10k in 30 days", "3x your pipeline"
- Emoji in B2B cold email — fine in newsletters, a strong pattern match for bulk mail in a cold context
- Unicode lookalikes and zero-width characters used to dodge filters (this is detected and penalized harder than the word you were hiding)
- Fake "Re:" or "Fwd:" prefixes on a first-touch email with no prior thread
- The word "invoice", "payment", or "account suspended" in anything that isn't one — these hit phishing heuristics
The fake "Re:" deserves a separate note. It still gets recommended. It does measurably raise opens, because it exploits the recipient's sorting reflex. It also raises spam complaints, because the first thing a real person feels is that they were tricked. Complaints are one of the few signals that directly damages your domain reputation, and a complaint costs you far more than an open earns you. Don't do it.
Length: write for a 40-character window
On an iPhone in the Gmail app — where a large share of B2B recipients triage — you get roughly 35 to 40 characters of subject before truncation, and Gmail's preview then borrows the first line of your body. Desktop Outlook shows more, maybe 60 to 70.
Practical rule: keep it under 45 characters, and front-load the specific word. "Question about your Q1 hiring plans" survives truncation; "Following up on something I noticed about your team's hiring process" becomes "Following up on something I noticed…", which says nothing.
Short also signals human. Nobody writing a real one-to-one email produces a nine-word marketing headline. Three to six words reads like internal mail.
The patterns that hold up
These aren't magic phrases — copy them verbatim and you'll sound like every other sequence in that inbox. They're structures. The value comes from the specificity you fill them with.
| Pattern | Example | Why it works | Watch out for |
|---|---|---|---|
| Company reference | "Acme's onboarding flow" | Reads like internal context, not broadcast | Requires the body to actually reference it |
| Plain question | "Question about your SOC 2 timeline" | Low-pressure, obviously human | Vague questions feel like a script |
| Observation | "Noticed you're hiring 4 SDRs" | Proves research in five words | Only use a trigger that's real and recent |
| Peer reference | "How Ramp handles this" | Curiosity plus credibility | Name-dropping a client without permission |
| Topic label | "Payroll onboarding — 2 min" | Sets scope, promises brevity | The email must actually be 2 minutes |
| Mutual connection | "Sarah Chen suggested I reach out" | Highest-performing when true | Never fabricate this |
Notice what's absent: benefit claims. "Cut your CAC by 40%" is a headline, not a subject line. Headlines belong in ads where the reader expects to be sold to. In an inbox, the claim tells the reader instantly that this is a sales email, and they decide before opening.
Subject-body coherence matters more than the subject itself
The most expensive subject-line mistake isn't a spam word. It's a mismatch that produces an open followed by an immediate delete or a spam-button click.
If your subject says "Question about your SOC 2 timeline" and the first line of your body is "Hi {{first_name}}, I'm the founder of X, and we help companies like yours…", you've burned the trust the subject bought. Gmail watches what recipients do after opening — delete without reading, archive, mark as spam, or reply. Those behaviors feed the personalization models that decide where your next message lands.
So write the body first, then pull the subject out of it. If the subject has to reach beyond what's in the first two sentences, either the subject is overselling or the body is too vague.
Follow-up subject lines
For follow-ups inside the same sequence, you have three options:
- Reply in-thread (Gmail shows "Re: original subject"). Most common, keeps context, and threading means a second message doesn't compete for attention separately. Downside: if the first email was filtered, the whole thread is buried.
- New subject, new thread. Gives a genuinely filtered first email a second chance at the inbox. Best used on step 3, not step 2.
- Same thread, but a new angle in the body. Our preferred default for steps 2 and 3 — see How to Write a Cold Email Sequence for the angle rotation.
One thing never worth doing: "Bumping this", "Re: Re: quick follow-up", or a subject that's only about your own persistence. If the follow-up has nothing new to say, that's a content problem a subject line can't fix.
How to test without fooling yourself
Three rules that keep subject-line testing honest:
Measure replies, not opens. If you must use opens as a directional signal, use a metric that filters bot and scanner activity — otherwise you're A/B testing corporate security appliances.
Respect sample size. Detecting a 1-percentage-point difference in reply rate at a 5% baseline needs several thousand sends per variant. If your list is 600 people, you cannot statistically test subject lines. Pick a sensible one and spend your energy on the offer and the targeting, which move reply rate in whole percentage points rather than fractions.
Change one thing. Testing "Question about hiring" against "How Ramp cut time-to-hire 30%" tests two variables — length and framing — and teaches you nothing transferable. Test pattern against pattern, holding the body constant.
One more honest caveat: subject lines are the smallest lever on this list. Targeting is the biggest, offer is second, the first body line is third. If your reply rate is 0.4%, no subject line will get you to 4%. Compare against the numbers in our reply rate benchmarks before you start optimizing words.
A pre-send checklist
- Under 45 characters, and it still makes sense truncated at 35
- No caps-for-emphasis, no emoji, no more than one punctuation mark
- No claim, number, or currency symbol
- No fake Re:/Fwd: and no urgency language
- The first two sentences of the body deliver exactly what the subject implied
- Merge tags render correctly on a blank field — "Question about " with a missing company name is a dead giveaway
- Run the draft through a spam-score check before launch, not after
Where Sendvanta fits
Sendvanta scores every draft with SpamAssassin before you send, so you see the actual rules a message triggers — subject rules included — rather than guessing which words are "spam words." You can also run the finished draft through seed-mailbox inbox placement tests to see whether it lands in Primary, Promotions, or spam at Gmail and Microsoft. And because engagement classification is bot-aware, the open and click numbers you compare variants against exclude security scanners instead of inflating with them.
Everything here works on the free plan — $0 forever, no credit card, 1,000 active leads and 3,000 emails a month across up to 10 of your own SMTP mailboxes. Paid plans start at $29/mo. More on the spam scoring and placement testing at /features.
Do spam filters look at subject lines at all?
Yes, but with small weight. Content scoring engines like SpamAssassin have explicit subject rules for all-caps, excessive punctuation, and certain phishing-adjacent phrases, each worth a fraction of a point. They rarely filter a message alone, but they stack with a young domain, a link-heavy body, or weak authentication.
Should I personalize the subject line with the recipient's first name?
Usually no. First name in the subject is now a recognized mail-merge tell in B2B. Company name or a specific observation about their business works better, because it's harder to fake at scale and reads like real context.
How long should a cold email subject line be?
Under 45 characters, ideally three to six words. Mobile Gmail truncates around 35 to 40 characters, so front-load the specific word that makes the message relevant.
Is it better to follow up in the same thread or with a new subject?
Same thread for step 2 — it preserves context. Consider a new subject and thread for step 3, which gives a first email that was filtered a genuine second chance at the inbox.
Can I A/B test subject lines on a small list?
Not meaningfully. Detecting a one-point reply-rate difference needs thousands of sends per variant. On lists under a few thousand, pick a reasonable pattern and invest in targeting and offer instead.
Ready to send outbound that lands?
Create your free Sendvanta workspace — no credit card required.
Start free