Cold Email Subject Lines: What the Evidence Shows and How Much to Trust It

By Laurel · Updated Oct 8, 2026

A cold email subject line is the one line a stranger reads before deciding whether your message deserves attention. The research on what makes one work is thinner than the advice implies. Most studies are vendors analyzing their own users, they contradict each other, and the metric they rely on, open rate, stopped being reliable in 2021.

That doesn't make subject lines unimportant. It means a copied benchmark is a poor guide. Below: which findings have some support, what Apple's privacy change did to open tracking, what to measure instead, and a table of subject line types.

Key Facts: Cold Email Subject Lines

• Apple says Mail Privacy Protection stops senders from using invisible pixels to collect information and from knowing when a user opens an email (Apple Newsroom, June 2021).
• Litmus put Apple's mail clients at 62.26% of tracked opens in July 2026, from over 1 billion opens in its own Email Analytics.
• Belkins reports 46% opens for personalized subject lines versus 35% without, and 7% replies versus 3%, from 5.5 million emails sent for its own clients in 2024 (vendor data).
• A field experiment in Marketing Science found the recipient's name in the subject raised opens from about 9% to about 11%. A 2023 follow-up did not reproduce the first-name effect.
• Gong analyzed 28M+ cold emails and found buzzwords and numbers cut open rates by up to 17.9% (vendor data, July 2025).
• Instantly advises at least 250 contacts per variant, judged on positive reply rate rather than opens.
• The FTC says a subject line must accurately reflect the content of the message, with penalties of up to $53,088 per violating email.

What a Subject Line Has to Do

The job is simple: get the right person to open an email that is relevant to them. The constraint is that the email has to deliver what the subject promises. A line that over-promises can buy an open and cost you the reply. How to write a cold email covers the body that has to deliver on it.

That's why optimizing for opens misleads. A line that lifts opens but lowers positive replies is a worse line, and one that earns fewer opens from the wrong people is a better one. The cold email strategy framework covers the rest of that stack, and outbound lead generation puts email in the context of the whole motion, as does the outbound sales playbook.

What the Evidence Says, and Who Produced It

Start With the Source

Public subject line numbers come from vendors analyzing their own customers' sends, email platforms analyzing newsletters, and a few academic field experiments. The table lists the main ones.

Five of the seven are vendors or publishers describing their own users' sends. Only the two academic studies ran controlled experiments. Observational data can't separate the subject line from the list, the offer and the sender behind it.

Length: Five Sources, Four Answers

Boomerang found that 3 to 4 words drew the most replies, with rates falling slowly as words were added. Belkins reports 46% opens for 2 to 4 words, 38% for one word and 34% at ten. Hunter says 5 to 6 words is the sweet spot for average reply rates. Gong's executive analysis found opens fall as lines get longer, with 1 to 4 words best. Backlinko's link-building study measured characters and favored 36 to 50.

The measures differ (replies versus opens, words versus characters) and so do the audiences. The consistent message is "short and specific," and none of these data can tell you whether two words beat five. Boomerang also found that only 14% of messages with no subject line got a response, so don't leave it blank.

Personalization: Best Tested, Least Settled

The strongest evidence is an academic field experiment. Researchers put the recipient's name in the subject line of half the emails in each campaign. For a test-prep company, opens rose from about 9% to about 11%, leads rose roughly 30%, and unsubscribes fell 17%. A later field experiment did not reproduce the first-name effect, though a surname did lift opens and clicks. The authors suggest recipients may have grown used to the tactic.

Vendor data points the same way, with the usual caveats. Belkins reports 46% opens with personalized subject lines versus 35% without. Backlinko's link-building study found personalized subject lines raised response rates by 30.5%. Hunter reports 35.4% opens with one custom attribute and 40.2% with two.

Put together, personalization probably helps, and its cheapest version, the first-name merge, looks worn out. What survives is relevance: a line that shows the sender knows what the recipient works on. That depends on clean data behind the merge fields, which is what lead data enrichment is for. Personalization at scale takes the same question into the body of the email.

Questions, Numbers and Lowercase

These three have the thinnest evidence and the most confident advice.

• Questions. Belkins gives 46% opens for question-framed lines in one place and 44% in another. Hunter found that subject lines using "Quick question" drew 2.5% replies and 28.7% opens, against 2.9% and 32.9% for other lines.
• Numbers. Belkins lists numbers at 44% opens in one chart, then reports 27% with numbers against 28% without. Gong says buzzwords and numbers together cut opens by up to 17.9%.
• Casing. Belkins reports all caps at 30%, title case and lowercase at 29%, and capitalizing only the first letter at 25%, and calls the all-caps lead a 1 to 2 point edge.

The same Belkins post shows 46% opens in some cuts and 25% to 30% in others, without explaining the different baselines. A gap of a few points, on sample sizes nobody reported, isn't a finding. And nobody should write in all caps because a table put it first.

Why Open Rates Stopped Meaning What They Used To

What Apple Changed, in Apple's Words

On June 7, 2021, Apple announced that in the iOS 15 Mail app, Mail Privacy Protection stops senders from using invisible pixels to collect information, helps users prevent senders from knowing when they open an email, and masks their IP address. Apple's current privacy documentation, where the setting is called Protect Mail Activity, adds the mechanism: Mail downloads remote content in the background when a message arrives, "regardless of whether you engage with the email." The content passes through two relays run by different entities, so neither sees both the IP address and the content.

Open tracking works by placing a one-pixel image in the email and logging when it loads. For anyone using the setting, the image loads at delivery, so a logged open only says the message reached an Apple Mail inbox. Apple doesn't describe this in terms of sales email.

How Much of Your Data Is Affected

Litmus put Apple's mail clients at 62.26% of tracked opens in July 2026, from more than a billion opens in its own Email Analytics. Those are Litmus customers' sends, so the share on a B2B prospect list will differ. Check the Apple share in your own sending tool.

Apple isn't the only source of false opens. Gong's own analysis, which is built on open rates, concedes that some recorded opens may be the prospect's spam filters. Hunter's data shows sequences with open tracking enabled averaging a 4.4% reply rate against 7.4% without. The report doesn't prove that tracking caused the gap, since senders who switch it off may differ in other ways. Hunter's own takeaway is that opens remain a rough lagging indicator of deliverability.

What to Measure Instead

Judge a subject line by what happens after the open. Outbound metrics and benchmarks lays out the full chain from delivered email to pipeline.

1. Positive reply rate. Replies that show interest or ask a real question, divided by delivered emails. Instantly's A/B testing guide says a valid test measures positive reply rate, not just opens and calls open rate a directional signal.
2. Meetings per 100 contacts. Slower, but harder to fake. The BDR/SDR metrics and quota math guide shows how the funnel ratios connect.
3. Opens without replies. Instantly calls high opens with low replies the "clickbait signal". It's a prompt to inspect the line, not a result.
4. Spam complaints. Google's sender guidelines ask for a spam rate under 0.30% in Postmaster Tools, with 0.10% as the target. A bait line is one way to push complaints up.

If your sequencing tool reports opens but not outcomes, log replies and booked meetings against the lead in a CRM so the test reads on pipeline. Rework's Sales Ops CRM is built for that view, and most CRMs can do it.

Testing Without Fooling Yourself

Instantly's guide sets the floor at at least 250 contacts per variant, with 500 or more for high confidence. Treat that as a floor. At a 3% reply rate, 250 contacts produce about seven replies. If variant B gets twelve (4.8% against 2.8%), that looks like a 70% lift, but a standard two-proportion test puts the chance of a gap that size from luck alone at roughly one in four. Standard sample-size formulas put the requirement for detecting 3% versus 6% (80% power, 5% significance) at about 745 contacts per variant. That's why small tests mislead.

Rules that keep a test honest:

• Change one thing per test: the subject line, nothing else.
• Split one list at random and send in the same window from the same mailboxes.
• Judge on positive replies, then rerun the winner on a fresh list before adopting it.
• Keep a log. After a few rounds the log is a benchmark for your market, which no vendor report can give you.

The cold email cadences guide has working subject line and template examples across follow-up steps. The sales engagement platform roundup and the Instantly, Lemlist and Smartlead comparison cover the tools that run sequences.

Patterns That Tend to Hold Up

None of these is a proven winner. They hold up because the mechanism is sound and the weak data above doesn't contradict them.

• Plain and specific. Name the topic the recipient owns: "Outbound hiring plan" rather than "Opportunity." Gong's executive analysis favored short topic lines, with examples like University achievement gaps and Student dropout risk.
• Tied to a real trigger. A funding round, a new hire or a posted role gives a reason to write today. Intent data and a sound prospecting strategy are where those triggers come from. The trigger has to be true and recent.
• Written like an internal message. "Question about onboarding" looks like it came from a colleague. Hunter's "Quick question" result is the warning: a pattern everybody copies stops working.

Patterns That Look Like Spam or Bait

• Fake Re: and Fwd:. Google's sender guidelines say not to start a subject line with Re: or Fwd: unless the message is an actual reply or forward. The FTC's CAN-SPAM guide says the subject line must accurately reflect the content of the message, that the law has no business-to-business exception, and that each violating email can draw penalties of up to $53,088. That's general information, not legal advice, and other countries have their own rules.
• Fake familiarity. "Our call tomorrow," "Invoice attached," or "Following up" on a first email claim a relationship that doesn't exist. Google asks that subjects and headers accurately represent the message content.
• Urgency and shouting. Belkins reports urgency terms like "now" and "ASAP" below 36% opens.
• Buzzwords, numbers and AI talk. Gong ties them to open-rate drops of up to 17.9% and advises leaving AI references out of subject lines.
• Decorative emoji and symbols. Google's guidelines say not to use emoji or non-standard characters to imitate graphic elements with the intent to deceive or influence recipients.

Subject Line Types and When to Use Each

Language and Register in Southeast Asia

Treat every figure above as untested in your market until your own sends say otherwise. Singapore's business environment is English-first, according to Rework's Southeast Asia business culture guide. In Jakarta, Hanoi, Manila or Kuala Lumpur, the right language depends on the buyer, and a subject line in the wrong one reads like a mass mailing. In relationship-first, hierarchical cultures a blunt, lowercase line may also come across as curt. The communication styles guide explains the differences. Split your tests by country so one market's result doesn't pass for another's.

What to Do With All of This

Pick a plain, specific line that matches the email beneath it. Test it against one alternative, on enough contacts, judged on positive replies. Write down every result. In six months that log will be worth more than any benchmark in this article, including the ones it cites.