Skip to content
From the blog

Cold email subject lines: the broken metric

Write a short, flat, lowercase label for what is inside the email, six words or fewer, then stop. The reason nobody says it that plainly is that the whole category grades subject lines with an instrument that counts machines, and in B2B that was true before Apple touched it.

By Rahul Bageria, co-founder · Updated August 2026 · 12 min read

The patterns

Five subject lines that are labels, not hooks

A subject line names what is inside, so the message reads as correspondence rather than a campaign. Five patterns cover almost every first cold email.

Pattern Write it like this What it labels Where it fails
The observation series a, then a revops hire The trigger event you actually saw, in their words. You inferred it rather than saw it.
The problem duplicate accounts before migration The problem the email argues about. The problem is true of the whole market.
The artifact the ops role you posted friday A document you opened and read. You never opened it, a tool found it.
The category outbound for seed teams What you sell, said flat and small. It is doing the persuading for you.
The inherited thread (nothing, you keep the original) Nothing new. It continues a real thread. There was no thread, and you wrote one.

Illustrative examples, written for this page, not campaign results.


The mechanism

Open rate is not a measurement any more

An open is a hidden image loading. Once something other than a person loads it, the number stops being about people.

Definition

An open is one request for a tracking pixel. Your tool logs the request and cannot see what made it. Also see open rate.

Three beats
1
The pixel

One image request, logged as a person

Open tracking loads a hidden image. Anything that fetches it is recorded as an open, and nothing in the record says whether a human was involved.

2
Sept 2021

Apple made the fetch automatic

Apple's newsroom, 20 September 2021: Mail Privacy Protection "prevents senders from learning whether an email has been opened". Apple's privacy documentation says Mail downloads remote content in the background by default, whether or not you engage with it.

3
July 2026

Most of the opens you count are Apple's

Litmus puts Apple at about 62% of tracked opens, calculated from over a billion opens inside its own analytics. Vendor data from Litmus customers, and Litmus states that privacy-impacted opens sit inside that Apple figure.

Nobody can tell you what share of your own opens this touches. Mail Privacy Protection is a property of the mail client, not the mailbox provider.

A prospect on Google Workspace who reads on an iPhone is an Apple Mail reader here, so no provider-side figure answers it.


The finding

In B2B it was broken before Apple touched it

Ten months before Apple shipped anything, the body that writes the messaging abuse standards published a B2B number. Nothing that ranks for this query cites it.

20 to 80%
Of B2B interactions

of B2B email opens and clicks were machine driven, ten months before Apple Mail Privacy Protection shipped.

M3AAWG, November 2020 A range, not an estimate

M3AAWG, "Exploring the Impact of Nonhuman Interactions on Email Send Metrics". Consumer mail in the same study ran under 10%.

The hedge, in their words

M3AAWG says the B2B data behind that range was "collected but not published", and that a summary effect is hard to quantify because filters work to mask detection. The width of the range is the finding.

It also found the range tracked engagement: a lower engagement base meant a higher probability of scanning. Cold outbound is the lowest-engagement mail there is, which is our reading rather than their measurement.

The cause is not your writing. M3AAWG puts it on security scans inside businesses running enterprise anti-spam: the system follows the links, visits the target site, and an open or a click gets recorded.

Apple did not break the B2B open rate. The security stack got there first, and Apple finished it.


The arithmetic

Why sending more does not fix it

The instinct is that a bigger sample settles it. It does not, and the reason is a difference between two kinds of error.

Noise averages out when you send more. Contamination does not, because its size is a fact about your list.

Three ways it breaks
Between lists

The gap is about their IT, not your writing

Two lists with different client mixes carry different amounts of machine traffic. The gap between their open rates is a fact about the recipients' security stacks first.

Inside one list

Randomizing balances it, and it still bites

Split one list and the machine traffic lands evenly on both sides. But most of the number is machine, so the human part is a minority share and the volume you need climbs.

At seed volume

You never get there

A founder sending tens of emails a day cannot power that test in any useful time. The guide to writing the message already says you cannot measure your way to good copy at seed volume. This is the arithmetic under it.

The experiment you can still afford runs one segment against another and scores on positive replies. That is what an ICP slice experiment is for, and it never grades a subject line on its own.


The escape hatch

The escape hatch is closed too

Every vendor that admits the open rate is broken points you at clicks instead. In B2B that door is shut, and by the same machines.

Three parts, sourced
The advice

Test on clicks instead

Litmus, June 2022: rather than trying to get an open, "measuring click-through rate or conversion rate when A/B testing subject lines is an option". Where that will not work, Litmus says it may be time to stop.

The B2B problem

The scan clicks before the reader does

Microsoft's Safe Links documentation, updated May 2026: "URLs are scanned prior to message delivery, regardless of whether the URLs are rewritten or not". The built-in preset applies it to all recipients on any Defender for Office 365 license.

The cold problem

There should be no link to click

A first cold email carries no link anyway. A tracked link is a deliverability cost before it is a measurement, which our guide to staying in the inbox works through in detail.

What survives is the reply. A reply is typed by a person, and no security stack has ever written one. It also grades the whole email rather than the subject on its own.


The receipts

What the famous subject-line data actually measured

The most-quoted cold email dataset in the category has a subject-line section. Read what its charts are plotted on.

How it gets quoted

"85 million cold emails prove that lowercase subject lines win."

  • Repeated as a result about people
  • No collection period, no methodology section
  • Privacy, pixels and MPP appear nowhere in it
What the report holds

"Four subject-line charts. One is plotted on reply rate. Open rate is an axis on the other three."

  • Chart titles read off the report's own text layer
  • The reply-rate chart is about length, not case
  • Its reply-rate findings are cited elsewhere on this site

That is the 30 Minutes to President's Club report with Gong, 2025. Its lowercase chart says what we say. We agree with its conclusion and we will not use its evidence.

The older subject-line rules, four studies with four different answers, are taken apart by name in the copywriting guide. All four measured opt-in marketing email, on open rate.

Want your outbound copy written and judged by the two of us?

Book a Fit Check

Our rule

How to judge one: qualified, not optimized

No measured subject-line finding survives on a metric that still works, so these four rules are mechanism and judgment. We would rather say that than dress them up.

Four rules
1
Inspect

Qualified by inspection, never chosen by a test

The metric that would grade it counts machines. The metrics that still work grade the whole email. So the honest procedure is a checklist before the send, not a chart after it.

2
The row

The unit of judgment is the inbox row

A reader decides on sender name, subject and the first line of the body together, in about a second. You control the subject least: the client decides where it gets cut.

3
Register

Match the register of what you are actually sending

Lowercase is not a trick that lifts a number. It is what a note typed by a person looks like. It works when the rest of the email is one, and it fails when it is a costume on a template.

4
Alarm

An open is a smoke alarm, not a thermometer

Read it in the negative, in aggregate, across a mailbox or a domain, and only as a question about infrastructure. Never as a verdict on a subject line.

A subject line is not a lever you pull. It is a gate you walk through.


The stance

The lowercase stance, and where it stops being true

We write subject lines lowercase. The reason matters more than the rule, because the reason is what tells you when to stop.

Don't

Write a headline for an email

Boost Your Q3 Pipeline by 3X

  • Title case reads as a campaign in the list view
  • A claim asks for belief before the read
  • Survives being sent to four hundred people
Do

Label the thing you actually wrote

the ops role you posted friday

  • Reads as correspondence in a one-second scan
  • Short enough to survive an invisible cut
  • True of this account and few others

An illustrative walkthrough of the method, not a specific client result. We report real numbers only when they are real.

Operator note
Where the rule breaks

Lowercase has become a tell. Every tool shipped it as a default, so a lowercase subject on an obviously templated email now reads as a costume. The rule survives only with its reason attached: lowercase because the message really is a note.

RB
Rahul Bageria
Co-founder, Real Good GTM

It also fails honestly in two places: markets where business correspondence is formal, and executives whose mail is triaged by somebody else. Both read lowercase as sloppy rather than human.


Failure modes

What never to write

Four of these cost you the read. The fifth costs you more than that.

Manufactured urgency

"Quick question", "urgent", a deadline you invented. Every sender before you has trained the reader on these, and the classification takes under a second.

A merge field in the subject

A broken token in the body is read by one person who already opened. A broken token in the subject sits in a list beside real correspondence. It is the most legible tell there is.

Emoji, for the unexpected reason

The problem is not that they lower opens. M3AAWG lists emoticons in subject lines among the styles that correlate with more machine handling of your mail, not less.

An empty subject line

The report everyone quotes calls this one out itself: empty subject lines "look good, however", and its own takeaway is to put an actual subject line in there.

!
Caution

The fake "Re:" is the one you cannot take back

Google's sender guidelines say not to send subject lines starting with Re: or Fwd: "unless the messages are actual replies or forwards", and ask that the subject accurately represent the message content.

US law is not softer. Under 15 U.S.C. 7704(a)(2) it is unlawful to send a commercial message whose subject heading is likely to mislead a recipient about a material fact.

Do this instead
Keep the inherited subject on a real thread. Write a new one on a first touch.

The ops

What an open still legitimately tells you

One reading survives, and it is smaller than anyone wants it to be.

Four limits
Negative only

A collapse means something, a rise means nothing

Opens falling to near zero on a mailbox that used to log them is a real event. A high number is not the mirror image of that.

Aggregate only

Per mailbox or per domain, never per send

The reading is about a sending asset over weeks. Comparing one campaign's open rate against another's compares two client mixes, which is not a comparison.

Infrastructure

It is a placement question, not a copy question

The machines generating most of your opens do not read subject lines. Whatever the number moved on, it was not your phrasing.

Better instruments

Postmaster Tools and seed tests answer it cleanly

They answer the same question without a pixel, which is why the deliverability guide keeps opens off the panel.

Operator note
How we run it

We turn open tracking off. The pixel and the tracking domain that produce the number are themselves things a filter can score, so you are paying deliverability for a count of machines. Switching it off is the cheaper option, not the principled one.

KM
Kshitij Maheshwari
Co-founder, Real Good GTM

Key takeaways

What to carry out of this

Key takeaways
4 points
  • 1 Write a label, six words or fewer, lowercase and flat.
  • 2 Do not A/B test one. Qualify it by inspection before it goes out.
  • 3 Read an open as a smoke alarm: negative, aggregate, infrastructural.
  • 4 Score the segment on positive replies. No machine writes those.

FAQ

Questions founders ask

What is a good open rate for cold email?
No honest answer has been published since September 2021. Apple's Mail app loads the tracking pixel whether or not anybody read the message, and in B2B the recipient's own security stack was already generating opens before that. Your open rate is a blend of people and machines in a ratio nobody can state for your list. Judge on positive replies.
Should I A/B test cold email subject lines?
No. The metric that would settle it counts machines, and the metrics that still work grade the whole email rather than the subject. Test angles and segments on positive replies instead, and qualify the subject line by inspection before it goes out.
Do lowercase subject lines actually work better?
We think so, and we will not point at the chart everyone quotes, because it is an open-rate chart. The reason to write lowercase is that a note typed by a person looks like a note typed by a person. It stops working the moment the rest of the email obviously is not one.
How long should a cold email subject line be?
Six words or fewer, and short enough to survive being cut by a mail client you cannot see. Every published length rule was measured either on opt-in marketing email or on open rate, so none of them is evidence about cold B2B email. Length is not what decides it anyway.
Can I use "Re:" or "Fwd:" to get more replies?
No. Google's sender guidelines say not to unless the message really is a reply or a forward, and US law makes a subject heading likely to mislead a recipient about a material fact unlawful. It also spends the reader's trust in the second they work out what you did.
Does Apple Mail Privacy Protection affect Gmail and Outlook recipients too?
Yes. It is a property of the mail client, not of the mailbox provider. A prospect whose address sits on Google Workspace but who reads mail in the iPhone Mail app is an Apple Mail reader for this purpose. That is one reason the true share is not knowable from provider data.
Rahul Bageria, co-founder of Real Good GTM
About the author
Rahul Bageria

Co-founder of Real Good GTM. He has been the first business hire and Chief of Staff at seed-stage B2B startups, building outbound pipeline before any playbook existed. This post exists because he went looking for the measurement behind the standard subject-line advice and found a standards-body document nobody in the category cites.

Connect on LinkedIn

Keep going

The rest of the message

The subject is a label for something already written. These three cover what it is labeling.

Want outbound copy judged on what still measures people?

Book a fit check. We'll look at the segments your market actually has, write the messages per segment, and score them on positive replies rather than on a number that counts security scanners.

Book a Fit Check

No hard sell. No fake numbers. Real good work speaks for itself.