Email Subject Lines for Cold Emails: The Tested-List Myth

Email subject lines for cold emails are not something you copy from a list. The ones that get replies name one thing only the recipient owns (their returns page, their job post, their new pricing) and put the ask in four words or fewer. The tested lists ranking for this query contradict each other, because a line's results belong to the sender, list and offer it was tested on, and none of those are yours.

The version that holds

A subject line's performance is a property of three things together: a noun only the recipient owns, an ask short enough to sit next to it, and a sender the inbox already trusts. The words are the smallest of the three. What transfers between senders is not the line, it is the generator, and the generator is the rule from Cold Email Subjects Are a Filter, Not a Hook: the shortest true label of your one ask, written after the body, plus the one thing you know about this person.

I gave Claude one fact, and only one: a Shopify store selling ceramic mugs whose returns FAQ says exchanges take 10 days, and a solo developer who wants 15 minutes to show a returns-tracking spreadsheet. The ten lines back included Your 10-day exchange window, Noticed "10 days" on your returns FAQ, 15 minutes on mug exchanges? and Kiln & Co returns, tracked in one tab. Nine of the ten name the fact, and none of them appears in any tested list I have seen. The tenth, A returns sheet built for small Shopify stores, could go to any store, and it is the weakest of the batch. The difference was one sentence of input no list can contain, because it is about one person.

So the rule for a solo sender: the research writes the subject line, not the swipe file. If you cannot name one thing the recipient owns, you have not earned the send yet; go find it. If you draft with an AI, give it that thing and the ask and nothing else, and if what comes back could go to anyone on your list, you gave it nothing. This is also why the Tim Ferriss cold email template has survived eighteen years of being copied: it is a generator, not a line. It refuses to let you write until you have named the specific thing.

And stop testing the words. To tell a 3% reply rate from a 5% one with an ordinary two-proportion test, you need about 1,500 sends per variant. At fifty sends a week that is over a year. Put that attention on whether the domain lands at all, which is the whole job of a warm-up service like Mailivery, and on how you follow up a cold email after the first silence, which decides more replies than the subject does.

The myth, stated in its own words

Search this query and you get a stack of lists: 20 lines from Brafton, 25 from Career Contessa, 50+ from cleanlist, 10 "that hit 80%+ open rates" from lemlist, a YouTube video that tried 225. Under each sits the same claim: subject line performance is a property of the words. Test enough of them, keep the winners, publish the winners, and whoever copies them inherits the result.

That is the myth, and it is not that lists are useless. It is that the result travels with the line.

“

A subject line that could have been sent to anyone carries no information for the one person who got it.

”

Why it circulates

Open rate is the only number a subject line can claim on its own. Replies depend on the body, the offer and the list; opens look like they depend on the subject alone. So it became the cheapest claim in cold email to make and to test. lemlist's ten lines qualified by being "sent at least 50 times" with 80%+ opens. Forty opens out of fifty gives a 95% interval from 67% to 89%, before you subtract the opens Apple Mail has generated on its own since Mail Privacy Protection shipped with iOS 15 in September 2021.

The tools that sell sequences also sell the lists. Four of the eight lists come from companies selling email or sales tooling. The statistics recycle with them: cleanlist's figures that 47% of recipients decide to open on the subject line alone and 69% report spam on it are sourced to an OptinMonster roundup. lemlist cites a Belkins analysis of 5.5 million B2B cold emails sent in 2024, personalized subjects at 46% opens against 35%. Real numbers, about other people's lists.

Give the myth what it has right. Short wins, because phones cut a subject somewhere between 35 and 50 characters. A line that names something specific to the recipient beats one that could go to anyone. And a consultant who says they tested lines across 150+ client campaigns concludes that the subject line barely matters next to deliverability. On that point I think they are closer to right than the lists.

Where the lists break

Take one line and follow it through the lists. "Quick question." The consultant's LinkedIn post calls it one of the best performing lines they have seen and tells you to pick a generic line and move on. Brafton lists "Quick question regarding [project]" at number 4 of 20. Exclaimer's mistakes section opens with it: no context, no relevance. Salesforce's don'ts read "be a cliche (or generic)" and "do what everybody else does". Four tested verdicts on the same two words. They cannot all be measuring the words.

They are measuring their own lists. A line that worked across 150 client campaigns gets reported as a property of "Quick question". Move it to a solo developer emailing store owners from a domain registered last month and the string carries none of what made it work: the reputation, the list fit, the offer behind it. The open rates agree from the other side: Career Contessa puts a well-run cold email at 20 to 25% opens, lemlist's list sits at 80%+, and both are probably true of the lists they were measured on.

Then there is the loop nobody mentions. I asked Claude for "10 cold email subject lines", no other context, and got back Quick question about {{company}}'s pipeline, Idea for {{company}}, Are you the right person for this?, {{mutual_connection}} suggested I reach out, 15 minutes next week?. Seven of the ten sat as bracketed templates on a single page, cleanlist's, two of them word for word. That is what "tested" means now: the lists trained the models, the models feed the sequence tools, and your recipient's inbox is where all of it lands at once. A subject line that could have been sent to anyone carries no information for the one person who got it. Compare that with the ten lines above, written from one fact about one store: same model, opposite result.

This week, read your current subject line as the next person on your list would. If it could reach them unchanged, delete it, write down the one thing you know about them, and let the line come from that.

FAQ

Is "quick question" a good subject line for a cold email? Only when the sender is already trusted and the question really is quick. To a stranger, from a new domain, it says nothing about why you wrote. Replace it with the actual question, shortened.

What open rate should cold email subject lines get? Whatever you see is inflated by machine opens, Apple Mail's since 2021 among others. The lists claim anything from 20 to 25% to 80%+ for the same kind of email. Count replies per hundred sends instead, and read opens as a deliverability signal, not a copy signal.

Can I A/B test subject lines as a solo sender? Not usefully. Separating a 3% reply rate from 5% takes about 1,500 sends per variant. Below that, the differences you see are noise, and your list changes faster than the test completes. Pick lines by rule, and test the offer and the list instead.

Should I use ChatGPT or Claude to write cold email subject lines? Yes, on one condition: give the model the one fact you know about the recipient and the ask, nothing else. With no context it returns the template lines that fill every list. With the fact, it returns lines only that recipient could receive.