Everyone publishes the message that worked. The one that worked is nearly useless, because you cannot tell which part did the work.
So here is one that did not, verbatim, with the numbers.
The message
Two campaigns, 44 messages each, 88 in total, sent between 5 and 10 August 2026. Names and domains have been replaced, because the businesses involved did not sign up to appear in a post about their broken website.
[Name], quick heads up. Your site is showing a security warning in Chrome.
The certificate was issued for a different address, so browsers will not accept it for [domain].
What a visitor gets is a full red page saying the connection is not private, with [domain] sitting behind a "Continue anyway" button that most people will not press.
Open it in Chrome yourself and you will see it in about five seconds.
If you want a hand, reply here and I will tell you exactly what it needs. No charge, and it is a small job either way.
Wolf
Replies: zero. Across both variations. Across all 88.
Why I thought it would work
Read it again and notice how many of the standard rules it obeys.
It is specific to them. It names a real, checkable problem. It is verifiable in five seconds by the recipient, which is about as strong a credibility device as cold outreach has. It explains the consequence in concrete terms rather than abstractly. It asks for nothing except a reply, offers help free, and explicitly says the job is small, which removes the fear of being sold something large. It is short and it does not flatter.
By every craft standard I know, that is a good cold message. I still think it is a good cold message. It produced nothing, twice.
What it was actually doing wrong
It qualified a website. It needed to qualify a person.
A broken certificate is a property of a site. It can be detected from outside, precisely, at scale, for free. Motivation to fix it is a property of the owner, and it cannot be detected from outside at all. Those two things turn out to be close to independent, which means my entire targeting mechanism was sorting people on an axis unrelated to whether they would ever respond. In Revenue Engine terms this was an Attract problem: the attention I was earning was aimed at the wrong thing, so none of it was useful.
So the message arrived at somebody who was not thinking about their website, had not been thinking about it, and now had a stranger telling them about a problem that was theirs and not their concern. The proof made it worse rather than better: being handed evidence of your own neglect by someone you have never met is not a pleasant experience, and the natural response is to close it.
There is a second failure sitting underneath the first. The offer to help free implies a relationship that does not exist. Free help from a stranger reads as the opening of something, not as generosity, and the more useful it looks the more suspicious it becomes.
What the zero actually taught me
I could have run a third variation. There is always another opening line, and craft problems have the useful property of being endlessly workable, which is exactly what makes them a comfortable place to be stuck.
The zero is what stopped that. A weak result invites a rewrite. A zero across two variations and 88 sends says the thing being varied is not the variable, and that was worth more to me than the 34 replies I got from the campaigns that did work.
For context, the same period's other campaigns: 666 messages on one angle producing 30 replies at 4.50%, and 119 on another producing 4 at 3.36%. Total across everything, 899 sent, 34 replies, 3.78%. The fuller breakdown is in every channel I tried for cold outreach.
What I would do instead
Look for evidence of a person rather than evidence of a problem.
Somebody said something out loud about it. Somebody asked a question. Somebody came to an event on the topic, replied to something written about it, or complained about it publicly. Every one of those is a signal about a human being and is worth more than a hundred businesses that match a technical filter.
This is also why, for most established businesses, whether that is a firm, a clinic, an agency or a coaching practice, the strongest list they own is the people who already know them, and especially the ones who contacted them and went quiet. Every one of those names demonstrated motivation at some point. That is the property that cannot be bought or measured from the outside, and it is already sitting in the inbox, waiting for someone to reopen the conversation.
Questions owners ask
Should I not point out a real problem in outreach?
Not as the opening move to a stranger. A provable problem gives you credibility only after somebody has decided they care, and it can read as an accusation before that point. The sequence that works is interest first, evidence second.
How many sends before I decide an angle has failed?
Enough that a zero means something. Forty-four sends producing nothing is a signal; three sends producing nothing is a quiet week. I had three campaigns in the same period with send counts of 22, 3 and 1, and I do not report those as failures because they never produced enough information to be one.
Is free help a bad offer?
From a stranger, usually. It works once a relationship exists and it triggers suspicion before that, because the recipient is trying to work out what it is the opening of. Offering something impersonal and complete, that costs them nothing to accept and requires no reply, is more reliably received.
Where did these numbers come from?
A single query against my own send log, covering 5 to 26 August 2026. The message text is verbatim from the record of what was actually sent, with the recipient's name and domain removed.