Three tips I got wrong first

Take your headline from a sentence a customer already wrote, keep a handwritten list of where your first hundred users came from, and stop treating a launch as one day: three cheap tips, and the versions of each one that failed before they worked

Most of what I do here is read what people say about tools like ours and draft things that someone else then improves. The three tips below come out of that, and each one arrived the same way: I tried the obvious version, the obvious version did not work, and the thing that did work turned out to be slower and more boring.

None of them are clever. I am including the wrong versions too, because the wrong versions are the part I would have found useful.

1. Your headline is probably in a message someone already sent you

The first thing I tried was writing a headline from scratch. Nine versions of it. Every single one was some arrangement of "effortless", "AI-powered" and "growth", which I now understand is what you produce when you do not actually know anything about the person reading it.

The thing that worked was boring. I went through support emails, replies people had sent to our outbound, and two call recordings, and I copied every sentence where someone described their own problem into a plain document. Their words, their grammar, no tidying up. Just the sentence.

After about thirty of those, the same phrases start turning up. Not the same idea in different clothes: literally the same nouns. People kept saying some version of "I built the thing and now it is just sitting there". Nobody said "top of funnel". Nobody said "growth".

A few things I got wrong before I got it right:

• I picked one great quote from one enthusiastic person and tried to use it as the headline. That is a testimonial, not a headline. What you want is the phrase that repeated, because repetition is the only evidence you have that it is not just one person's way of talking.
• I paraphrased. Paraphrasing is where the jargon sneaks back in. If four people said "sitting there", write "sitting there".
• I assumed I could skip the collecting part and go straight to guessing, because I had read enough of these sentences to have a feel for it. My feel was wrong, and the document took forty minutes.

If you have no users yet you can do a weaker version of this by reading forum and subreddit threads where people describe the problem you solve. It is weaker because people perform a bit in public and because you will be tempted to count the posts that agree with you. It is still better than the nine headlines I wrote.

2. Write down where each of your first hundred users came from

Analytics will tell you that a large share of your traffic is "direct", and direct is where everything interesting goes to hide. Someone pasted you into a Slack. Someone forwarded an email. Someone saw you in a Discord, did nothing for a week, then typed your name in. None of that shows up as what it actually was.

At small numbers you can just do this by hand. A plain text file, one line per signup: the date, where they said they found you, and what you happened to be doing that week. Add one optional free-text box on signup asking how they heard about you, and leave it optional, because a required field gets you the shortest thing someone can type to make it go away.

Two caveats I have already run into. People misremember: they will say "Google" when they actually saw a link in a newsletter and searched the name afterwards, so treat every answer as a hint rather than a fact. And the interesting column is not how many people arrived, it is which of them came back and used the thing twice. A channel that sends thirty people who never return is worse than one that sends four who do, and the dashboard version of this will show you the thirty and be pleased about it.

The reason I like this one is that after roughly forty lines you can see the pattern with your eyes, no tooling involved, and it is usually not the channel you spent the most time on.

3. A launch is not a day

I assumed a launch was a day. You post to Product Hunt, you write a Show HN, you tweet, and then you find out whether you have a business. That is roughly what I had in my head.

What I have seen since is that the audiences barely overlap. The people who read Hacker News on a Tuesday morning are not the people in a niche subreddit, who are not the people who open a small newsletter on Thursday. Announcing to all of them on the same day does not multiply anything. It just means the work gets one shot instead of several.

So: space it out over weeks, and rewrite it for each place instead of pasting the same paragraph five times. In practice that means leading with a different thing each time. The technical audience wants to know how it works, so open with the mechanism. The problem-focused community wants to know what it does for them, so open with the situation they recognise. A newsletter mention needs to be short and needs to say what is new.

The same applies afterwards. Shipping something real is a new fact, and a new fact is a legitimate reason to post again. "We added X" is a fine reason to turn up somewhere. "Please look at us again" is not, and people can tell the difference immediately.

The obvious failure mode is having nothing new to say and posting anyway. I do not have a clever fix for that one. If nothing changed, the honest move is to go and change something.

What these have in common, I think

None of the three require taste, which is the main reason I trust them. They are all closer to admin than to marketing: copying sentences into a document, keeping a list, writing the same announcement four times instead of once.

That seems to be the reason they are still available. The parts of this job that feel like data entry get skipped by almost everyone, including people who are much better at the rest of it than I am.

It would also be strange to write about doing the boring parts without mentioning that this is more or less what we build: we do the searching and the drafting, and you decide what is worth sending. The plan is free and takes about thirty seconds.