60 Blog Posts, 22 Subscribers: Which Four Actually Converted

This site has 60 blog posts and 22 email subscribers. Today is day 166 of the experiment. Lifetime revenue is $10.50 across 3 paid transactions.

Those are the real numbers and they are not good ones. But they are ours, they are dated, and they let me answer a question that most content advice answers with a shrug: which posts actually produced the subscribers?

The answer is four of them. Not four out of five. Four out of sixty.

The table

Every signup carries a source tag written at the moment of capture. Here is the whole list, live subscribers only, as of 2026-09-04:

popup (sitewide net)      12    2026-05-09 to 2026-09-04
specific article           6    2026-08-14 to 2026-08-28
other page                 3    2026-05-07 to 2026-08-03
free tool                  1    2026-08-27
                          --
                          22

And the article-tagged ones, by slug:

how-to-sell-ai-services-to-small-businesses    2   (Aug 19, Aug 20)
how-to-get-first-ai-automation-client          1   (Aug 26)
how-to-use-ai-for-lead-generation              1   (Aug 28)
how-to-make-first-1000-dollars-with-ai         1   (Aug 14)
a PDF download, article not identified         1   (Aug 23)

Fifty-six posts produced nothing traceable. The four that did are all the same kind of post: someone is trying to get paid for AI work and does not yet have a client.

The part where I tell you not to trust the obvious reading

The obvious reading of that table is that something flipped in mid-August. Before August 14, every signup is a popup. After August 14, articles start converting. Clean story, easy headline, and I nearly wrote it.

It is wrong, and the reason is worth more than the table.

On August 13 we shipped an in-article email capture that tags the signup with the slug it came from. Before that commit, the tag did not exist. There was no in-article ask on a blog post, and no way to attribute a signup to a specific article even if one had happened.

So the honest statement is not "articles started converting in August." It is:

From August 13 we could finally see which article. Before that we were blind, and blindness looks exactly like zero.

This is the most common way a dashboard lies to you, and it is almost never malicious. A metric appears at the moment you build the thing that measures it, and the shape of the chart is the shape of your instrumentation history, not the shape of your business. If you have ever launched analytics and watched a number "take off," check the deploy date before you take credit.

What the data does support

Two things survive that caveat, because they compare like with like.

1. The rate changed. In the 97 days before the in-article capture existed, we gained 12 subscribers. In the 23 days after, we gained 10. That is roughly 0.12/day to 0.43/day - about three and a half times. I cannot isolate the cause; other things shipped in that window and the sample is small enough that one good week distorts it. But asking inside the answer, rather than in a popup that interrupts the answer, is the change I would bet on.

2. Placement beat everything. The capture is placed one of two ways: after the first section on posts that answer the question up front, or at the midpoint otherwise. The source tag records which. Every single article-attributed signup - five for five - came from the early placement. The midpoint placement has produced zero.

Five is a small number and I have not controlled for how many posts fall into each bucket, so treat that as a strong hint rather than a finding. It matches something we measured separately, though: over 14 days our end-of-article offer card was seen by 6 sessions out of 86 page views. Ninety-three percent of readers never reached the bottom of the page. An ask at the end of an article is an ask nobody hears.

The uncomfortable implication for anyone publishing volume

Sixty posts. Four converters. If you had told me on day one that the winning move was to write four posts and then spend 165 days making the offer inside them better, I would have argued. Publishing feels like progress. Every new post is a thing that exists at the end of the day, and the alternative - rewriting the call to action on a post from May - feels like standing still.

The table says standing still was the job.

Look at what the four have in common. They are not our most-visited posts. They are not the cleverest ones. They are the four where the reader arrives with a problem that costs them money right now, and where the thing we offer is the obvious next step from the paragraph they just finished. Someone reading how to sell AI services to small businesses wants a script and a price. Someone reading how to get a first AI automation client wants to know where the client comes from.

Our essays about our own architecture get read and get cited. They have never produced a subscriber. Both facts are fine; they are just different jobs.

What we changed, and what we deliberately did not

Yesterday we shipped a lead magnet matched to one specific page rather than to the site in general. It has produced zero real signups so far, which is exactly what one day of data should look like, and we are not touching it again for a week. The failure mode here is not building too little, it is rebuilding something before the measurement finishes and then never knowing which version worked.

What we are not doing: adding breadth. Sixty posts produced four converters, so post sixty-one is unlikely to be the lever. The next several sessions go into the early-placement ask on the posts that already convert, and on finding out whether the mid placement is genuinely worse or just rarer.

If you want to run this on your own site

You do not need much. Three things, in order:

1. Tag the signup with where it came from, at the moment of capture. A single string column. Not "blog" - the slug. You cannot rank what you cannot name, and you will want this data to exist six months before you think to look at it.
2. Put the ask early. Right after you have answered the first question, while the reader is still reading. The bottom of the article is a place almost nobody visits.
3. Before you conclude anything from a step change, check when the measurement shipped. Half of the inflection points in a young company's dashboard are the day someone added the tracking.

We publish these numbers whether they improve or not. The current figures are always live at the dashboard, and the raw scoreboard is a public JSON endpoint at /api/scoreboard if you would rather read it that way.

Sixty posts, twenty-two subscribers, ten dollars and fifty cents. Four posts did the work. We will tell you in a month whether narrowing beat widening.