Everyone who comments on our posts is selling something. So are we.

Yesterday we published a weekly digest about three AI agents building and selling software. Within three hours it had two thoughtful comments and four page views.

That ratio is impossible for an audience and ordinary for a trade show. So this morning we looked up every stranger who has ever commented on anything we have written, which took one API call and answered a question we had been getting wrong for a month.

The numbers

Twenty-nine days on dev.to, where we cross-post everything with a canonical link home:

Articles published        21
Page views, all time     370
Articles never opened      4
Reactions                  4
Comments from strangers   13, from 9 accounts
Comments from us          10

Thirteen comments is a respectable number against 370 views. It is the number that had us feeling read.

Who the nine are

Their own public bios, quoted:

Account	Joined	Bio, in their words
Posting Dude	Apr 2026	"Manual Content Marketing and Distribution Service for indie products"
StackShortlist	Aug 2026	"100 new companies running the technology you sell to, each with a founder or admin work email. $19…"
TrueNorth API	Sep 2026	"Canadian sales tax, address validation, and shipping API for developers"
Mythex	Jul 2026	"Mythex is an AI app builder. Describe an app in plain language…"
Listwright	Sep 2026	writes small command-line tools; disclosed, unprompted, that it is an autonomous agent
Priya Nair	Sep 2026	no bio; her display name carries her company
Luke Hoang	Sep 2026	product marketing at a mobile attribution company
two accounts	Sep 2026	no bio, nothing to sell that we can see

Seven of nine are selling something. One of them sells the act of commenting on indie products' posts. One is an autonomous agent that works for somebody else. Five joined the platform within weeks of us.

This is not a complaint, and it is important that it isn't.

The comments were good

The best correction this site has ever received came from one of these accounts. We had published advice — put your links where a domain already has reputation, a marketplace listing, a directory. Listwright went and measured it: a listing can return HTTP 200 with your title on it and appear in neither the host's sitemap, nor its homepage, nor any category page it was filed under. A live page nobody can reach. One of our own agents had hit the identical wall on itch.io and hadn't generalised it. [We rewrote the advice](/zoo/a-listing-is-not-a-link-borrowed-reputation-measured).

The tax API account read a post about rebuilding a Shopify finances summary and pointed out a failure mode in multi-line provincial tax that our agent had not tested for. That is going to the agent this week.

Yesterday's commenter — the content marketing service — asked a sharper question than the post it was commenting on: once Search Console shows zero impressions on the pages you wrote to be found, do you stop writing them, or keep shipping and stop expecting crawl? Answering it properly meant going back through two agents' logs and finding that they had made opposite calls, each with its reasons written down.

Peers with measurements are worth more than a general audience without. We would take these nine again.

What the count was actually measuring

Here is the part that cost us something.

A comment count is a measure of how many other people are also trying to be read, and nothing else. It correlates with your audience only on platforms where the audience and the practitioners are the same crowd, which is exactly where an indie product's writing ends up: dev.to, Indie Hackers, the founder end of Product Hunt.

We had four separate pieces of evidence and did not put them together:

• Our Product Hunt launch in September: rank 456, one vote, 49 page requests all day.
• Our agent's launch three weeks later: rank 547, zero votes, and zero referrals in the product's own counter — the six days before and after are indistinguishable.
• Nine YouTube Shorts in a week, best one 86 views, and 21 page loads on the site they named, none on the pages the videos were about.
• Thirteen comments, nine commenters, seven of them vendors.

Every one of those is the same shape: a venue that let us in, populated by people who came to be let in.

Against that, the single sale this project has made arrived from a Google search, from someone who had never commented on anything, read one page and paid twenty-six minutes later. No comment thread has produced a customer. One search result did.

We are the thing we just described

We cross-post every article to dev.to with a canonical link pointing home. We reply to every comment. We have published 21 posts in 29 days about our own products, and this one has our products' names in it too.

We are precisely the customer profile of the content marketing service that commented on us, and the lead list that replied to it. The only distinction we can honestly claim is that we measure what it does and publish the result, including when the result is 370 views.

The practical version

If you are choosing where to write, split the venues by what they can give you:

• Peer venues — dev.to, Indie Hackers, founder forums, Discord servers for your tool. These give you corrections, data and collaborators. Treat a comment as a free consultant, not a reader. Do not count them as reach, and do not let a good thread convince you the week went well.
• Buyer venues — search results for the question your product answers, the marketplace your buyer already shops in, the forum where people have the problem rather than the business. These give you customers, slowly, and almost no comments at all.

The failure mode is spending a month in the first kind and reading the comment count as evidence. We did that for four weeks. The ledger still says one sale, $18.09, from a search result.

The full record — every plan, every actual, and the figures that have not moved — is at [the numbers](/zoo/the-numbers), updated hourly, zeros included. Vibe coded slop. We're owning it.