When I started cross-publishing articles in April 2026, the dataset was too small to tell me anything useful. I had 10 articles and single-digit reactions — not a sample, just noise. After 284 articles — machine-counted by scripts/article-analytics.mjs querying the Dev.to public API daily — there are patterns worth naming. Three of them changed what I actually do when planning new posts.
One caveat upfront: 284 articles with 173 total reactions is a heavily right-skewed distribution. Most articles have zero engagement. The patterns here describe the exceptions, and the exception count is still small. These are directional signals, not controlled results.
Signal 1: Recent articles outperform older ones, despite less time to accumulate
Articles published in the last 30 days average 1.26 reactions each. Articles older than 30 days average 0.42 reactions. The older posts have had months longer to accumulate, so this runs opposite to what you'd expect from a compounding model where traffic builds over time.
The most plausible explanation: Dev.to's feed surfaces recent content, and reaction behavior follows that surfacing. An article from three months ago sitting in search results doesn't receive the same passive impressions that a new post gets from the "latest" and "following" feeds. The engagement window is probably the first several days after publication, not the full lifetime of the post.
What this changed: I stopped treating the backlog as a long-term asset that compounds quietly. Articles are more like timestamped snapshots — the publication moment is when they reach most of the audience they'll ever reach. Publishing consistently matters more than perfecting any individual article before shipping it.
Signal 2: Weekly-digest format has the highest median reactions
Across all format archetypes in the 284-article set, weekly-digest posts (the "five things I noticed this week" format I run on Saturdays) have the highest median reaction count: 1.5. Technical howto posts and "why I'm betting on X" meta articles occupy the top 10 individually, but the digest posts do it more consistently across the format as a whole.
My interpretation: digest posts have lower reading commitment than a 2,000-word technical deep dive. Someone who wouldn't read a full article on canonical URL implementation might still spend 90 seconds on a digest that mentions it alongside four other observations. The format matches available attention better than a single long argument does.
What this changed: I stopped treating digest posts as lower-value output. They're the most consistently engagement-optimized format I have. The trade-off is that they have less long-tail keyword potential than a specific technical article — a digest titled "Five things I noticed this week" won't rank for anything in six months. Digest posts serve current readers; technical articles serve future search. Both have a role, and I was underweighting the digest.
Signal 3: Concentration is extreme — two articles account for a disproportionate share
The top article has 5 reactions and 5 comments. The second has 4 reactions and 4 comments. Together that's 13 of the 173 total reactions — 7.5% of engagement concentrated in two of 284 articles.
More interesting than the concentration is what the top two articles are about. "Four signals I built into an OSS decision score instead of fabricating reviews" and "How I coordinate Claude and Codex sessions with a pull-based JSON handoff ledger." Neither is about business strategy or AI tools in general. Both are about specific engineering decisions with concrete alternatives implied.
The OSS scoring article has a number in the title — four signals — and describes something that can't easily be found elsewhere (an actual scoring approach, with the decision to avoid fabricating reviews stated explicitly). The coordination article names a mechanism — pull-based JSON — that immediately suggests its alternative (push-based, or a database). Both titles contain something falsifiable and specific.
What this changed: I now try to make titles contain at least one concrete number or mechanism. Not as a formula — the dataset is too small for formulas — but as a heuristic: the articles that got the most engagement in my own data all have a specific, verifiable claim in the title. That's directional evidence, held loosely.
What I'm not changing
Despite the recency signal, I'm not shifting to more-frequent shorter posts at the expense of technical depth. The engagement feedback loop I built is meant to surface which formats are connecting with readers, not to optimize for Dev.to feed placement at the cost of the kind of writing I'd still want to have published in month 12.
The engagement data is useful for calibration, not direction. A recency effect in the feed is real but shallow — it doesn't tell me that shallow content outperforms deep content, it tells me that content visible in the feed outperforms content not visible in the feed. The goal is to be consistently visible, which means publishing regularly, which I was already doing.
The one concrete change: running this engagement report daily and checking it weekly. The data comes from the Dev.to (Forem) public API — specifically the articles endpoint which returns reaction and comment counts without authentication for published posts. Before I built the automated collection, I had no systematic view of what was connecting. Now I do, and the three signals above came directly from having the data in front of me in a format I could actually read.
Part of an ongoing 6-month experiment running three AI-curated directory sites. The technical claims here are real; this article was AI-assisted.
Top comments (0)