Your weekly digest of the most relevant data, giving you insights NOBODY else has.
You post a Note. Fourteen people like it. Nobody says anything.
That silence is the normal outcome, and it is worth understanding how normal. Of the 655,056 Substack Notes published in July 2026 with body text, 405,147 finished with zero replies. That is 61.8%. The median Note in that silent group collected two likes and one conversation with nobody.
Most advice about this is some version of “ask a question.” Which is fine, except it is also what everybody already does, and 18.1% of English Notes in my sample contain a question mark while only 38% of all Notes get even one reply. So the advice is either wrong or badly underspecified.
It turned out to be badly underspecified. There is a version of a question that pulls replies at roughly double the rate of an equivalent Note with the same reach, and a version that performs no better than no question at all. The difference is mechanical and easy to copy once you can see it.
Not here for the nitty gritty? Scroll down to the TL;DR section to get the gist of the article
Finding 1: A question opener nearly doubles replies, even against Notes with identical reach
Across the full 655,056 Notes, question openers average 2.41 replies against 1.36 for the “other” opener category (n=32,452 question openers). That gap could easily have been a reach artifact.
It is not. Inside every single like bin, question-opening Notes pull between 1.8 and 2.6 times the replies of Notes with the same number of likes. Pooled across bins and weighted by bin size, English question openers average 10.83 replies against 5.57 for everything else at matched reach. That is a lift of 1.94.
Average replies at matched like counts, question opener vs everything else
The consistency is the part worth noticing. A confound that produced this would have to inflate replies by roughly the same factor at 3 likes and at 500 likes, which is not what reach artifacts look like.
Mechanism
A like is a one-tap exit. A reply requires the reader to compose something, which means the Note has to hand them a sentence they already know how to write. A question at the top of a Note does that before the reader has decided whether to keep scrolling, so the composition starts while attention is still on the screen. Every line after that is the reader rehearsing an answer they are already half-holding.
Finding 2: The question belongs in the last line, not the first
Once I split question-containing Notes by where the question actually sits, the opener stopped being the strongest position. Notes that end on a question average 11.38 replies at matched like counts. Notes that open on a question but end on something else average 9.96. Notes with the question buried in the middle average 7.33. Notes with no question at all average 5.24.
Average replies at matched like counts, by where the question sits
Restricting to multi-line Notes only, so that position is genuinely a choice, closing on a question beats putting it anywhere earlier by 1.46 times (n=1,137 against 5,477). Only 4.6% of English Notes in my sample end on a question mark, which makes this one of the cheapest unclaimed moves on the platform.
Mechanism
The last line is the one the reader is holding when their thumb reaches the reply box. A question at the top has to survive the whole body before it gets acted on, and by the last paragraph the reader has usually replaced it with whatever the Note actually argued. Ending on the question removes the gap between the prompt and the box.
Finding 3: Ask for something the reader can retrieve, not something they have to work out
This was the hypothesis I most wanted to break, because it is the one everybody states without evidence. The claim is that a question pulls replies when it leaves a gap the reader can fill from memory, and dies when it asks them to think.
It held. Among Notes that contain a question, questions that address the reader directly average 10.88 replies at matched like counts against 7.92 for abstract questions with no “you” in them, a lift of 1.37. And the hard version genuinely does nothing: questions opening with “why” or “how” average 0.94 times the replies of other questions at matched reach (n=730), which is to say no lift at all, and at 100 likes and above they run slightly behind.
Average replies at matched like counts, by kind of question
Narrowing further, questions in the closing line that ask for a specific retrievable item (”what’s your”, “which”, “name one”, “what are you”) run at 1.84 times the matched baseline (n=284, small, so treat the size loosely and the direction as solid).
Mechanism
“Which novels made you want to write fiction” and “how do we fix publishing” both look like questions. Only one of them has an answer sitting in the reader’s head already. The second one asks for an essay, and the reader correctly decides they do not have time to write an essay in a comment box, so they like it and leave. The gap has to be the size of a sentence the reader has said out loud before.
Finding 4: A link in the Note costs you a quarter of your replies
This one was not in my hypothesis list. English Notes containing a URL average 5.07 replies at matched like counts against 6.95 for Notes without one, a lift of 0.73. The penalty appears in nine of nine like bins, ranging from 0.52 at zero likes to 0.84 at 50 to 99 likes.
Reply rate of Notes containing a URL, relative to Notes without one at the same like count
Narrowing to the closing line specifically, Notes whose last line is a link or a subscribe pitch run at 0.81 of matched baseline. Nine percent of English Notes in the sample carry a URL, and they are systematically buying reach at the expense of conversation.
Mechanism
A link is an exit. It converts the Note from a place to say something into a doorway to somewhere else, and the reader who taps it is no longer in the feed to reply. It also changes what the last line is asking for. A closing URL asks for a click, and readers give the platform exactly one action per Note.
Finding 5: At the same reach, high-reply Notes convert about 20% better
Same method. Bin by likes, split each bin at its median reply count, compare subscriber acquisition between the two halves. Bin-weighted, high-reply Notes acquired 26.0 subscribers per 100 Notes against 21.7 for low-reply Notes with the same number of likes, a lift of 1.20. A permutation test that shuffles reply counts within like bins puts that at p = 0.031, so it is unlikely to be noise, but it is not a landslide either. Clicks moved by the same 1.20 factor.
Subscribers acquired per 100 Notes, high-reply vs low-reply half of each like bin
Two honest qualifications. The lift is consistent in direction across every bin I could measure but it shrinks as reach grows, from 1.93 in the 1-to-2 like bin down to roughly parity above 100 likes. And a reply half is not a randomised treatment, so some of this is that better Notes get both. The useful reading is modest: at the reach you actually have, a Note that starts a conversation is worth about a fifth more than a Note that does not.
Finding 6: The like cost is real but it lives entirely in the tail
Question openers do earn fewer likes. Across the full 655,056 Notes they average 14.6 likes against 18.4 for the “other” category. That is the trade the platform makes you take, and it is worth naming precisely, because “questions get fewer likes” is true in a way that does not mean what it sounds like.
The median question Note and the median non-question Note both get exactly 3 likes. The gap opens at the 90th percentile, 22 likes against 28, and widens from there into the averages. Question Notes are not disliked. They are less likely to produce a monster.
Likes for question openers vs everything else, at three points in the distribution
A within-author check says the same thing from a different angle. Among 344 authors who published at least two question-opening Notes and at least five others in July, the median author’s question Notes earned 0.99 times their own median likes and 1.46 times their own median replies. Roughly half of those authors saw no like penalty at all on their own feed.
What this means for the trade
If your goal is one Note that reaches 40,000 people, a question opener is the wrong instrument, because the ceiling is lower. If your goal is a Note that produces twenty conversations you can convert, the median outcome is unchanged and the reply count roughly doubles. Most writers below 5,000 subscribers are not competing for the tail, and are trading away something they were never going to get.
For completeness, Open Letter openers pull the highest reply rate of any category I measured, 4.22 replies against a 1.42 platform average, but on only 897 Notes in the entire month. The sample is too thin to build a habit on.
Six closers that pull replies
Each of these is a template abstracted from Notes in the sample that outperformed their like-matched peers. Swap the bracketed part for your own subject.
The constrained retrieval. “In one sentence, what’s your [thing] about?” A word limit removes the reason not to answer. Retrieval asks in the closing line run 1.84× matched baseline (n=284).
The lookup. “Which [book / song / mistake] made you [outcome]?” “Which” outperforms “why” because the answer is a noun, not an argument.
The small specific. “What are your favourite [very small everyday thing]?” The smaller the category, the more people have one.
The solidarity check. “Anyone else, or is it just me?” Worth 1.71× matched baseline on a thin sample (n=124), so treat the number loosely. The reply it invites is one word long.
The direct ask. “Tell me yours below.” Explicit ask verbs are worth 1.44× (n=2,787), and this is the version that works when your Note has no natural question in it.
The narrowing follow-up. Ask a broad question, then a narrower one underneath it. Notes with two or more question marks earn 1.66× matched baseline (n=2,468), because the second question tells the reader what shape of answer you meant.
TL;DR
Run this over the Note before you post it. Every line is a comparison at matched like counts, so each one is a real change in outcome rather than a change in reach.
Does the last line end in a question mark? Closing on a question is worth 2.05× the replies of a Note with the same likes (n=2,387). Only 4.6% of Notes do it.
Does that question contain the word “you” or “your”? Reader-addressed questions beat abstract ones by 1.37× (n=3,162).
Can the reader answer it from memory in under ten words? If it starts with “why” or “how,” it earns 0.94×, which is nothing (n=730). Rewrite it as “which,” “what’s your,” or “name one.”
Is there an explicit ask verb? “Tell me,” “drop yours,” “let me know,” “curious” are worth 1.44× (n=2,787).
Is there a URL anywhere in the Note? Take it out. Notes with a link run at 0.73× on replies (n=4,692). Put the link in the first reply instead.
Is the Note more than one line? Four or more line breaks earn 1.18× over shorter Notes at the same reach (n=13,051). Single-line Notes run at 0.84×.