A common way to write in Threads is the way one writes on a website: a useful list, a few points, a couple of emoji. It looks considerate. In our snapshot such posts had no edge over the rest, while posts that asked readers to share their experience had the most replies. Below is what we counted, how to read it, and what does not follow from it.

What we analysed

This is one short snapshot taken in September 2026, not months of observation. We collected 539 posts from one account's feed and from search under the “Recent” tab. The search ran on 8 queries across the types of business we work in. We read the counts of likes, replies and reposts from the labelled buttons under each post.

The analysis covered 183 posts aged 3 to 240 hours. Younger ones have not had time to collect reactions, older ones are no longer comparable with the rest. For each feature we computed the median of likes and replies, with the mean next to it where it matters. The median shows a typical post, while a mean is easily pulled up by one lucky case, so the two should not be confused.

What came out

Asking the audience for experience or advice - “any recommendations”, “how do you do it”, “who has dealt with this”, “share your story”. These posts had a median of 18 likes and 42 replies (mean 28 and 48). A post without such a request: a median of 0 likes and 0 replies (mean 13 and 16).

First person - “I”, “me”, “my”. With it, a median of 4 likes and 5 replies (mean 41 and 36); without it, 0 and 0 (mean 7 and 19).

A provocative claim - “unpopular opinion”, “actually”, “stop doing this”. A median of 12 likes and 28 replies; the mean for replies is 133, far above the median, so a small share of posts pulls it up. Without a claim, a median of 1 like and 0 replies.

A question mark ending the first line. A median of 10 likes and 14 replies against 1 and 0 without it.

A list of points or numbering. A median of 0 likes and 2 replies against 1 and 1 for the rest, so a list gave no advantage. A post with 6 or more emoji had a median of 1 like and 0 replies.

Length. Posts up to 150 characters collected the most replies: 33 on average. For posts of 300-600 characters the count recorded is 7. A post with two or three “?” had a median of 7 replies, one with none had 0.

Separately, about the posts with the most replies: among them were plain requests for recommendations, for example an online English tutor. Each of those posts got 40-70 likes and 70-120 replies. What follows is our view, not something measured: these are posts a business answers, so they interest us from both sides, as a model of wording and as a place to show up.

How to read this

This is an observation, not an experiment. We did not split identical posts in half or change one feature at a time. Features can overlap: a first-⁠person post can also ask for advice. We did not record per-⁠group sizes, so the comparisons are indicative: in small groups a median easily comes out random.

There is also a bias in the selection itself. The feed belongs to one account, so it is picked by the algorithm for that account. The search ran on 8 business queries, which could pull in posts from people looking for advice; we did not compare the snapshot with a live feed. The conclusions apply to this context: one snapshot, one feed, eight topics. So we do not say that a question mark “produces” replies. We say that in this snapshot posts that addressed the reader collected replies noticeably more often than the rest.

Why it happens, we do not know. A plausible explanation is simple: a question is easy to answer, and a list leaves nothing to add. But that is an explanation, not a measured result.

What to do with it

  • Open with a question from the reader's world. Not your product, but a situation where a person has something to tell: a choice, daily life, a trip, a child's education.
  • Ask for experience, not a vote. This is our recommendation: a poll with options A, B and C feels like a form, not a conversation, and we did not measure it separately. The question “who has dealt with this and how did you solve it?” belongs to a feature we did measure.
  • Keep it short. In the snapshot, posts up to 150 characters did best; a 150-200 character limit and paragraphs of one or two sentences are our recommendation, there is no measured threshold here.
  • Write as yourself. In our snapshot the first person won on both likes and replies.
  • Do not count on a list. It is not forbidden, but in the snapshot it gave no edge.
  • The brand answers in the comments, not in the post itself. Once replies appear under a question, that is the place to help with real expertise. That is where the conversation starts, and a customer can grow out of it.
  • Check on your own account. After two weeks compare the median replies per post before and after, and the share of posts with no reaction at all. Your numbers matter more than someone else's.

What this means for finding customers

This is our conclusion, not part of the measurements. The same logic works in reverse. A person who writes “recommend who to go to” has already done what we advise accounts to do: asked a question that invites an answer. A reply to such a post is short, to the point and from a real person. That is the reply we help prepare.