How Wren labels your posts

What's working groups your published posts by their characteristics and compares how each group performed. Instead of generic advice about hooks and formats, you see how your narrative posts did against your listicles, on your own audience.

To do that, Wren reads each post and labels it along a fixed set of axes: what kind of post it is, how it's built, how it opens, what it asks for, what it shows. Those labels are what you see on a post's analytics page, and they're what the groups in the patterns view are built from. This article defines every one of them.

What's working isn't included in every Wren plan. The pricing page shows which plans have it.

How a label gets made

Wren analyzes a post once, shortly after its analytics are captured, and stores the result. The text axes come from reading the post's words; the visual axes come from looking at its images. Each axis is judged independently, so one label never decides another: a build-in-public post can be an announcement, a personal story, or a lesson, and Wren records both facts separately.

When Wren can't tell, it says so. Axes it couldn't determine come back as Unknown or Unclear, and those are hidden rather than guessed. They don't appear on the post's labels and they never form a group in the patterns view.

If your analytics came from a partial text capture, the post's label card says so. The read is based on the text Wren actually had.

Text axes

Post type

The post's primary communicative move.

Label	What it means
Personal story	Something that happened to you, told as the point of the post
Lesson	A takeaway or principle, whether or not it's carried by a story
Commentary	Opinion or analysis of the world, with no personal narrative
Announcement	News about you, your product, your company, or your life
Promotion	Content whose main purpose is selling or driving signups
Building in public	Sharing your own in-progress work — progress, metrics, decisions, misses
Other	None of the above fit

This is the axis people argue with most, because real posts mix. Three rules settle it. A post whose main move is showing your own in-progress work is Building in public, even when it's shaped like an announcement — a founder walking through a new feature of their own product is building in public, not promoting. Otherwise, a post that tells your own story to deliver a takeaway is a Lesson if the takeaway dominates and a Personal story if the narrative dominates. And asking your audience a question is not a post type: engagement asks belong to the Ask axis, so a commentary that ends with "what do you think?" is still Commentary.

Structure

The post's form, independent of its subject.

Label	What it means
Story	A chronology — something unfolds, beginning to end
List	A numbered or bulleted list dominates
Essay	Flowing text making a point or argument — no chronology, no list
Mixed	More than one of the above, with none dominant

First-person story

Yes or No: does the post contain a concrete first-person story or experience, something that actually happened to you? This is narrower than Post type. A lesson that opens with a story from your last job is Post type Lesson, First-person story Yes.

Opening

The structural device of the first line only. Nothing later in the post affects this label.

Label	What it means
Question	The first line asks something
Number/stat	The opener leads with a figure
Direct claim	A bold assertion
Story open	Dropping straight into a scene or moment
Announcement	Leading with news
Other	A device that isn't one of these

Two tie-breaks matter here. An opening question is Question even when it contains a number ("Why did 40% of them churn?"), and an opener that leads with the figure is Number/stat. If the first line is too short or too ambiguous to place, Wren returns Unclear and the label is hidden.

Ask

What the post asks the reader to do, at the close.

Label	What it means
None	No ask
Asks the audience	Invites replies, opinions, or experiences
Direct ask	Follow, sign up, DM, register, buy

Topics

Up to five lowercase subject-matter labels describing what the post is about, like "fundraising", "hiring", or "ai agents". Wren reuses labels already in your vocabulary before minting a new one, so you get "ai agents" consistently rather than a scatter of "AI", "genAI", and "artificial intelligence". Topics are subjects only. Words that describe the form of a post, like "storytelling" or "announcement", never become topics because those axes already exist.

Media axes

Media

What the post carried, counted rather than judged: Text only, Single image, Carousel (two or more images — the count shows on the post's own page), Document, or Video. This is deliberately separate from Visual style: a carousel of five charts is Carousel plus Chart, never one blended label, so you can see whether it's the carousel format or the charts that travel. In charts, every carousel is one group regardless of its count. When the capture never observed a post's media, there's no media label at all, because unknown is not the same as zero.

Visual style

What the post's primary image is. Later images only inform a Mixed call.

Label	What it means
Real photo	A camera photograph of the real world
AI illustration	AI-generated art or illustration
Graphic	A designed graphic, meme, or quote card
Screenshot	A screenshot of an app, UI, or document
Chart	A data visualization
Mixed	Multiple images of clearly different kinds
No visual	The post had no image to analyze

Two things shape this label. Images generated in Wren are known to be AI without anything having to look at them, so they're marked AI illustration from the record itself. And when Wren looks at an image but can't confidently separate a real photograph from an AI-generated one, it degrades to Unknown instead of committing to a coin flip. An unlabeled visual style usually means exactly that.

What Wren saw in the image

Alongside the visual style, Wren records two small facts about the primary image: how many people are visible, and whether the image contains text. They appear on the post's labels so a wrong read is obvious at a glance.

Length

Length isn't classified, it's counted from the post's words: Short is up to 150 words, Medium is 150 to 350, Long is more than 350.

Reading the numbers

The summary at the top of What's working goes further than the charts: nothing appears there unless it survives a statistical shuffle test. How Wren decides a pattern is real explains that test in plain language.

Medians first — and what the mean is for

The default number everywhere in What's working is a median. One post that unexpectedly reached 80,000 people would drag an average up far enough to make a whole group look like a winner. The median describes the middle post in the group, which is a better answer to "what usually happens when I write this kind of post".

The comparison chart also lets you flip to the mean (and the total). That's not a different opinion about statistics — it's an outlier detector. When a group's median and mean sit close together, the group is evenly strong. When the mean is far above the median, one post carried the group, and that's a completely different lesson from "this format works for you". The gap between the two numbers is the finding.

Wren still doesn't publish percentage deltas between groups. At the sample sizes involved, "31% better" would be a made-up precision. The one ratio you will see is on the best-posts view: how much more often a label appears among your top posts than across your whole corpus — and anything under 1.25× is omitted entirely rather than drawn small, because a barely-there bar reads as a finding it isn't.

Three lenses that can disagree

Each group is shown against three measures at once, with no default and no winner:

• Reach — median impressions
• Engagement — median engagement rate
• Profile views — median profile views

They measure different things and they often point different ways. A contrarian take can travel a long way and convert nobody; a quiet, specific build-in-public post can reach a fraction of the audience and send far more people to your profile. That disagreement is the useful part. A single blended score would hide it.

At least three posts per number

A group needs at least three posts with that metric captured before its number appears for that lens. Below that, you see a dash. The count next to a group is its total size, so a group of four can still show a dash for one lens if only two of those posts had profile views captured.

The whole view stays quiet until at least five of your analyzed posts are eligible, and tells you that's what it's waiting for.

Posts under seven days old are excluded

A post keeps accumulating impressions for days after it goes out. Comparing a two-day-old post against a two-month-old one measures age, not quality. So posts younger than seven days sit out of the groups and are listed separately as still settling. They join automatically once they age in.

When labels get upgraded

If Wren's analysis is improved, every post is re-analyzed and the view shows Recomputing until that finishes. Groups are never built from a mix of old and new analysis, because half-upgraded labels would produce comparisons that mean nothing.

Where you see this

Two places. On a single post's analytics detail page, How Wren read this post shows that post's labels, its topics, and what Wren saw in its image. And the What's working tab on the Analytics page turns those labels into views: a comparison chart (group by one label, split by another), the posts behind each group as dots, your best posts and what they have in common, how your mix of post types has changed over time, and any two labels crossed in a grid.

The charts don't group by every axis. They compare post type, opening, structure, first-person story, media, visual style, and length. The Ask axis is recorded per post and shown on the post's own page.

What Wren does not claim

Wren reports what its groups did. It doesn't claim to know why, and it won't tell you to write more of anything.

Nothing here is a causal claim. If your listicles show a higher median reach, that is a description of eleven posts you happened to write, not evidence that the list format caused the reach. You wrote those posts about particular subjects, at particular moments, to an audience that was growing the whole time. Wren can't separate those, and it doesn't pretend to.

Small groups are characterized, not measured. Three posts is enough to stop the number being noise-on-noise, and it is nowhere near enough to be a finding. Read a small group as a hint worth testing.

And a label is a judgment. Wren decides whether your takeaway outweighed your narrative, or whether an image is a photograph or a good render, and sometimes it gets that wrong. There's no way to correct a label today, which is exactly why the per-post view exists: check a few of your own posts before you trust the aggregate. If the labels look wrong on posts you know well, treat the groups with the same suspicion.