What is the Foldline Benchmark?

The Foldline Benchmark is an open, versioned standard for judging whether a LinkedIn post is ready to publish. Version LI v3.0.0 defines 26 weighted tests. 18 apply to every post and are worth 119 points; the remaining 8 are scoped to the format being posted, so a video is scored out of 147 points and a text post out of 119. 3 tests are blocking: fail any one and the post is not ready regardless of the score.

What are the blocking tests?

LI-05, external link discipline: no external link in the post body. LI-06, no engagement bait: no phrasing such as "comment YES" or "agree?". LI-26, character limit: a post over 3,000 characters cannot be published at all. The first two are behaviours LinkedIn suppresses directly; the third is a hard platform limit. No amount of craft elsewhere compensates for any of them.

What do the evidence grades A, B and C mean?

Grade A means the test measures something directly observable, such as character count before the fold. Grade B means the rule is corroborated across multiple published 2026 breakdowns. Grade C means the rule is directional: the direction is agreed, the magnitude is not.

How is the Foldline score calculated?

Each applicable test returns pass, warn or fail. A pass earns full weight, a warn earns half, a fail earns none. The total earned is divided by the points that apply to that format, not the whole catalogue, and expressed as a percentage. 85 percent or above is Ship it, 60 to 84 is Tighten it, below 60 is Rework it, and any blocking failure overrides all of them.

Open standard · free to cite

The Foldline Benchmark — LinkedIn v3.0.0

The Foldline Benchmark is an open, versioned standard for judging whether a LinkedIn post is ready to publish. Version LI v3.0.0 defines 26 weighted tests. 18 apply to every post; 8 more are scoped to the format you are actually posting.3 are blocking: fail any one and the post is not ready, whatever the score.

Version: LI v3.0.0
Published: 2026-09-01
Tests: 26 (18 always-on)
Blocking: LI-05, LI-06, LI-26
Total points: 119–167 by format
Engine: v1.0.0
Licence: CC BY 4.0
Maintainer: DB23

01What this measures

LinkedIn’s 2026 feed does not rank by keywords or hashtags. It ranks with a large language model that reads each post for meaning, intent and topical relevance, and matches it to the specific readers most likely to find it useful. Reach is narrower and more relevant than it was — by design, not by failure.

The benchmark therefore measures one thing: whether a relevant reader will stop, stay, and respond with something real. Every test below serves that outcome. Nothing here measures vanity reach.

The benchmark is a publishing gate, not a prediction. It tells you whether a draft is ready and what is costing it distribution. It does not forecast impressions, and no honest tool can.

Tests split into three groups. Craft and compliance (LI-01 to LI-11) covers how the post is built and the behaviours the platform suppresses outright. Interest graph (LI-12 to LI-17) covers how the ranking model reads the post, its author, and who it should reach. The asset (LI-18 to LI-25) covers the image, document or video attached to it.

Scoping: a test only counts when it applies

The first 18 tests apply to every post. The 8 asset tests apply only to the format being posted — a text-only post is never marked down for lacking a video hook, and a video is never asked how many slides it has.

Tests that do not apply are excluded from the denominator as well as the result, so the percentage stays comparable across formats.

Format	Asset tests that apply	Points
Text-only, poll, link-out	—	119
Newsletter	LI-18, LI-20	127
Image + text	LI-18, LI-19, LI-20	133
Document / carousel	LI-18, LI-20, LI-21, LI-22	141
Native video	LI-18, LI-20, LI-23, LI-24, LI-25	147

The asset tests score what you declare, not the file. Foldline reads text; it cannot open your MP4 or look at your cover slide. Every asset test is a stated intention, and the report labels it as one. A benchmark that pretended otherwise would be lying about its own reach.

02How scoring works

Every test returns one of three states, and each earns a fraction of that test’s weight:

Pass — full weight.
Warn — half weight. The rule is not broken, but reach is being left on the table.
Fail — no weight.

Earned points are divided by the points that applied and expressed as a percentage. Weights are not uniform: originality (LI-03, 12 points) carries six times the weight of hashtag discipline (LI-09, 2 points), because that is roughly the ratio of their effect.

Verdict bands

Verdict	Score	Meaning
Fix before publishing	any	A blocking test failed. Overrides every other result — nothing compensates.
Ship it	≥ 85%	Both blocking tests pass and the craft rules hold.
Tighten it	60–84%	Publishable, but working the fix list first is worth more than posting sooner.
Rework it	< 60%	Below the bar on too many tests. Rebuild around one idea, one reader, one takeaway.

Two overrides apply. A blocking failure forces Fix before publishing at any score. A draft under 50 words forces Rework it: a clean stub is not a shippable post, and the benchmark refuses to let an empty draft score well by passing the declarative tests.

03The 26 tests

Test IDs are stable and citable. A retired test keeps its ID and is marked retired; new tests take new IDs. Weights may change between minor versions, IDs never do.

Craft & compliance

How the post is built, and the behaviours the platform suppresses outright.

LI-01

Hook lands in the first 1-2 lines

Pass: The opening 1–2 lines run under 220 characters and carry a number, a tension or a contrarian claim.
Fail: The post opens with throat-clearing (“excited to share…”), so the hook falls below the fold.
Why: The fold is the only gate that matters. Nothing below it is read unless the reader taps.

10 ptsA

LI-02

One clear idea, audience and takeaway

Pass: One idea, declared by the author, and under 320 words.
Fail: The author flags more than one idea in the post.
Why: The ranking model matches a post to readers by topic. Two topics halve the match quality of both.

5 ptsB

LI-03

Original, first-hand insight

Pass: First-person voice, two or more concrete numbers, or a named framework — and no filler phrases.
Fail: Two or more generic-marketing filler phrases, or no first-hand signal at all.
Why: Recycled thought-leadership earns no dwell, and repeated low-value posting erodes the account's topical authority.

12 ptsB

LI-04

Format fits the job

Pass: Format suits the job: a document for a framework, video for a demo or story, text for a sharp opinion.
Fail: A link-out post, which is throttled at the format level.
Why: Format sets the ceiling before a word is read. Documents and native video hold dwell in a way text cannot.

10 ptsB

LI-05

External link disciplineBlocking

Pass: No external link in the body. Any link is parked in the first comment and signposted in the post.
Fail: One or more external links in the post body.
Why: A body link sends the reader off-platform. Published estimates put the cost between a fifth and half of median reach.

10 ptsA

LI-06

No engagement baitBlocking

Pass: No bait phrasing, and the post closes on a question a real reader could answer.
Fail: Bait detected: “comment YES”, “agree?”, “tag someone”, “like if”, and similar.
Why: Bait is suppressed directly by the platform. It is one of the few behaviours that is penalised rather than simply ignored.

8 ptsA

LI-07

Built for the thumb

Pass: Three or more blocks, none longer than 240 characters.
Fail: A block over 400 characters — a wall of text on a phone.
Why: Almost all reading happens on a phone. Density costs dwell before content ever gets a chance.

5 ptsB

LI-08

On one of this account's 2-3 themes

Pass: The post sits on one of the account's 2–3 defined themes.
Fail: Off-theme.
Why: Topical authority compounds per account. Hopping stops the model recognising the account as an authority on anything.

8 ptsB

LI-09

0-3 relevant hashtags, no stacks

Pass: Between zero and three hashtags, not stacked.
Fail: Four or more hashtags, or a stack at the end.
Why: Hashtags barely affect distribution now that ranking is semantic. Stacks read as spam without buying reach.

2 ptsB

LI-10

Right account and posting time

Pass: Personal profile, weekday morning in the audience's timezone.
Fail: Company page and off-peak.
Why: Personal profiles consistently out-reach company pages, and early engagement is easier to earn when the audience is online.

5 ptsB

LI-11

Present for the first 60-90 minutes

Pass: Someone is committed to replying to every comment in the first 60–90 minutes.
Fail: No reply plan.
Why: The first 60–90 minutes are the test the model runs before widening reach. Substantive replies count and lift the thread.

7 ptsB

LI-26

Within LinkedIn's character limitBlocking

Pass: The post body is 3,000 characters or fewer.
Fail: The body exceeds 3,000 characters.
Why: A hard platform limit, not a guideline. The composer refuses a longer post outright, so no other score means anything.

6 ptsA

2026 interest graph

How the ranking model reads the post, its author, and who it should reach.

LI-12

Profile matches what you post about

Pass: Headline and experience line up with the post's topic.
Fail: Profile does not match what the post is about.
Why: The model scores post and author together. A mismatch caps distribution before anyone reads a word.

6 ptsB

LI-13

Names the reader it is for

Pass: A specific role is named and the reader is addressed directly.
Fail: Nobody is named and no second person is used.
Why: Distribution is interest-based. An unaddressed post gives the model no audience to match it to.

5 ptsB

LI-14

Knowledge and advice, not an update

Pass: The post teaches something — a how-to, a lesson, a framework, a mistake.
Fail: Pure announcement or promo with nothing to learn.
Why: Advice content is the class that still earns multiples of baseline reach; company news is the class demoted hardest.

6 ptsC

LI-15

Length in the dwell sweet spot

Pass: Between 800 and 1,000 characters.
Fail: Under 500 or over 1,600 characters.
Why: Long enough to earn dwell, short enough that readers finish. Unfinished posts leak the signal the model is measuring.

4 ptsC

LI-16

Reads human, not model-generated

Pass: No obvious model tells.
Fail: Three or more tells: em-dash rhythm, “not X, it's Y”, model vocabulary, no numbers, uniform sentence length.
Why: Generic model prose earns no dwell, so the feed stops distributing it — and the account's authority erodes with it.

5 ptsB

LI-17

Cadence builds topic authority

Pass: Three or more posts a week for eight weeks or more, on the same pillars.
Fail: Sporadic posting.
Why: Baseline reach per post is an account-level property built by consistency. It is worth more than any single post.

5 ptsB

The asset

Scored only for the format you are posting. Declared by you — Foldline reads text, it cannot open your file.

LI-18

Alt text written for the asset

Pass: Alt text of five words or more describing what the asset shows.
Fail: No alt text at all.
Why: Alt text is the only part of an image or video the ranking model can read, and the only version of it a screen-reader user gets.

4 ptsA

LI-19

The image carries meaning

Pass: The author confirms the image carries information the text does not.
Fail: The image is decorative.
Why: A stock photo takes the space a second paragraph could have used and adds no reason to stop.

6 ptsB

LI-20

Aspect ratio owns feed height

Pass: 4:5 for documents and images, 9:16 or 1:1 for video, 16:9 for a newsletter header.
Fail: 16:9 on a feed post — it wastes vertical space on a phone.
Why: Feed height is attention. A landscape asset gives away half the screen it could have held.

4 ptsB

LI-21

Cover slide carries the hook

Pass: The cover slide states the claim, the number or the tension.
Fail: The cover is a title card or a logo.
Why: Most people see only the cover. It is the thumbnail and the hook in one frame.

8 ptsB

LI-22

One idea per slide, legible on a phone

Pass: One idea per slide, and 5–10 slides.
Fail: Slides carrying more than one idea each.
Why: Every swipe is a fresh decision to continue. A crowded slide ends the sequence.

6 ptsB

LI-23

Hook in the first 3 seconds

Pass: The claim lands inside the first three seconds.
Fail: The video opens on an intro, a logo sting or a slow build.
Why: Autoplay gives you three seconds before the thumb moves. Spending them on branding spends all of them.

8 ptsB

LI-24

Captions burned into the video

Pass: Captions are burned into the video file.
Fail: No burned-in captions.
Why: Most video in the feed is watched with no sound. Without captions the audience watches a mute clip and scrolls.

7 ptsA

LI-25

Video length in the retention band

Pass: Between 30 and 90 seconds.
Fail: No length declared, or over three minutes.
Why: Completion rate is the signal video is judged on, and it collapses with length.

5 ptsC

04Reach index

Alongside the score, Foldline reports a reach index: the product of a set of multipliers, expressed against a clean baseline post from the same account. An index of 1.40× means the draft should out-reach that account’s own baseline by roughly forty percent. It is not a prediction of impressions, and it is clamped to 0.20×–2.20× so it never reads as precision it does not have.

Class	Factor	Multiplier	Note
Format	Document carousel	×1.40	Highest-reach format; holds dwell on every swipe.
Format	Native video	×1.30	Captions burned in, hook in the first 3 seconds.
Format	Image + text	×1.05	Only when the image adds meaning.
Format	Newsletter	×1.00	Bypasses the feed into the inbox.
Format	Poll	×0.95	Reach has fallen sharply; not an engagement shortcut.
Format	Text-only	×0.90	Works when sharp and short.
Format	Link-out	×0.55	Throttled at the format level.
Penalty	Link in the body	×0.65	Published range runs roughly 0.80×–0.50×.
Penalty	Link in comment 1	×0.95	The cheapest way to carry a link, though not free.
Penalty	Engagement bait	×0.55	Suppressed directly.
Penalty	Reads model-generated	×0.70	Three or more AI tells.
Penalty	Some AI tells	×0.90	One or two tells.
Penalty	Profile–topic mismatch	×0.85	Author and post scored together.
Penalty	Off the account's themes	×0.85	Breaks compounding authority.
Penalty	Company page, not a person	×0.75	Pages under-reach profiles.
Penalty	Hook doesn't hold	×0.75	Throat-clearing above the fold.
Penalty	Soft hook	×0.90	Clear but flat opening.
Account	3+/week, 8+ weeks	×1.15	Consistency lifts baseline reach per post.
Account	1–2 posts a week	×1.00	The baseline.
Account	Sporadic posting	×0.85	Keeps resetting topical authority.
Account	Author present in hour one	×1.10	Replies in the first 60–90 minutes.

Format multipliers are relative to each other, not absolute. The link-in-body figure of 0.65× sits deliberately in the middle of the published range — estimates run from roughly 0.80× to 0.50×, and the benchmark does not pretend to know which is right.

05Signal hierarchy

Depth of attention now outweighs volume of taps. The tests are weighted to serve this order:

Saves — the strongest signal, reportedly around five times a like. Frameworks, checklists and reference lists earn them; opinions rarely do.
Substantive comments — multi-sentence replies that add a perspective. A handful of real comments beats hundreds of reactions.
Dwell time — how long someone reads or watches before scrolling. The core passive signal, and what hooks, white space, carousels and video all ultimately serve.
Reshares with a genuine take — a reshare with commentary, not a bare repost.
Reactions — baseline only. Do not optimise for these.

The first 60–90 minutes decide distribution. Early meaningful engagement is the test the ranking model runs before widening reach. Plan to reply to every early comment in full sentences — replies count, and they pull the thread back up.

06Method, evidence and limits

Every test carries an evidence grade. This is unusual and deliberate: a benchmark that hides its own confidence is worth less than one that states it.

Grade	Means	Tests	Example
A	Directly observable in the text or the declaration. Measured, not inferred.	6	LI-01 — characters before the fold.
B	Corroborated across multiple independent published 2026 breakdowns.	17	LI-12 — profile–topic alignment affects distribution.
C	Directional. The direction is agreed; the magnitude is not.	3	LI-15 — the 800–1,000 character dwell band.

What this benchmark cannot do

It cannot predict impressions. No external tool can. Distribution depends on your specific graph, your history and who happens to be online.
It cannot read your images, video or slides. The asset tests score what you declare, not the file.
It cannot judge whether your idea is any good. A well-built post about nothing still scores well on craft. LI-03 catches filler, not banality.
It cannot see your audience. Timing and theme fit are declared by you, not verified.
Multipliers are modelled, not measured. They encode published estimates and are revised as better data arrives.

Roadmap for evidence

This version is built on published breakdowns of the 2026 ranking changes plus direct observation of the composer. It is not built on a dataset, and it does not claim to be. Foldline is collecting post-publication outcomes — score against actual reach — and a correlation study will be published under this benchmark when the sample is large enough to mean anything. Until then, every grade C claim should be read as a working assumption.

07Benchmark family

The scoring engine is platform-agnostic. A platform is defined by its own test set, weights, fold length and multipliers; the engine that runs them is shared. Test IDs are namespaced by platform, so LI-05 and a future X-05 are unrelated tests that never collide.

Benchmark	Prefix	Tests	Status
LinkedIn	LI-	26	v3.0.0 live
X / Twitter	X-	—	Planned
Instagram	IG-	—	Planned
Email subject lines	EM-	—	Planned
YouTube titles	YT-	—	Planned

Planned means exactly that: designed for, not built. No date is promised, and nothing here should be read as available.

08Changelog and versioning

The benchmark is versioned PLATFORM vMAJOR.MINOR.PATCH. A major bump adds or retires tests. A minor bump changes weights, thresholds or bands. A patch corrects wording or a detector without changing any score. Citations should always name the full version, because scores are not comparable across major versions.

LI v3.0.0 2026-09-01

Added LI-26, the character limit, as a third blocking test. LinkedIn refuses a post over 3,000 characters outright, so an over-long draft can never be publishable however well it scores elsewhere.
Until this version the limit was only a warning shown when copying, so a 3,500-character post could return a passing verdict for something the composer would not accept. A defect, found by a reader running real content through the tool.
LI-26 also warns when the first comment exceeds its own 1,250-character limit.
Major bump: the catalogue gained a test and every denominator moved up by 6 points.

LI v2.0.0 2026-09-01

Added the asset group: LI-18 to LI-25, covering alt text, aspect ratio, image purpose, carousel cover and slide discipline, video hook, captions and length.
Tests are now scoped by format, and non-applicable tests are excluded from the denominator as well as the result. Text-post scores were unchanged at that version and remained out of 113.
Major bump, per the policy above: the catalogue gained tests, and carousel and video scores from v1.0.0 are not comparable to v2.0.0. Text-only scores are.
Asset tests are marked as declarations throughout. Foldline does not read media files and does not claim to.

LI v1.0.0 2026-09-01

First public release. 17 tests, 113 points, two blocking.
Reach index introduced, clamped to 0.20×–2.20×.
Evidence grades A/B/C published for every test.
Native video corrected to 1.30× after review — an earlier internal draft ranked it below text-only, which contradicted the format guidance.
Link-in-body multiplier set at 0.65× with the published range disclosed rather than a single asserted figure.
Substance floor added: drafts under 50 words cannot return a passing verdict.

09How to cite this

The benchmark is published under CC BY 4.0. Quote it, table it, teach from it, build on it — attribution is the only condition. If you extend it, say which version you started from.

DB23. (2026). The Foldline Benchmark — LinkedIn v3.0.0. Retrieved from https://db23.co.za/foldline/benchmark/

Citing a single test

Foldline Benchmark LI v3.0.0, test LI-05 (External link discipline, blocking, 10 pts).

For corrections, disputed figures or evidence that would move a grade, write to DB23. Substantive corrections are credited in the changelog.

10Run a draft against it

The benchmark is free to read and free to cite. The tool that runs a draft against it is free to score.

Score a post free

Independent standard by DB23. Not affiliated with, endorsed by, or connected to LinkedIn Corporation. Reach figures are directional models, not guarantees.