LinkedIn dwell time claims are unverified

Every LinkedIn growth guide written in the last few years agrees on the same mechanism: the feed measures how long you stop scrolling, dwell time is the dominant signal, and therefore you should write posts that force people to hold still. Go looking for LinkedIn's own statement of that and you will not find one. Not on the help centre, not in the engineering blog, not in the press material.

That is not the same as saying the claim is false. It is saying the citation does not exist. Worth knowing which of your operating assumptions sit in that gap.

What LinkedIn actually publishes about video

The documented surface is narrower than most people assume, and it is almost entirely mechanical rather than algorithmic.

Formats and limits. LinkedIn publishes three first-party spec sheets, and they are not the same document. Share videos governs a normal feed post: 15 minutes maximum, minimum 3 seconds from desktop and 2 seconds from the app, resolution 256x144 to 4096x2304, aspect 1:2.4 to 2.4:1. Page and Career Page video specifications cap duration at 10 minutes, with the same resolution and aspect window. Video ads specifications list 16:9, 1:1, 4:5 and 9:16, bound 9:16 between 360x640 and 1080x1920, and allow 3 seconds to 30 minutes. The widely repeated "15 minutes on desktop, 10 on mobile" split as a maximum is not on the organic sheet. That page's device split is the minimum. The 10-minute figure lives on the Pages sheet.

Provenance labelling. LinkedIn reads C2PA Content Credentials and shows an icon on images and video that arrive cryptographically signed. Clicking it reveals whether AI was used, which tool produced the asset, and when the credential was issued. LinkedIn states plainly that it is not yet possible to identify and label all AI-generated and modified content. There is no creator-facing disclosure toggle, and no stated reach or eligibility penalty attached to the label. It is passive metadata pass-through. Our note on why C2PA manifests die on upload covers the part of that chain most teams break without noticing.

Monetization. There is no organic creator ad-revenue share on LinkedIn. Monetization is sponsorship and lead generation. This has a useful consequence: there is no demonetization risk to manage on this platform, only reach. Every optimisation decision is a distribution decision.

Ranking. Nothing. No published weighting, no stated signal hierarchy, no video-specific guidance on what the feed rewards.

What a documented ranking signal looks like

The contrast is the point. Other platforms in the same competitive set publish exactly this.

Instagram's ranking explanation names its top predictions for Reels in order: how likely you are to reshare the video, how likely you are to watch it all the way through, how likely you are to like it, and how likely you are to tap through to the audio page. That is a specific, first-party statement. Sends-per-reach is a named lever.

YouTube's search and discovery tips are similarly first-party: videos are ranked on performance and viewer personalization, including whether recommended viewers watch, how long they stay, and whether they enjoyed the video. Separately, YouTube's Trending Charts page says charts aim to surface videos that grow quickly and that are not misleading, clickbaity, or sensational. That is Charts guidance, not a Home-feed formula.

Meta even publishes a ranking page for the Threads feed, describing it as a machine-learned mix of content from accounts you follow and accounts you don't.

None of these are complete. All of them are more than LinkedIn offers. When Instagram, YouTube, and Threads each publish a ranking page and LinkedIn does not, the absence is information.

Why the dwell-time story is so durable

Three things keep an unsourced claim alive.

It is plausible. Feed products do measure engagement, and time-on-post is an obvious thing to measure. Plausibility is not evidence, but it feels like it.

It is unfalsifiable in practice. You cannot run a clean experiment against a ranking system you cannot observe, with a sample size of one account, in a feed that also varies by viewer, time, network and topic. Any result confirms the theory.

And acting on it produces content that would be good anyway. "Write a strong first line, front-load the substance, make the reader want the next sentence" is correct advice regardless of whether dwell time is a weighted feature. The claim survives partly because its prescription works for reasons that have nothing to do with its mechanism.

Believing the dwell-time story will not ruin a LinkedIn presence. The next unsourced mechanism might: post at 7:41am, drop links from the body, pad a video to a specific runtime. Those prescriptions are not harmless, and they come from the same place.

Separating documented from inferred

Run every claim you act on through three questions.

1. Who published it? A platform help page, engineering blog or newsroom post is a source. A tool vendor's blog citing another tool vendor's blog is not. Ask for the link, then open the link.
2. Is it a mechanism or a result? "Dwell time is weighted heavily" is a mechanism claim about a system you cannot see. "Our posts with a first-line hook outperformed our posts without one" is a result claim about your own account. The second is verifiable and the first is not, and only the second should carry weight.
3. Would you still do it if the mechanism were false? If yes, it is craft advice wearing an algorithm costume, and you can adopt it without adopting the theory. If no, you are betting your calendar on someone's guess.

Working under an undocumented ranking system

You cannot reverse-engineer the ranker. You can measure your own account against itself.

Set a channel-relative baseline. Take your last 20 posts of a given format, record impressions, reactions, comments and click-throughs, and take the median. That median is your zero. Every subsequent post is a delta against it, not against some benchmark from a report about a different industry.

Change one variable per test, and hold it for at least ten posts. Opening line structure, video length, whether the video carries burned-in captions, whether the asset is 4:5 or 9:16. One at a time. A week is not a sample.

Measure what LinkedIn shows you, not what a blog claims it measures. Impressions, reactions, comments, reposts and profile views are reported. Dwell time is not. Optimising for a metric you cannot see means optimising for a story about a metric you cannot see. The framework in brand video metrics that matter is a good place to fix a definition list before you start logging.

Separate reach from outcome. On a platform with no revenue share, impressions are not the goal. Meetings, replies, and qualified inbound are. Our piece on getting clicks off LinkedIn without a link deals with the mechanical side of that, and B2B video distribution beyond LinkedIn covers what to do when the platform's ceiling is lower than your pipeline needs.

Write down what you concluded and when. Six months later, nobody remembers whether "we stopped using 16:9" came from a test or from a conference talk. A one-line decision log with a date and a reason is what stops inherited folklore from calcifying into house style.

Spend that effort on the things that hold across every documented feed: the first frame decides whether the second happens, a forwarded post travels further than a liked one, and a consistent format beats a clever one-off. Those are properties of how people read. A LinkedIn content engine for teams and the company page video playbook start from a weekly format, not from a ranking theory.

FAQ

So is dwell time a real LinkedIn ranking factor or not?

Unknown, publicly. LinkedIn has not published a statement on it that I can point you to. If someone tells you it is confirmed, ask for the first-party link. Third-party inference repeated often enough starts to sound like documentation, and that is the whole problem.

Should I stop writing hooks then?

No. A strong opening line is good writing, and good writing works whether or not a specific feature weight exists. The change is in how you justify it. Adopt it as craft, not as algorithm compliance, and you stop being vulnerable to the next unsourced mechanism that comes along.

Does the C2PA label on my AI-generated video hurt reach on LinkedIn?

LinkedIn publishes no reach penalty for labelled content and gives creators no toggle either way. The label appears when signed provenance metadata survives to upload. LinkedIn also says it cannot detect all AI-generated content, so its presence is inconsistent by the platform's own admission. See AI content label for what the marker communicates versus what people assume it does.

How long should a LinkedIn video be?

Organic feed posts: 15 minutes, on the share-videos page. Pages: 10 minutes. Ads: 30 minutes. The 15/10 device split as a maximum is not on the organic page; that page's device split is the minimum. Measure completion rate against your own median before you change length.