The LinkedIn algorithm and outbound: what LinkedIn says, what is folklore, what to measure

LinkedIn has published research on how its feed retrieves and ranks posts, using language models to match posts to members' interests, but it has not published a list of signal weights or a magic posting time. Claims about golden hours, comment multipliers and dwell time formulas come from third parties and are unverified. For outbound, the algorithm matters less than your own funnel: acceptance, replies, meetings, and the health of your account.

Every few months a new "LinkedIn algorithm update" post claims to have cracked the formula. This one takes a narrower line. It sets out what LinkedIn itself has put in writing, marks the popular claims we could not verify, and then turns to the part you control if you do outbound. The research statements below come from LinkedIn authored material, read on 25 October 2026.

What has LinkedIn actually published?

The best primary source we found is a research paper by LinkedIn authors, "Large Scale Retrieval for the LinkedIn Feed using Causal Language Models", submitted to arXiv on 16 October 2025 with 23 authors. Its abstract says:

• The retrieval stage of the feed must narrow hundreds of millions of candidate posts to about 2,000 within a latency budget of a few milliseconds.
• The authors fine tune Meta's LLaMA 3 as a dual encoder that produces embeddings for members and for content, using only text.
• Quantizing numerical features in the prompt helps the embeddings encode that information and improves alignment with the ranking layer.
• Offline metrics and an online A/B test showed substantial gains in member engagement, with the largest improvements among newer members who have few connections.

Read plainly: LinkedIn describes a pipeline in two stages, one that picks a few thousand plausible posts from a huge pool, and one that ranks them for each member. The retrieval stage matches on meaning in text, which is why the paper mentions benefits for people with small networks. LinkedIn also publishes engineering posts at its engineering blog. Some trade press reports a March 2026 post about rebuilding the feed; we could not retrieve it on the day, so we do not rely on it here.

What the paper does not give you: weights for likes against comments, a ranking of post formats, a best hour, or a penalty for links. Those are the things most "algorithm guides" claim to know.

Which popular claims are unverified?

We checked the claims that recur in third party guides against the LinkedIn material above and found no support for them. Treat each as folklore until LinkedIn says otherwise.

This is not a claim that these things are false. It is a statement that nobody outside LinkedIn can prove them, and that guides which print exact numbers are guessing. The honest way to learn what works for your audience is to test it.

How do you test what works for your audience?

Run a simple experiment, labelled as ours:

1. Pick one variable, such as the first line or the posting day.
2. Publish at least 8 to 10 posts per variant over several weeks, so one lucky post does not mislead you.
3. Record impressions, reactions, comments and profile views in a sheet.
4. Compare medians, not the best post.
5. Keep the winner and test the next variable.

Our posts on LinkedIn content strategy for B2B and LinkedIn posting rules that matter cover what to post. This one is about what to measure.

Does the feed algorithm affect your outbound?

Mostly not directly. Invitations and messages are not ranked in the feed. They reach a person because you sent them, and are accepted or ignored on their merits. What decides the result:

• Who you contact. A fit audience accepts more often than a broad one.
• Your profile. People check it after an invitation. A clear headline and recent activity build trust.
• The message. Relevance to the person beats clever phrasing.
• Your account's standing. LinkedIn restricts invitations under its own rules, described in its Help page on types of restrictions for sending invitations, and those rules are separate from ranking.

Posting has an indirect effect. A person who sees your posts before your invitation arrives knows who you are. Our post on pre invite engagement describes that effect.

What should you measure for outbound?

These are the numbers your own sending produces. None depends on a hidden formula.

Be explicit about the denominator. Our post on defining reply rate shows how the same data produces very different numbers, and the acceptance rate post covers the other end of the funnel.

For scale, here are our own numbers. In our own workspace between September 2025 and October 2026, 10,978 of the 27,737 people we messaged on LinkedIn replied (40%), and 351 showcase meetings were booked through the agents' booking link. Separately, we count over 1,500 calls booked in the last 12 months, using on average 7.5 LinkedIn accounts. These are one company's results from one offer and audience, not a benchmark for yours. The Sellinger case study has the method.

A worked example (made up)

A consultancy runs two LinkedIn campaigns for a month. Campaign A sends 400 invitations to operations directors at logistics firms. Campaign B sends 400 to general "managers" in any industry. The numbers below are invented to show the method, not real results.

Nothing in this table depends on how the feed ranks posts. The difference comes from the audience and the message, and the team can act on it the next day by shifting volume to Campaign A and reading the replies in B to see what went wrong.

A weekly routine

1. Monday: review last week's acceptance, reply and meeting numbers per campaign.
2. Tuesday: change one thing, such as the first line or the segment, and write it down.
3. During the week: read twenty replies yourself, both good and bad.
4. Friday: check account warnings and daily volumes.
5. Monthly: look at the posting experiment, if you run one.

The point is a feedback loop on things you control. Chasing an algorithm you cannot see is the least efficient use of a week.

What this means for Sellinger users

The LinkedIn agent sends invitations and messages from your connected account within per account limits: a default of 20 invitations a day and 150 a week, a quiet window with no sending, and a schedule you set. It does not try to influence feed ranking. No tool can guarantee that LinkedIn will never restrict an account. The numbers above are what you should watch inside the app.