A cast-iron anvil in shadow beside a worn leather work glove on a dark forge bench.

What belongs on the company page and what belongs on a person

LinkedIn's feed is ranked by a large language model reading meaning, which changes what a brand page can do and what only a named person can.

LinkedIn now ranks its feed with a single large foundation model rather than a stack of separate task-specific models. The engineering work was published in January 2025 and the production rollout was confirmed in March 2026, described as generative recommenders paired with large language models.

The consequence for your business is not subtle. The system reads what a post means, so metadata gaming stopped working and specificity started paying. That changes which content belongs on the company page and which content only works from a person.

The mechanical change

The old approach matched on signals around the content: engagement history, connection graph, keywords, tags, the topics you had followed. The new approach reads the content itself and works out topical relationships that a keyword system missed. LinkedIn's own example was that someone interested in electrical engineering who engages heavily with small modular reactors will now be matched correctly, because the model understands the relationship between the two.

Two things follow immediately.

Hashtags are close to worthless as a distribution lever. They were a metadata shortcut for a system that could not read. It can read.

Vague content has no match surface. A post saying the team is proud to serve clients across the region to the highest standard contains no topic. There is nothing in it to relate to anyone's interests, so it goes to the people who already follow you and stops.

What the company page is genuinely for

The page has not become useless. It has become a different instrument.

It is the record. Announcements, milestones, product releases, contract wins, safety and compliance updates. Things that need to exist in an official voice at a permanent address.

It is the credibility check. A buyer who meets your sales lead will open the company page before the second meeting. What they need there is evidence of a real, active, competent business, not reach.

It is the foundation for paid. Your ads run from it, and a dead page undermines the ads pointing at it.

It is a citation source. AI answer engines and search both read company pages when assembling an answer about a business. That is a reason to keep the page current and factual even when its organic reach is modest.

None of those jobs require the page to be a growth engine. Stop measuring it as one.

What only a person can do

Match to a topic interest at scale, because the specific first-hand content that a model can classify precisely almost always comes from an individual.

Consider two posts about the same event. The page version says the team attended an industry conference and enjoyed valuable conversations about the future of the sector. The person version says that in a session on remedial waterproofing, three separate builders raised the same detail failure at balcony thresholds, that it matches what their crew has been finding on 1980s stock in the inner west, and here is the detail that has actually held up.

The second post has a topic. It has terms a model can relate to other terms. It has a point of view, which is a thing only a named human can credibly hold.

Where employee advocacy programmes fail

The standard programme sends the same copy to thirty employees and asks them to post it. Under semantic ranking this is worse than doing nothing, because the model reads the sameness. Thirty near-identical posts describe one piece of content distributed thirty times, and the individual accounts get no benefit from carrying it.

The version that works is harder and slower.

Pick five to eight people who genuinely know something, not the whole company. Give them a subject and the underlying facts rather than a script. Have someone edit for clarity and legal risk without flattening their voice. Accept a lower posting frequency in exchange for posts that are actually theirs. And measure at the person level, not as a programme total.

Five people posting something real once a fortnight will beat thirty people posting the same paragraph.

The split, written plainly

Company page: the record, the credibility surface, the paid foundation, the factual reference. Consistent, accurate, unexciting, and updated.

People: the meaning. Specific, first-hand, one subject per post, named accountability, a genuine opinion where they have earned one.

Trying to make the page do the second job is where most mid-market LinkedIn budgets go to die.

What to do in the next fortnight

Take your last twenty company page posts and mark each one as record or meaning. If more than a handful are meaning posts, they are in the wrong place and should be republished from the person who actually knows the subject.

Then find the three people in your business who know something a buyer would pay to hear, and ask them what they have argued about with a client this month. That conversation is your content plan.

The model reads for meaning now, so put the meaning where a human can sign it.

Written by David Eid. Published .