LinkedIn revamps feed algorithm with LLM-powered ranking and retrieval

LinkedIn has updated its feed algorithm to use large language models for both ranking and retrieval, according to Search Engine Land. The change moves the professional network toward the transformer-based architecture that other major platforms have already adopted for content distribution.

The shift, which LinkedIn disclosed in technical materials cited by Search Engine Land, consolidates retrieval and ranking into a single LLM-based layer. Retrieval selects the candidate posts a member might see; ranking orders them. Combining the two in one model family differs from a typical multi-stage pipeline, which uses separate models for candidate generation, scoring, and re-ranking.

What does the LLM layer change for content?

The shift makes textual alignment a more direct input to ranking. Posts whose language matches a member's profile signals—job title, skills, industry, recent activity—stand to benefit from the new stack. Carousels and documents with descriptive captions may gain relative advantage over image-only frames. Video will continue to be prioritized, but on-screen text and captions now carry more weight in retrieval and ranking.

For B2B marketers, who have built LinkedIn playbooks around engagement velocity and early-post momentum, the change suggests a longer optimization horizon. Posts that take time to gain traction may surface more reliably if the LLM layer picks up on topical relevance rather than early engagement signals.

Why does the architectural change matter?

LLM-powered ranking and retrieval align the feed's selection and ordering around the same representation of language, profile data, and engagement history. This matters because the two functions typically optimize for different goals: retrieval favors recall, surfacing the broadest set of relevant posts; ranking favors predicted engagement, picking the ones most likely to hold attention. Sharing a model between them changes how those tradeoffs play out in practice.

For LinkedIn, the feed is the surface where most member time is spent. Any change to ranking affects organic reach, conversation starter visibility, and the cost dynamics of sponsored content—all of which sit atop the platform's talent and marketing revenue.

What should practitioners do now?

Three actions follow from the rollout:

• Audit post copy for alignment with target audience job titles, skills, and industry vocabulary.
• Lead documents and carousels with descriptive text rather than image-only frames.
• Treat post copy as a primary ranking input, not a caption afterthought.

What's next for the feed?

LinkedIn has not disclosed a full deployment timeline. The change signals continued integration of generative AI into the platform's core surfaces, with ad products, conversation starters, and recruiter search likely candidates for the next phase. For marketers and creators, the practical takeaway is straightforward: copy quality and topical precision now carry more algorithmic weight than they did under the previous pipeline, and the metric worth tracking is no longer first-hour engagement but sustained topical match.