Generative Recommendation (GenRec): The Future of Recommendation Systems

How Transformers are changing the way AI recommends.

Rohan Mistry

6 min read

·

Oct 23, 2026

--

1

Listen

Share

For a decade, every recommender system worked the same way.

Retrieve candidates. Rank them. Re-rank them. Repeat, with a different team owning each stage.

Now a different idea is spreading fast: treat recommendation like language modeling, and let a single Transformer generate what a user will want next.

Recommendation is becoming a next-token prediction problem. That’s the whole shift.

This article explains what generative recommendation (often shortened to GenRec) is, how it works, the production results behind the hype, and where it still falls short.

Non-Member’s Link!

Press enter or click to view image in full size

First, What Does “GenRec” Mean?

The name is used two ways, so it helps to separate them:

• GenRec the paradigm: Generative recommendation in general, where a model generates items instead of scoring them.
• GenRec the papers: Several research papers share the name, including a 2023 Rutgers paper that fine-tunes an LLM to generate item titles, and an April 2026 JD.com paper on a production framework.