LinkedIn boosts its messaging with smart replies, pre-written, AI based interactions
Context & Ripple Effects
Smart replies look small in isolation, but they are the earliest entry in a decade-long arc the coverage traces clearly: LinkedIn first bought predictive insights with Refresh.io in 2015, then put pre-written AI into messaging in 2017, and by 2023 was publishing AI-generated conversation starters and offering GPT-4 writing suggestions for profiles and recruiter job posts.
The through-line is that each generation of features moves drafting work from the member to the platform — and by 2025 LinkedIn had enough of its own data to launch free Jobs Match and recruitment-agent tools built on its own AI rather than OpenAI's, with premium-only AI people search following. This 2017 messaging update is where that substitution of machine-drafted text for human-typed text began.
First-order effects
- Members replying to messages get one-tap suggested responses, so routine networking exchanges — acceptances, follow-ups, thank-yous — get answered faster and with less typing.
Second-order effects
- Every accepted or edited suggestion becomes labeled training signal inside LinkedIn's own messaging corpus, which is precisely the proprietary data the company later cites when building its own AI stack instead of licensing OpenAI models.
Third-order effects
- If the pattern holds — drafted conversation starters, drafted profiles, drafted replies — professional networking drifts toward what The Verge called semiautomated social networks, where the platform composes and the human merely approves, raising the question of how much signal remains in interactions both sides machine-wrote.
The trend: LinkedIn has spent eight years moving text generation from members to the platform, one surface at a time, from smart replies to conversation starters to its own in-house AI tools.