AI summarisation: extractive vs abstractive, and when each one wins

"Summarise this for me" hides two very different techniques. Knowing which one your tool uses changes how much you can trust the output.

Extractive summarisation

An extractive summariser picks real sentences from the source and stitches them together. The summary is guaranteed to be faithful — every word came from the original — but it can read a little choppy.

Abstractive summarisation

An abstractive model (the kind powered by today's large language models) generates new sentences in its own voice. The output is smoother and shorter, but the model can hallucinate facts that weren't in the source.

When to use each

• Legal, medical, financial text → extractive. Faithfulness matters more than style.
• Meeting notes, casual articles, brainstorms → abstractive. You'll get a much more readable summary.
• Mission‑critical reporting → run both and compare. Trust nothing without verification.

Treat every AI summary as a first draft. Always spot‑check at least three claims before forwarding it.