AI, audio, and agents
AI summary
Also called Article summarizer, Automatic summarization
An AI summary is a shortened account of a source text produced by a language model. Extractive summaries reuse sentences from the original; abstractive summaries — what modern models produce — rewrite the material in new words, which is more readable and also where factual drift can enter.
Extractive versus abstractive
Extractive summarization ranks the original sentences and returns the top ones. It cannot invent anything, but it reads like an outline and misses points spread across several paragraphs. Abstractive summarization rewrites the content in the model's own words, which reads better and can compress an argument, at the cost of introducing claims the source did not make.
Grounding is the safeguard
A summary is only as trustworthy as its link back to the text. Summarizing from a retrieved document, keeping that document available beside the summary, and making every claim checkable against it are what separate a useful summary from a plausible one. A summary should be an entry point to the article, not a replacement for it.
In smry
In smry the article stays the grounding context: summaries, chat, and highlights all sit beside the retrieved text, so any claim can be checked against the source in the same view.
Common questions
Are AI summaries accurate?
They are usually directionally right and occasionally wrong in specifics — numbers, attributions, and hedged claims are the common failure points. Keeping the source alongside the summary is the practical check.
What is the difference between a summary and an overview?
Nothing technical. Both are condensed accounts of a source; the term varies by product. What matters is whether the output is grounded in a retrieved document.
Can a summary replace reading the article?
For triage, often. For anything that depends on precise wording, evidence, or nuance, no — a summary flattens exactly the details that carry the argument.