Outlook search now ranks seven emails, with a small language model doing the sorting

Microsoft has revamped Outlook search, showing seven ranked emails instead of three and using a small language model to order them. The update also brings offline search fallback to Classic Outlook, where a poor connection need not leave the inbox searching in vain.

Microsoft AI Watch analysis

What happened

The new ranking is available in new Outlook and Classic Outlook, according to a ghacks report citing a Microsoft blog post. Microsoft ranks messages using signals including recency, sender and interaction, and says the model was built specifically for Outlook using distillation, a technique for training a smaller model under the supervision of a larger one.

The change also treats common words such as “the”, “and” and “is” differently, aiming to stop them dragging irrelevant messages into results. In Classic Outlook, searches can fall back to locally indexed mail when connectivity is poor. Microsoft says recent improvements have reduced latency for the slowest 5 per cent of Outlook searches. The same offline fallback is planned for new Outlook, but no release date has been given.

Why it matters

This is a practical example of a small language model being put to a specific job: ranking email without requiring Outlook to run a large model for each search. For users, the visible changes are straightforward to check: seven ranked messages rather than three, less noise from common words, and locally stored results in Classic Outlook when the connection falters.

Our read

Search is one of those features people notice most when it fails, usually while looking for the email they absolutely did not delete. More ranked results and better handling of ordinary words sound useful; the offline fallback is the more concrete win for Classic Outlook users with unreliable connectivity. Try a search you know well and see whether the right message has moved up the list.

What to watch

• When the offline fallback reaches new Outlook.
• Whether common-word searches produce more relevant results in practice.
• Whether Microsoft shares further performance or user-feedback findings.

Discussion spark: For email search, would you rather see more ranked results at the top, or fewer results with a stronger guarantee that the first one is right?

not affiliated with or endorsed by Microsoft