Check what you already have before buying anything. Embedded features avoid moving your data around.
Do now: List the five tools you use most and find the AI feature in each.
Convenience is not consent. Check what the embedded feature does with your content.Embedded features have a real advantage over separate tools: your material is already there, so nothing has to be copied out. They also carry a real risk, because the data-handling question is easy to skip when the feature simply appears.
Where the features usually hide
Spreadsheets: formula explanation, pattern filling, cleaning suggestions and chart recommendations. Word processors: drafting, rewriting, summarising and tone adjustment. Email: reply drafting, summarising long threads, and extracting actions.
Messaging and meeting tools: transcription, summaries and action extraction — often the single highest-value feature most people ignore. Design and presentation tools: layout suggestions, image editing and background removal. Phone keyboards and cameras: dictation, translation and text extraction from photographs.
The advantage is proximity. Your document is already in the word processor, your data is already in the spreadsheet, and nothing has to be copied into a browser tab, which removes an entire category of privacy mistake.
The best feature is often the one already sitting in the software you have open.
Ask the data question anyway
Embedded features feel like part of the software, which makes it easy to forget that content is still being processed somewhere. Find out what happens to it, whether it is retained, and whether your organisation has approved that use — the same questions you would ask of any separate tool.
Business and consumer versions of the same product frequently differ substantially here. Assuming the terms your employer negotiated apply to the copy on your personal laptop is a common and consequential mistake.
Check whether the feature can be turned off for specific documents or folders. For anything sensitive, that switch is worth finding before you need it.
Decide whether you need anything more
Run the embedded feature against your actual task for a fortnight before concluding it is insufficient. Many people conclude too early, having tried it once on the wrong kind of task.
Where an embedded feature genuinely falls short, name the specific gap — it cannot handle the length, the format, the language, or the depth — before shopping. A named gap makes the comparison quick and prevents buying a tool that has the same limitation.
And accept that the embedded version is often good enough for the ninety per cent, with a separate tool for the remainder. That combination usually costs less and moves less data around than replacing everything.
Name the specific gap before you shop. 'Not quite good enough' is not a specification.