Which Copilot mode, and when.
Copilot can answer straight away, think for longer first, or decide for itself which to do. Matching the choice to the task gets better results in the time you have.
About five minutes to read, plus six practice tasks. No technical knowledge needed.
A line runs from Quick response at one end to Think deeper at the other. Auto, the default, sits above the line and picks a point along it for each request.
The three modes
One assistant working at three different paces.
-
Quick response
Answers immediately. Best when you already know what a good result looks like.
- Rewriting an email so it's shorter or clearer
- Titles for a document or an event
- A fast list of ideas
- A straightforward factual question
-
Auto the default
Decides how long to think for each request, balancing quality against speed.
- Drafting emails, documents and messages
- Summarising meetings, reports or long threads
- Pulling together information from several places
- Anything where you're not sure which mode fits
-
Think deeper
Takes longer before answering, and usually reasons more carefully.
- Comparing options and recommending one
- Planning work with several steps or constraints
- Analysing a long or complex document
- Anything where the reasoning matters as much as the answer
Choose in two questions
Before you send a prompt, ask these in order.
-
Is it short and simple, and can you check it at a glance?
Yes → Quick response
No → question two
-
Does it involve weighing up options, planning, or working through something long?
Yes → Think deeper
No, or you're not sure → Auto, which will decide for you
Pick a named model instead when you're repeating a task you've already tested and want the same model each time, or when you want to compare two models on one prompt.
Practise
Six everyday tasks. Every mode would produce something usable, so the feedback says which fits best and why. Change your answer as often as you like.
Answered: none yet
Try it on something you know
The fastest way to learn the difference is to run one task twice, on a document you've already read.
- Choose Quick response, attach the document, and ask for a five-bullet summary plus the one point you should check against the original.
- Start a new chat, choose Think deeper, and send the same thing.
- Compare where the two answers differ, not how long they are. Because you know the document, you can judge which one caught what mattered.
If the quick summary covered everything, that kind of task suits Quick response. If the slower one found something you'd have wanted, use it for similar work in future.
Check every response
Every mode gives you a starting point, not a finished thing. Match the depth of your check to what the output is for.
-
A reworded sentence
Read it once and make sure it still says what you meant.
-
A summary you'll send on
Compare it against the original and confirm nothing important dropped out.
-
Analysis behind a decision
Check the facts, the figures and the sources before you rely on any of it.
About this guide
I wrote and built this as a portfolio piece. It's hand-coded rather than authored in a tool, and the things that make it work for everyone aren't visible unless you need them: every diagram carries a written description for screen reader users, all animation is disabled under reduced-motion settings, every control clears the 44-pixel minimum touch target, and the practice activity announces its results to assistive technology.
Mode names and menu wording change as Microsoft updates Copilot. The two-question rule doesn't.