Copilot Discover: turn an idea into checked research

To make a Copilot Discover topic useful, turn it into one question and check the answer against original sources. The Microsoft guides cited here do not establish a current Discover feed location or its controls, so a verified feed-opening procedure is still missing. Once you have a topic, the research steps below help you decide whether it deserves more work. fenbs can keep your checked findings and the next question outside the chat.

Where should you start when looking for Discover?

First identify which Microsoft product and account you are using. Microsoft Copilot, Copilot Studio and GitHub Copilot have different documented capabilities; a setting in one should not become an instruction for another. The Microsoft AI tools guide helps with that product choice without assuming a Discover feature is present.

Microsoft’s Pages guide↗ (opens in a new tab) documents editable content and its account requirements. It does not document the Discover feed requested here. If your question is which exact control opens Discover, that cited guide cannot settle it; avoid treating a screenshot from a different account as proof of your own access.

If a Discover feed is visible in your account, use an interesting topic as a lead to investigate. Write its wording down before expanding it. If you cannot locate the feed, you can still begin with your own question; the checking method does not depend on a particular recommendation screen.

How do you turn a suggested topic into a research task?

1. Choose one topic connected to a real decision or piece of work. Leave other interesting ideas for later.
2. Write one question you could answer with evidence. Include the audience, location or situation when it changes the answer.
3. Say what a useful result looks like: a short explanation, a list of unresolved questions, or a recommendation you can review.
4. Ask for original sources and for uncertainty to be kept separate from confirmed information.
5. Open the cited material yourself, check the relevant passage, and decide what still needs investigation.

For example, imagine you run a small design studio and notice a topic about accessible checkout forms. “Tell me about accessibility” is too broad. “Which public guidance should we check before reducing the number of fields in our checkout?” gives the research a purpose without pretending that an AI answer is an approval.

Keep the first request small enough that you can inspect every important source. You can borrow the structure in these Copilot prompt examples, then replace the subject and acceptance check with your own. A useful prompt specifies what evidence you need rather than asking for a persuasive answer.

What should you check before trusting the answer?

Read the source for each claim that would affect your next action. Check who published it, what it actually says, and whether the passage applies to your situation. A link to an organization’s home page is not enough evidence for a detailed claim about its current instructions.

• Scope: does the source discuss the same product, account type, audience and task?
• Timing: is the information current enough for the choice you are making?
• Support: does the original passage support the claim, or only a nearby idea?
• Uncertainty: what remains unanswered after reading the source?

Write “needs checking” when evidence is incomplete. In the studio example, a general explanation of readable forms would not establish that your particular checkout meets an accessibility requirement. Keep the proposed change and the remaining review separate. That gives the next person something concrete to inspect.

Save the source, the conclusion you checked and the open question together. The work record described in how fenbs works is useful here because a later assistant needs your checked conclusion and its limits, rather than a confident summary with its evidence removed.

Why might an interesting topic lead to an unhelpful result?

A topic can be interesting without being useful for your current task. Before asking for more research, finish this sentence: “I need this answer so I can…” If you cannot name the next action, save the idea for later instead of expanding it into a large research job.

Another problem is a question that combines several decisions. Comparing form layouts, choosing a payment provider and approving a launch require different evidence. Split them into separate questions and finish the one that blocks the next step. This makes disagreements easier to trace to a specific claim.

If you cannot find the conversation you used, the guide to finding Microsoft Copilot history covers that separate problem. Microsoft’s conversation-history documentation↗ (opens in a new tab) also distinguishes sharing a conversation from exporting it. Neither a shared link nor a transcript replaces a checked research conclusion.

What should you keep when you stop researching?

Keep a short record with the question, the original source links, the answer you verified, and the next check. Include any reason you rejected a source or abandoned a proposed approach. Someone returning tomorrow should be able to see why the remaining question matters without reading every message.

If the work turns into a learning exercise, use the separate Study and Learn workflow for practice and unanswered learning questions. Here the goal remains a checked research task: what did you find, how do you know, and what action follows?

How to keep your research with fenbs

Keep the useful work so you can carry on from exactly where you left off, on any device and in another AI. Start with one question you have already checked instead of copying a whole conversation.

1. Create a free account.
2. Add a task named after your research question. Write the answer you checked and add the original source links.
3. Write what still needs checking and the first small step to take next.
4. If you use Copilot Studio, connect Copilot Studio to fenbs using an account allowed to add tools. Microsoft documents custom servers in Studio; these instructions do not establish a connection in the ordinary Copilot app.
5. Before the next session, ask your connected AI to read the saved work. When you stop, save the new findings and next step.

The Studio boundary comes from Microsoft’s custom-server guide↗ (opens in a new tab). fenbs keeps the research record; it does not provide a Discover feed or check your sources for you. For more practical workflows, browse the AI at work guides.