Can Copilot in Teams summarize old feature requests?
No, not by default, and not without work. Microsoft's own support documentation for Copilot in Teams states that it "references information from the message thread in which it is opened, with a 30-day history as the default time frame unless you specify otherwise." That default window applies to whichever single chat, channel, or thread you're currently in when you ask. A feature request a customer raised four months ago, in a channel you're not currently sitting in, is outside both boundaries at once.
You can push past the 30-day default inside a single thread by naming a date range in your prompt, something like "summarize requests about the mobile app from the last 90 days." Microsoft's own prompting guide for channel summaries shows the pattern as "What's new in [channel] of [team]?", which only works if you already know which channel to ask about. Neither the support docs nor the prompting guide describe a way to ask Copilot to summarize across several channels in one pass. If your customer's feature requests are spread across a general channel, a Teams Connect channel shared with your account team, and a support escalation chat, you're running that same prompt three times and reading three separate summaries by hand.
Why the 30-day default exists, and where it bites
Copilot's scoping isn't a bug someone forgot to fix. It's a real-time chat assistant reading a live thread, and a live thread's useful context is usually what happened recently, not what happened at the start of the year. The same support page that states the 30-day default also lists what Copilot can't read at all: "Copilot can't summarize images, loop components, or files shared in the chat thread." A feature request pasted as a screenshot of a mockup, or attached as a Loop doc with the actual spec, is invisible to a summary request regardless of how you phrase the date range.
Put those two limits together and the failure mode is specific. A request raised in March, described partly in a Loop component, in a channel your prompt didn't name, will not show up when someone asks Copilot in September "what feature requests has this account made." Not because Copilot is broken. Because none of the three conditions it needs, recent, in-thread, and text, were true.
A support engineer runs into the gap
Soren Bakke leads customer support at Amberline Robotics, a company that sells warehouse picking arms and runs its account relationships almost entirely through Teams, since most of its customers are manufacturers already standardized on Microsoft. In late September, one of Amberline's largest accounts asked when a specific safety-interlock override would ship. Soren opened Copilot in the account's main Teams channel and typed: "Summarize any feature requests this account has made about the interlock system."
Copilot came back with one item from three weeks earlier, a minor request about the override's confirmation dialog. It didn't surface the account's original ask, filed in a Teams Connect channel back in May, or a follow-up mentioned in passing during a support escalation chat in July. Both were in scope, technically, they were real messages that existed. Neither was in the thread Copilot was reading, and the May message was five months outside the 30-day window besides.
Soren's next move was the honest one. He opened each channel separately, ran a summary in each, and manually stitched the results into a reply. It took about twenty minutes and produced a paragraph that was accurate but assembled by hand, the same reconstruction work Copilot was supposed to save.
Getting more out of Copilot without leaving it
Short of switching tools, a few habits close part of the gap:
- Ask in the channel where the request lives, not the channel you happen to have open. Copilot's summary is scoped to the thread you're in, so cross-channel asks always come back partial.
- Name a date range explicitly rather than relying on the default. "Summarize requests since January" reaches further than the unstated 30-day window, though reply quality on old messages still degrades, a pattern also documented in why Teams search misses old messages.
- Keep the actual ask in plain text, not a screenshot or a Loop component, if you want it summarizable later. Anything Copilot can't read today, it still can't read in six months.
- Pin or bookmark the message the moment a customer raises something you'll need to reference, since Teams search itself becomes unreliable past about a month, and Copilot inherits that same recency bias rather than fixing it.
These habits help a single channel. They don't solve the cross-channel problem, because Copilot has no index that spans channels in the first place, only a live read of whichever thread you asked it from.
When correlating across channels becomes the job
Once feedback about one account, or one feature, is scattered across more than two or three Teams channels, plus whatever the same customer said in a support ticket or an email, manually running Copilot in each location and merging the answers stops scaling. It's not that Copilot got worse. It's that the job changed from "summarize this thread" to "correlate everything said anywhere," which is a different piece of software.
That's the point where we build Modem. Modem's Microsoft Teams integration subscribes to the channels you connect, including Teams Connect channels shared with customers, imports the last 30 days of history on connection so it isn't starting from an empty feed, and then reads new messages, replies, edits, and reactions as they happen from then on. Every message gets classified and clustered into a topic regardless of which channel it landed in or how many months ago it was said, so Amberline's interlock request would already sit in one topic carrying the May message, the July mention, and the September follow-up together, with Soren's account attached to all three. There's no lookback window to run past and no second channel to remember to check, because nothing here depends on which thread happens to be open when someone asks. Modem is what we build, so read the pitch with that in mind. What matters more is the mechanism. A persistent index across channels solves a problem that a better prompt to a same-thread assistant cannot.
The direct answer again
Copilot in Teams can summarize feature requests, but only the ones in the thread you're currently reading, and by default only from the last 30 days. Ask it about a different channel or an older message and you get a partial answer or nothing, not because the request wasn't real, but because it was never in scope. For a single channel with recent activity, that's often fine. For a request that moved across channels or aged past a month, the honest fix is checking each channel yourself, or reading from something built to hold all of them at once.
