The Truth / scorecard

Copilot Chat: the best feature nobody knows they have

"Cross-app synthesis that borders on organizational telepathy — hidden behind naming chaos and a grounding toggle most users never notice."

8/10
Best for: Anyone whose job is reassembling context scattered across email, chats, and meetings — managers, PMs, account owners Skip if: Your work lives in one app, or your tenant's file hygiene is so chaotic that retrieval would be noise updated June 2026

What's genuinely good

  • Cross-source synthesis is the killer skill: one question pulls from your mail, Teams messages, meetings, and files simultaneously — nothing else in the suite does this
  • Pre-meeting prep ('catch me up on everything related to my 2pm') is the single highest-ROI prompt in M365
  • It's your organizational memory: 'what did Sandra promise about the renewal, and where?' finds the email, the chat, and the meeting
  • A free tier exists — web-grounded chat costs nothing, and tenant grounding is available metered via Copilot Credits without a full license
  • Pinning files and people as sources turns a vague answer into a precise one in one step

What sucks

  • The work/web grounding toggle is the most consequential UI element most users have never consciously clicked — wrong mode, useless answer, no warning
  • Retrieval is a lottery without pinned sources: it searches your tenant, finds *something*, and presents it with the same confidence whether it found the right thing or a 2023 draft
  • Most licensed users don't know it exists, or confuse it with consumer Copilot — Microsoft's naming (Copilot, Copilot Chat, M365 Copilot, BizChat) actively sabotages adoption
  • Answers inherit your tenant's permission and indexing quirks — oversharing surfaces things, undersharing hides them, and you can't tell which happened
  • Long, multi-step research holds together worse than focused single questions — it's a synthesizer, not yet a researcher

The honest assessment

Copilot Chat — the work-grounded one, the thing insiders still call BizChat — is the most underrated product in the entire stack, and the reasons are entirely self-inflicted by Microsoft’s naming department.

What it actually does is something no embedded Copilot can: answer questions that span your mail, your Teams threads, your meetings, and your files in a single shot. “What’s the latest on the Henderson deal?” is not an Outlook question or a Teams question — the answer is smeared across four apps, and Copilot Chat is the only place that can reassemble it. Pre-meeting prep is the canonical version: ask it to catch you up before a call and it pulls the email thread, the shared deck, and what was said in last week’s meeting. When it lands, it feels like organizational telepathy.

The catch is the word when. Without pinned sources, retrieval is a lottery: it searches your tenant, grabs what scores highest, and synthesizes it with total confidence — whether that’s the final contract or an abandoned draft from eighteen months ago. The answer reads identically either way. And before any of that, there’s the grounding toggle: work mode searches your tenant, web mode searches the internet, and a user in the wrong mode gets a confidently generic answer and concludes the product is broken. That single toggle is responsible for more “Copilot is useless” verdicts than any model limitation.

The workarounds that change the score

Three habits push Copilot Chat from an 8 to a personal 9 or 10:

  1. Check the toggle before you judge the answer. Work questions in work mode, world questions in web mode. If an answer feels weirdly generic, you’re almost certainly in the wrong mode — switch and re-ask before deciding anything.
  2. Pin your sources, end the lottery. Reference the specific file, person, or meeting in the prompt (the / and @ pickers exist for this). “Summarize the risks in this document” is deterministic; “summarize the risks in the project” is a dice roll.
  3. Make prep a ritual, not an experiment. “Catch me up on everything related to [meeting], including unanswered emails and open action items” before every significant call. This one habit is worth more than everything else in the suite combined for meeting-heavy roles.

What Microsoft won’t tell you

  • The naming collision is costing real adoption. Users who tried consumer Copilot assume they’ve “tried Copilot.” They haven’t tried the work-grounded version, which is a different product wearing the same name.
  • The free tier changes the rollout math: web-grounded chat is free for everyone, and tenant grounding can be metered via Copilot Credits. You can let unlicensed users taste work grounding pay-as-you-go before committing to seats — almost nobody knows this.
  • Every answer is a permissions audit. Copilot Chat surfaces whatever the asking user can technically access, which in overshared tenants is far more than anyone realized. Run the oversharing review before the rollout, not after the incident.

Bottom line

This is the second-best product in the suite hiding behind the worst naming in the suite. Users who learn the toggle and pin their sources get a cross-app memory no competitor matches. Users who stumble in cold get a lottery ticket and walk away. The gap between those experiences is about ten minutes of training Microsoft never tells you to run.

← All scorecards