AI meeting feedback that argues back, not another summary.

Most AI meeting feedback stops at the summary: a tidy recap, a list of action items, maybe a sentiment score. Useful, but it never asks the one question that matters after a hard call, which is whether the decision is any good. A meeting critique is the missing half. Instead of compressing what was said, it pressure-tests what was decided, the way a sharp colleague would if they were not worried about the mood in the room. This post covers what a critique is, why it beats a summary, and exactly how Recard produces one.

What a meeting critique actually is

A meeting critique is a structured, adversarial review of a decision that was captured in a conversation. It assumes the summary already exists and goes one step further: it names the decision on the table, builds the strongest possible case for it, and then goes looking for where that case falls apart.

That structure is the point. A summary answers "what happened." Action items answer "who does what." A sentiment score answers "how did it feel." None of those tell you whether the plan is sound. A critique answers a harder question: "should we, and where does this break?" It is the difference between a transcript that flatters the room and a second opinion that is willing to disagree with it.

Why devil's-advocate review beats a summary

Summaries are recall. They are genuinely helpful for remembering commitments, but they are structurally incapable of pushing back, because their whole job is to reflect the meeting faithfully. And meetings are bad at disagreeing with themselves. Sunk cost, social pressure, the loudest voice, and the simple fact that the person who should challenge the plan is often the one who called the meeting all conspire to leave the weak assumption unexamined.

A critique externalizes the disagreement. Nobody in the room has to volunteer to be the skeptic and spend social capital doing it. The tool takes that role, on every recording, without ego or fatigue.

Inside a Recard Critique, block by block

Here is what the output looks like on a real decision. In this example a team is weighing whether to switch a core vendor. The critique opens with the strongest version of the argument for switching, then splits into the conditions where that call is right and the ones where it quietly goes wrong, and closes with a source it can point to.

Notice the order. The critique does not open by tearing the idea down. It builds the best case for switching first, so the criticism that follows lands on the real decision rather than a strawman of it. Then the two verdicts do the work a summary never can: they name the precise conditions that separate a good call from a bad one, and they flag the risk nobody in the room had claimed.

The seven moves it makes

Every Critique runs the same repeatable sequence, so you get judgment, not vibes:

• Steelman first. It restates your decision in its strongest, most charitable form before it says a critical word, so the critique is aimed at the real idea.
• Holds-when and breaks-when. Instead of vague pros and cons, you get the exact scenarios where the decision is right and the ones where it fails.
• Contradictions. Where the conversation argued against itself, said one thing at 09:00 and the opposite at 41:00, it points to both.
• Blind spots. Harder than contradictions: what the room never raised at all. The risk that was simply never on anyone's radar.
• Decision reversibility. It weighs how expensive the call is to undo, because a cheap, reversible bet deserves less agonizing than a one-way door.
• Premortem. It fast-forwards, assumes the plan failed, and writes the post-mortem in advance, so you can fix the failure while it is still cheap.
• Cite-or-confess. Every factual claim is either linked to a source or clearly labeled unverified. It would rather admit it does not know than bluff.

When to run a critique

A critique earns its keep on the conversations that actually carry weight:

• Pre-decision. Before you lock a roadmap, sign a vendor, extend an offer, or set a price, run the plan through a premortem and find the failure mode that is still cheap to fix.
• Post-mortem. Right after a high-stakes call, ask what you missed while it is fresh, before the story hardens into "that went well."
• Irreversible calls. When a decision is hard to walk back, the reversibility read alone is worth the two minutes.

How Recard handles this

Recard's Critique reasons over the full transcript of that one recording, plus any context notes you add, so it is grounded in what was actually said rather than a generic template. Under the hood it runs on a top-tier reasoning model with live web grounding, and it is honest about the limits of that grounding through cite-or-confess sourcing.

You can turn web research off for any Critique, on any plan, when you want it reasoning only over what was said. In mobile Ultra Privacy, your audio is temporarily uploaded and processed by Google Cloud Vertex AI, the complete transcript and analysis return to your phone, and Recard deletes and verifies deletion of its cloud recording content. Afterwards, supported private features run from the transcript held on your device with web research off. This is zero-retention application processing, not on-device AI, and it uses your normal plan allowance. The capture needs no meeting bot on the invite and no pendant to buy: your phone records the room, and the critique follows. Export the reasoning to Word or Markdown on any device, plus PDF on Android, when you want to share more than the recap.

One honest note on the category: general critical reasoning about the actual decision, steelman through premortem, is the hardest thing in the product to copy, and we are not aware of another notetaker that does it. We would rather show you the output than quote a number we cannot back up.

FAQ