The Feedback Your Tool Never Sees
Most teams who run async feedback well will recognise the following two weeks.
First, the async loop working exactly as designed. A client records a session on your staging site, walks the checkout flow, narrates what they are thinking, flags a misaligned button and a confusing error message. It arrives as a transcript tied to every click and becomes three tasks before lunch.
Then, on a status call that was supposed to run fifteen minutes, the same client mentions they are no longer sure the checkout flow is the right thing to be polishing, because their sales team keeps saying the pricing page is where people drop off.
The pricing-page remark was never going to show up in a feedback session. Nobody opens a staging site and records a voice note about their own strategic doubt. It surfaced for one reason: two people were talking, one of them said something half-formed, and the other asked what they meant.
This article is part of our hub on Mastering User Acceptance Testing (UAT) for Modern Web Projects. Our async vs. sync UAT breakdown makes the case that async collection is faster, and it is. This piece is the counterweight. The meetings you still hold carry information your async tooling structurally cannot capture, and most teams let that information evaporate.
---
The Two Channels Are Good at Different Things
The instinct is to pick one channel and standardise on it. Resist it. Async and synchronous feedback fail in opposite directions, and covering one gap opens the other.
Async capture is good at anything the client can point at. A broken layout. A wrong label. A form that rejects a valid phone number. The client is looking at the thing while describing it, so the description arrives anchored to a page, a state, and an interaction. Precision, for free.
Async capture is bad at anything the client cannot point at. Doubt about direction. A budget conversation that has not happened yet. The reorganisation that means the product owner you have been working with is about to change. A reviewer sitting alone with a staging URL has no prompt to raise any of it, and a feedback widget gives them nowhere to put it if they wanted to.
Meetings are good at exactly that second category. A conversation has follow-up questions in it. "You said you're not sure about the checkout flow. What's driving that?" is a question no form asks. Half-formed concerns become specific when somebody probes them.
Meetings are bad at precision, and worse at survival. Nobody on a call can tell you which of the forty-seven pages they meant. And the record of what was said is whatever one person managed to type while also talking.
Neither channel covers the other's blind spot. Running only async means you find out about the pricing page three sprints late. Running only meetings means you are back to waiting days for a window that four stakeholders can all make, and losing the 30 to 40 percent of verbal feedback that our async vs. sync breakdown estimates live note-taking misses. The hybrid version is not a compromise between the two. It is using each one where it is strong.
---
Why Meeting Notes Lose the Thing That Mattered
The two channels are recorded to completely different standards. Your async feedback becomes a permanent, searchable, time-aligned record that anyone on the project can go back to a year later. Your meetings become four bullet points in somebody's notes app.
The loss is not evenly distributed, either. What survives a meeting is the part that sounded like an action item. What evaporates is the qualifying clause, the hesitation, the aside — which is precisely where the strategic signal lives. "We should probably look at the pricing page at some point" gets written down, if at all, as "review pricing page." The word *probably* carried the entire meaning, and it is gone.
Whatever the notes miss, memory finishes off. Without rehearsal, people forget roughly half of new information within an hour and up to 70% within a day, which we covered in memory decay and bug reporting. And a note-taker is not writing down what was said. They are writing down what they understood, filtered through what they thought mattered at the time. A client's offhand doubt rarely clears that bar in the moment.
You cannot fix this by taking better notes. Note-taking competes with listening for the same attention, and asking the person who owns the client relationship to also be a stenographer makes both jobs worse. Stop asking a human to be the recording device.
---
Closing the Loop: Make the Meeting Leave a Record Too
The hybrid loop is simple to state. Every channel a client speaks through should end up as text you can search, quote, and hand to whoever (or whatever) is doing the work.
Async, for anything they can point at. Keep the givefeedback.dev widget on staging. Reviewers narrate while they click, and you get a transcript time-aligned with every interaction. Your coding agent can read that transcript directly over the Model Context Protocol (MCP), with your codebase already in context. The Ask, Categorize, Act, Follow-up (ACAF) loop describes how to run this as a repeatable system rather than an inbox.
Meetings, for direction, disagreement, and anything ambiguous. Keep these short and keep them for the things async cannot hold: priority changes, budget, scope disagreements, the check-in where you ask what is worrying them. And record them, so the qualifying clauses survive.
For the recording half, you want something that produces a transcript without turning the call into a production. Memory Notch is the shape to look for, with one caveat worth stating before you get attached to it: at the time of writing, downloads are not yet available. Treat it as a spec for what to look for rather than something to install this afternoon. It is a Mac app that records from the notch and captures your microphone and system audio together, transcribing live with numbered speaker labels.
Each session lands in a folder under ~/memory holding the audio, a transcript, and a Markdown note. Recording and transcription run on-device, which is worth knowing when the conversation includes a client's revenue numbers. The licence, once you can buy one, is a one-time US$49 rather than a subscription.
Be clear-eyed about what a tool like this does not do. Memory Notch writes no automatic summaries. Its library search covers note titles and metadata, not every spoken word. Speaker labels are numbered rather than named, and overlapping voices are hard to separate. There is no API and no MCP server either. What you get is a plain Markdown file on your own disk, which, as it happens, is all an AI tool with local file access needs.
Then converge the two. This is the step teams skip. At the end of a meeting, anything that turned into a concrete request should go where your async feedback goes, as a task with the transcript quoted in it. Anything that stayed strategic should go wherever your project already keeps decisions, with a date attached. We do not provide that; a dated section in the project README or a pinned doc is enough. Six weeks later, when somebody asks why the pricing page jumped the queue, the answer is a quotation and not a recollection.
---
Recording Clients: Ask, and Mean It
None of this works if it damages the relationship, so handle it plainly.
Ask at the top of the call, out loud, every time. "Mind if I record this so I don't have to take notes?" Consent buried in a contract is not consent in the room. If anyone hesitates, do not record. Take notes like it is 2015 and move on without making it strange.
Say what happens to the recording. Where it is stored, who can see it, how long you keep it. "It stays on my laptop and I keep it until the project closes" is a complete answer, and a genuinely better one than most cloud note-takers can offer.
Know the law, then ignore how lenient it is. Consent law varies by country and by state: some places require only one party to consent, others require everyone. Asking out loud satisfies the strictest of those rules, so the rule above does not change when you cross a border into a more permissive one. If you record across borders routinely, get an actual answer from an actual lawyer.
Decide what gets recorded. Commercial negotiations and anything involving a third party's confidential information are often better left off the record. The point of recording is to preserve product signal, not to build a surveillance archive of your client relationship.
---
When to Break the Pattern
Some situations do not want the hybrid treatment.
A conversation that has already gone bad. If a project is sideways and the client is angry, recording raises the temperature and makes people guard their words. Take the call, fix the relationship, write your own notes afterwards.
Regulated or highly sensitive clients. Some organisations cannot have their staff recorded by a vendor at all. Do not negotiate this one. A structured recap email sent within the hour is the fallback, and it is a decent one.
Anything that could have been a feedback session. If a meeting exists only so somebody can walk you through a list of visual issues, that meeting is async work wearing a calendar invite. Send a staging link instead. The reduce UAT cycle time piece covers how much calendar time this one substitution gives back.
---
The Bottom Line
The argument was never async versus meetings. It is that one of your two feedback channels produces a durable written record and the other one produces four bullet points, and the second channel is the one carrying the information you would most regret losing.
Fix the asymmetry rather than the channel mix. Let clients point at things asynchronously and capture that with full fidelity. Hold fewer, shorter meetings for the things they cannot point at, and record those so the hesitations and qualifying clauses survive the week. Converge both into the same place, so that the answer to "why are we building this?" is something you can quote.
Run the async half properly and the meetings get shorter on their own, because nobody needs to spend twenty minutes describing a misaligned button. The free plan covers 10 recorded sessions a month, which is enough to test the idea on one project. See the plans, or try the demo first.