Ten minutes before a renewal call, an account manager opens Claude and asks one question: what did this client say about pricing and next steps on our last few calls? Claude pulls up the relevant calls and notes for that client and answers from them. She reads the answer, walks into the meeting knowing what was said, and hasn’t opened a single recording.
That is possible now, and it works because the assistant can read a record of her team’s calls. Most people don’t have that in place, so they either paste notes into a chat window or go without.
When AI help with meetings falls flat, the tool is rarely the weak point. It knows nothing about a meeting unless it has something to work from, and what you give it decides how good the answer is. The rest of this guide covers how to supply that context yourself, and where doing it by hand stops being practical.
Start from what the assistant can’t know
Both tools are good at writing and summarising whatever you put in front of them. Neither can see your recordings or your colleagues’ notes unless you connect them. A prompt like “write a follow-up to today’s call” gives the assistant a task with no material, so it fills the space with generic phrases about “circling back” and “next steps”.
The fix is context, and it matters more than clever prompt wording. A plain request with the right material attached will beat a polished prompt with nothing behind it.
Choose what to hand over
You have three realistic options, and they suit different jobs.
- The full transcript: the most detail, including the exact wording of a promise or an objection. It is also long and full of filler, and very long transcripts can strain how much a model handles well in one go.
- Meeting notes or a summary: shorter and easier to work with, though the reasoning behind a decision often gets lost on the way to a bullet point.
- Your own bullet points: the quickest route, and only as good as your memory an hour after the call.
For anything that matters, notes plus the short stretch of transcript around the key moment tend to work better than either alone. The notes give the shape of the meeting and the transcript gives the exact words when you need them.
Say what the material is and what you want
Assistants do better when you explain the situation before the pasted text. Tell it who was in the meeting, what the meeting was for, and what you want back. Something like this works:
“This is a transcript of a call with a prospect who is deciding between us and one other supplier. Draft a follow-up email that confirms the two next steps they agreed to. Don’t mention pricing, because we didn’t discuss it.”
That last instruction does more work than it looks like. Telling the tool what to leave out stops it from inventing detail to sound helpful.
Ask in stages and check the facts
Paste the material once and then ask for things one at a time: a short summary first, then the action items, then the email. Each answer gives you something to correct before the next one builds on it. If you ask for everything at once, a small misreading at the start carries through the whole response.
Check names and figures against the source before anything goes out. Assistants can state something with confidence that nobody said in the meeting, particularly when the transcript is messy or two people spoke over each other. Asking the tool to point to the line where a commitment was made is a cheap way to catch this, because a line that isn’t there is easy to spot.
Be careful what you paste
A transcript can hold customer names and prices, and sometimes personal or HR matters. Before you put one into a chatbot, check your company’s policy on AI tools, and look at your plan’s data settings to see whether conversations can be used to train the model. If the call was recorded, the people on it should have known. A good test is whether you would send the same text to an outside contractor without asking anyone first.
Where copy and paste runs out
Pasting works well for one meeting. It gets tedious for a team with dozens of calls a week, because each new chat starts with no knowledge of the last one unless you rebuild the background every time. The person who handles an account ends up re-pasting the same history, and anyone else who picks it up starts from nothing.
A growing number of teams deal with this by setting up meeting memory for AI, a searchable record of past calls that the assistant can query when you ask about an account or a project, so nobody has to paste the background in first. It doesn’t remove the need to check what comes back, but it takes the copying out of the routine.
A sensible way to start
Pick one recurring meeting, such as a weekly customer check-in, and build a short notes template for it with the same few headings each time. Use it for two weeks and keep a note of what you find yourself explaining to the assistant over and over. That list tells you which context is worth keeping ready, and which is a one-off that is fine to paste when it comes up.
You will probably find that a handful of facts account for most of the repetition, such as who the client is and what has already been promised. Those are the ones to sort out first, and the rest can wait until it becomes a nuisance.