AI Meeting Follow-Up: Summaries, Action Items, and Emails

Why follow-up is where AI saves the most time

Most of the work after a meeting is rewriting. The same conversation becomes an email to the client, a note for colleagues, a CRM update, and a list of tasks. Each version says roughly the same thing to a different audience, and each one takes longer than it should when you have three more meetings that afternoon.

That is a writing problem, and AI assistants are good at writing problems. What they cannot do is know what happened in the meeting beyond what you give them. A follow-up that gets the client's request wrong is worse than one that arrives a day late.

This post covers turning notes into follow-up for a single meeting. For capturing the notes in the first place, including when to use an AI notetaker, see AI notetakers in client meetings. For following up a whole conference, see our post-event follow-up plan.

Start with notes worth summarising

An assistant can only summarise what it is given. There are two common starting points:

• A transcript from a notetaker. Complete, but long, and full of small talk, false starts, and crosstalk that the assistant has to interpret.
• Your own notes. Shorter, and already filtered by someone who understood the conversation, but only as good as you made them.

If you rely on your own notes, the most valuable habit is a two-minute record straight after the meeting: what they asked for, what was agreed, who owns each action, and anything you promised to send. Typed into your phone in the corridor or spoken into a voice memo, it is accurate. Reconstructed the next morning, it is guesswork.

Then give the assistant context it would not otherwise have: who attended and their roles, what the meeting was for, and which side you were on. A first name in a transcript could belong to your colleague or to the client's chief financial officer.

Ask for decisions and actions, not a recap

The default output of "summarise this meeting" is a recap: a few paragraphs restating the conversation in order. Recaps are rarely read. What people actually need is:

• Decisions made, stated plainly.
• Actions, each with an owner and a date. If the notes do not name an owner, the assistant should say so rather than pick one.
• Open questions that nobody answered.
• Concerns the client raised, in their words where possible.

A prompt that asks for exactly that:

From the notes below, list: (1) decisions made; (2) action items with owner and due date, writing "not agreed" where the notes do not say; (3) open questions; (4) concerns the client raised, quoting them where possible. Use only what is in the notes. Do not add actions or commitments that are not there.

Then read the list against your memory of the meeting. An action marked "not agreed" is useful information: it is the thing to settle in the follow-up email.

Write the follow-up email

A good follow-up is short, specific, and moves things on. The structure that tends to work:

1. One line that refers to something specific from the conversation, to show it was listened to.
2. One line restating what they care about, in their terms rather than yours.
3. The agreed next step, with a proposed time or a way to book it.
4. Anything you promised to send, attached or linked.
5. A sign-off from the person they met.

Ask the assistant for exactly that, in four or five sentences, and in the tone you would use with this person. Our AI meeting prompt builder has a "Follow-up email" option that builds this prompt from your notes and tells the assistant not to add commitments that are not in them. What you enter is sent to us when you copy the prompt, so leave out anything confidential.

Tailor the email to the reader. An executive sponsor wants the outcome and any decision needed from them. A technical evaluator wants specific answers to their questions. If both attended, it is often better to send one short email to everyone and a detailed answer to the person who asked.

Then rewrite the result until it sounds like you. Clients notice when a follow-up reads like everyone else's.

Write a separate note for your own team

What you tell colleagues and what you tell the client are different documents. The internal note can be candid: the budget is not confirmed, the champion seemed to lack authority, a competitor is further ahead than expected, the meeting probably should not have been booked.

Keep the two apart. If you generate both from the same transcript in the same conversation with an assistant, check carefully that nothing from the internal note has leaked into the client email. A frank remark about the client's chances of getting budget approved, sent to the client, cannot be taken back.

The internal note is also where the handoff to a colleague starts, if someone else will own the next meeting.

Update the CRM while it is fresh

The step that most often gets skipped is the one that makes the meeting useful to anyone else: getting the outcome into the CRM.

An assistant can map your notes onto the fields you use (next step, next-step date, stakeholders and roles, stated needs, objections, competitors mentioned) so the update takes a minute rather than ten. Many CRMs now offer a version of this built in. Either way:

• Check every field against the notes. A wrong next-step date quietly breaks forecasts and reminders.
• Keep opinions out of shared fields. "Champion seems weak" belongs in an internal note, not in a field that may be synced to other systems.
• Log the next meeting as a meeting, not as a note saying one will happen.

Checks before you press send

Before any AI-assisted follow-up leaves your outbox, check that:

• Every name, title, and company is spelled correctly.
• Numbers, dates, and times are right, with time zones where it matters. Our time zone converter helps with that.
• Every commitment in the email was actually made in the meeting.
• Every action has an owner, and the client's actions read as requests, not instructions.
• Nothing from internal notes or pre-meeting chat has crept in.
• Anything you promised to send is attached.
• It sounds like you.

Speed helps a follow-up land while the meeting is fresh. Accuracy is what keeps the client's trust. AI makes the first easy; the checks take care of the second.

CallShark is being built to keep agreed next steps and owners attached to each meeting, so follow-up starts from a record rather than memory. The product is temporarily down.