How I Automate Client Updates With AI So Clients Stay
A client cares about three things. Is my thing moving, is anything waiting on me, and is anything wrong. So keep the update to three or four short paragraphs. One line to open, what got finished, what is moving now, and the one thing you need from them. Let the machine write the first draft from the board, the code, the channel, the calls, their accounts and the end of day log. Then a person reads it and sends it.
- A client cares if their thing is moving, if anything is waiting on them, and if anything is wrong
- Three or four short paragraphs, and I stopped using bullet points
- The facts come from the board, the code, the channel, the calls, their accounts and the log
- The machine writes the first draft and a person reads it

Client updates are the most important thing that makes a client stay with you.
So this post is how I write mine, and how I use AI to draft them. It covers what a client wants to hear and the shape of the message. Then where the facts come from, what to cut, and the weekly one.
This is the written version of my video. Watch it here: https://youtu.be/7sXI4dK1pug
The exact skill I use is free on GitHub: github.com/qemoza/client-updates-skill
The three things a client wants to hear
Clients care about three things.
- Is my thing moving?
- Is anything waiting on me?
- Is anything wrong, or is there a problem?
The first two are the most important.

The shape of the update
An update can look something like this.
- One line to open. I usually send something like hey, quick one or quick update.
- What got finished. Just one line. What is now true, what is already done.
- What is moving right now. If it is still being tested, you can say so.
- The one thing you need from them, in a sentence. If you need nothing, you just don't talk about it.

An example
Hey, quick update.
We just finished phase one of the project, which is X, Y and Z.
Right now we are moving to phase two. The goal is to finish X, Y and Z on this date.
By the way, I need one thing from you. If you can send me the clients list, that would be awesome. Nothing else is needed from your side. Talk to you soon.
It should feel human. I used to send bullet points. I don't anymore. Just three or four short paragraphs. It should not be a huge thing for them to read.
Where the facts come from
These are the sources the update comes from.
| Source | What it tells you |
|---|---|
| The board | The tasks that closed that day |
| The code | What actually shipped |
| The channel | What they already know, and what they sent us |
| The calls | What you promised them |
| Their accounts | The numbers |
| The log | What we wrote down at the end of each day |
The board is the project management tool you use. I have my own. It is where we track which tasks closed that day.

The code is what actually ships. If you ship code, you can have a skill that tracks what is live.
The channel is what they already know, so we don't say it twice. It is also what they already sent us, so we don't neglect their questions. We connect to Slack or email, whatever you use to talk to them.
The calls are usually where you promise them something. Always read your calls when you write an update. Use them as a data source.

Their accounts are the numbers. If you run ads for them, that is where the numbers are.
The log is what we wrote down at the end of each day. What we said yesterday should be taken into account when we say something today.
The machine writes the first draft
The machine writes the first draft, and a person reads it.
You can do this every day or every week. You can have it as an automation, or you can have it as a Claude skill.
The way I have it is a skill called /update. When the team ends their day, they run it in their Claude session. It checks all the sources. It checks what they did in that session. Then it writes the update and adds it to the CRM.

When more than one person works on the same client
The skill does the same thing, and it also checks the CRM to see if we already have an update. Then it only adds the new things, one time.
So the client is not getting one update from the media buyer, one from the marketer and so on. They only get one update. That is all they care about.
The next day, I send it
The next day I check the CRM. I have a skill that double checks everything. Then I check myself what the team said, and I send it to the client.
What to cut before it goes
You usually need to cut:
- What they already know.
- How it works under the hood. All the technical things, unless they ask about it.
- The scary number they cannot even act on. You should not scare them for nothing.
And you always say so when something is still being tested, or when there is something that needs them.

The weekly update
If it is a weekly report, you do the same thing. You can also do it as a Loom video. And you just have two or three numbers that they make a decision on.

Get the skill free
The exact skill I use for my updates is free. It is a GitHub repo with all the steps. You download it and give it to Claude. Claude will take it from there.
github.com/qemoza/client-updates-skill
Or comment UPDATE on the video and I will send it to you: https://youtu.be/7sXI4dK1pug
FAQ
What should a client update say? What got finished, what is moving right now, and the one thing you need from them. A client cares if their thing is moving, if anything is waiting on them, and if anything is wrong.
How long should a client update be? Three or four short paragraphs. It should not be a huge thing for them to read.
Should a client update use bullet points? I used to send bullet points. I don't anymore. It should feel human.
Can AI write client updates? The machine writes the first draft and a person reads it. I check what the team said myself, and then I send it to the client.
How often should I send client updates? You can do this every day or every week.
What goes in a weekly client report? The same thing, plus two or three numbers that they make a decision on. You can also do it as a Loom video.


