Using AI to write client updates without sounding like a robot

The short answer

An AI writes a robotic client update when it has nothing specific to say. Give it the facts from the project, a fixed shape, and a short list of things it must not write, and the draft gets close to what a person would send. Then cut the first and last line, check every date, and add one sentence only you could write.

The facts matter more than the prompt. "Write a weekly update for the Harbour project" produces filler. "Here are the tickets finished this week, the ones due next week and the two answers we are waiting for" produces an update.

Why AI updates sound like a bot

Most of the tells are habits the models picked up from a lot of corporate email. Once you know them you see them everywhere.

• The opener. "I hope this email finds you well" or "I wanted to give you a quick update on our exciting progress". Nobody needs it.
• Mood instead of facts. "Things are moving along nicely" says nothing. The client wants to know what moved.
• Lists of three for rhythm. "We have been busy refining, polishing and perfecting." Two of those words are padding.
• Bold headings on a five-line email. It reads like a report template, because it is one.
• Hedging. "We may potentially be looking at a slight delay." Either the date moves or it does not.
• The closer. "Let me know if you have any questions at all." The client knows how email works.
• Dashes everywhere, and a summary paragraph that repeats what the email just said.

None of these is wrong on its own. Together they tell the client nobody looked at their project this week.

What a client update needs

A useful update answers four questions, in this order, in as few lines as it takes.

1. What was done since the last update.
2. What happens next, with dates.
3. What we need from you, and by when.
4. Anything at risk, said plainly.

That is it. If there is nothing at risk, leave the fourth line out. If nothing is needed from the client, say so: "Nothing needed from you this week" is a sentence clients like reading.

Most agencies send this weekly on long jobs and at each milestone on short ones. Send it on the same day every week. A client who gets an update every Friday stops writing "where are we?" emails on Wednesday. How much of this a client portal can carry instead of email is in what a client portal should do.

Give it facts, not a mood

The input decides the output. Before you ask for a draft, collect:

• The tickets or tasks finished since the last update, by name.
• What is due next, with the dates.
• What you are waiting on from the client: approvals, content, feedback, a decision.
• Anything that moved: a date, a scope change, a deliverable added.

If your project tool has an AI assistant that can read the project, ask it for this list first and check it. If it does not, copy the list out of your board by hand. It takes three minutes and is the part that makes the update true.

A prompt that works

Paste the facts under this and keep the rules in. The rules do more than the instructions.

Write a weekly update to a client from the facts below.

Shape: four short parts, no headings.
1. What was done this week.
2. What is next, with dates.
3. What we need from the client, with a date.
4. Risks, only if there is one.

Rules:
- Under 120 words.
- Plain sentences. No opener, no sign-off line, no summary.
- No "excited", "thrilled", "great strides", "moving along nicely".
- No dashes, no bold, no emoji, no exclamation marks.
- Use the ticket names as they are. Do not invent dates or tasks.
- If a fact is missing, write [CHECK] instead of guessing.

Facts:
[paste here]

The [CHECK] rule is the useful one. It turns the AI's habit of filling gaps into a visible marker you can search for before sending.

Before and after

Here is a typical first draft from a vague prompt:

Hi Sarah,

I hope you're doing well. I wanted to share a quick update on the
exciting progress we've been making on the Harbour rebrand.

The team has been hard at work refining the logo concepts, and we're
thrilled with how things are shaping up. We're also making great
strides on the brand guidelines.

Let me know if you have any questions at all.

And the same update from the facts and the prompt above, after a two-minute edit:

Hi Sarah,

Done this week: the two logo routes you picked are now in colour, and
the type pairing for both is set.

Next: the brand guidelines draft goes to you on Thursday 13 November.

From you: the final company name for the legal line, by Tuesday, so
it is in the guidelines.

Risk: if the name changes after Tuesday, the guidelines move by two
days.

Tom

The second one is shorter, has three dates and one request, and the client can answer it in a line.

The edit pass

Read every AI draft once before it goes out, and make these four changes.

1. Delete the first sentence if it is a greeting dressed up as content.
2. Delete the last sentence if it offers help or summarises.
3. Check every date and every deliverable name against the project. This is where the AI is most often wrong.
4. Add one line that comes from you: the thing the client mentioned on the last call, or a plain opinion on which logo route you would pick.

The fourth step is what makes it read like a person. It is also the line clients reply to.

Email or status page

An update by email gets read once and then buried. A status page the client can open at any time keeps the history. Many agencies do both: a short email on Friday with a link to the page that has the detail.

If you post updates to a page the client sees, keep internal notes and client-facing lines apart. The AI should draft the client line; it should never be able to publish an internal one to the client by mistake.

How this works in OIO

In OIO, each project has status updates, and every update has an Internal or Client switch. Updates are Internal by default, so a draft or a team note never reaches the client unless you switch it. Client updates show on the client's status page next to the tickets, and the page refreshes itself while it is open.

The assistant can read a project's tickets and draft the update from what actually moved, then post it as a status line when you tell it to. On Studio, an outside AI app connected through MCP can do the same: read the tickets, draft the update, post it to the project. Every change is in Logs. For the setup, see project planning with AI, and for the rest of the workflow around it, AI project management for agencies.

If a client goes quiet after an update and an invoice is involved, the payment reminder email tool writes the follow-up for you.

Questions