How do I stop my clients from asking where their project stands?

https://youtu.be/t4Dopx7fnII

Make the project readable instead of writing updates. When a client asks where things stand, they are telling you the job went dark on their end. The fix is one project per client, structured so someone who was not in the meeting can read it in thirty seconds, fed by the tools that already record your work. I run ten clients this way. Writing updates takes me nothing now, and checking the system takes about thirty minutes a day.

Not sure whether your shop is ready for this? Run the AI-First Assessment. Five minutes, free, no email.

Why does the client keep asking?
What does answering by hand actually cost?
What has to be true before you automate anything?
How do you structure a project a client can read?
What feeds the project?
The mistake that makes this worse than doing nothing
The 90-Second Rule
Where it still breaks
Questions Builders Are Asking
What's Next
Sources

Why does the client keep asking?

Because they cannot see the job.

That is the whole answer. Most of us hear that question as a scheduling problem. It is not.

A client who asks where things stand has been guessing for a while. They got tired of guessing. They are not checking up on you.

Now count what that week cost. The client spent it uncertain. You spent it not knowing they were uncertain. Somewhere in there they told their spouse they were not sure what was going on.

That conversation is the one that shows up later. It shows up as a scope argument. Or a referral that never came.

The money version of this is well documented in remodeling. About 40% of change orders never get invoiced. Roughly $2,847 per project walks out the door in changes nobody wrote down. Three disputes a project over work that was decided out loud and recorded nowhere.

Those look like billing numbers. They are visibility numbers. Nobody disputes a change order they watched get approved.

Four questions your client should be able to answer without calling you:

What is in scope?
What are you working on right now?
Are we on pace?
What happens next?

Every one of those has an answer. The problem is where the answer lives.

What does answering by hand actually cost?

Six hours a week, in my shop. That was time spent moving tasks around and writing updates.

Six hours a week. Fifty weeks. 300 hours a year.

That is the number that is easy to quote. It is also the smaller of the two costs.

The bigger cost has no clean number on it. That is exactly why nobody manages it. It is the client who could not see the job.

I fixed the six hours first and nothing improved. Then I fixed the visibility, and the six hours went away on their own.

Fix what the client can see. The time savings follow.

What has to be true before you automate anything?

Three things. Only one of them is software.

1. Decisions have to get said somewhere a system can hear them.

A scope change decided in a text thread is invisible. So is one decided on the job site. So is the one you made in the truck on the way home.

No tool fixes that. This is where most builds die, and people blame the AI. The AI was fine. The input was never there.

2. One project per client. Not one per topic.

Ten clients means ten projects. If a client's work is spread across four lists and a folder, you do not have a project. You have a filing system.

3. Pick your status words and stop changing them.

"On pace" has to mean the same thing in week nine as it did in week two. When the words drift, the client learns they mean nothing. Then they start calling again.

How do you structure a project a client can read?

Four sections. The number is the point.

In scope. What you agreed to.
Moving now. What is being worked this week.
Waiting on you. Decisions sitting on the client's side.
Next. What comes after this block.

The temptation is to build something elegant. Fourteen categories. Nested tags. Technically perfect.

Nobody reads a taxonomy. Four sections a client scans in thirty seconds beat fourteen nobody opens.

The harder change is how you write the tasks.

Most of us write tasks for ourselves. Shorthand. Three words that mean something because we were in the room.

Rewrite them as full sentences with the decision inside.

Not this: Kitchen selections.

This: Client picked the matte quartz over the honed, which adds four days to the countertop lead time.

Someone who was not in the meeting can read that. That is the whole test.

There is no AI in this step. It is the step that does the most work.

If a project has to be explained, it is not a project. It is your notes.

What feeds the project?

Whatever you already run.

This is where builders get sold a portal. A portal is usually the wrong first move. It is one more place where somebody has to type the truth a second time.

Here is my stack:

Source	What it captures
Day.ai	Client meetings and what got decided in them
Google Workspace	Email and calendar
Quo	The business phone line, calls and texts

A Claude skill reads all three and writes into the project twice a day. Morning and evening.

Morning catches yesterday before it goes stale. Evening catches today before you have to reconstruct it.

What the system is allowed to do matters more than how it works. It can add, update, and flag. It does not close anything and it does not change scope. Those two stay with a human.

The task tool itself is boring on purpose. I use Todoist. Not because it is special. Because it is cheap and I already had it.

The tool is not where the value is. The value is in what feeds it and who checks it.

That is the same conclusion I came to when I shut down Drone Pros in 2021. Selling a tool into a business with no process makes shelf-ware, not results. Build the process, then pick the simplest thing that runs it.

Other operators are landing in the same place. Kyle Rossignol came out of contracting and now works on AI tooling for construction. Luke Thomas is a CPA co-founding BlueCollar Cloud Solutions, building software for the same trades operator. Both started from the work, not from the software.

Want a read on which part of your operation is ready first? Run the AI-First Assessment.

The mistake that makes this worse than doing nothing

Call it the transcript trap.

A recording system is very good at capturing that a sentence happened. It has no idea whether that sentence was a decision or somebody thinking out loud.

Here is how it plays out.

Your client floats an idea on a call. "We could maybe move the pantry wall." Nothing was decided. Nobody committed.

The system writes it down. The sentence happened.

Now it is in the project. A week later the client reads it back as a commitment you made. You are managing an expectation you never agreed to, and you handed it to them yourself.

This is why an automated update can be worse than no update.

Silence leaves a client uncertain. A wrong update leaves them certain and wrong. Certain and wrong costs far more to unwind.

I have not found a technical fix. Better prompting does not solve it. The fix is a rule.

The 90-Second Rule

Every client project gets ninety seconds of human eyes per run, twice a day. If a project needs more than ninety seconds to check, the system did not write it clearly enough. That is my problem to fix, not my client's problem to decode.

Here is the math:

10 clients × 90 seconds × 2 runs = 30 minutes a day
15 clients × 90 seconds × 2 runs = 45 minutes a day

Forty-five minutes a day is 3.75 hours a week. I was spending six hours a week before.

At fifteen clients, this stops paying for itself.

That is why fifteen is my ceiling and not my goal. I run ten. I know where the wall is and what it is made of.

The rule catches something else too. When a project runs past ninety seconds, that is a signal about the writing. It sends me back to the task titles, not to the client.

And when you read, do not check whether the system worked. Check something harder: would a person who was not in the room come away with the right picture?

Different question. Catches different problems.

This is not a fully automated system and I would not want one. Automation writes the draft. A person decides what a client is allowed to believe.

The same principle is showing up across residential construction tech. Grace Tsao Mase is building HomeRES for the homeowner side of that relationship, and the company was named a winner in Fast Company's 2026 World Changing Ideas Awards. The direction of travel is the same everywhere: give the client a view, do not make them ask for one.

Where it still breaks

Four things, said flat.

It only sees what got said somewhere it can reach. Text threads, job-site conversations, and decisions made in your head stay invisible. That is not an AI limitation. That is physics.
It writes what was said, not what was decided. The transcript trap. I have not solved it. The ninety seconds exist because of it.
It cannot tell whether the client understood. It makes the picture available. A human still has to notice when a client goes quiet.
Fifteen is arithmetic, not experience. I have run ten. If the real review time turns out to be two minutes, the ceiling drops.

Questions Builders Are Asking

Do I need a client portal for this? No, and a portal is often the wrong first move. A portal is a second place where somebody has to type the truth, so now the truth is maintained twice. Point the system at where your work already gets recorded, then make one of those places readable.

What software do I need? A task tool you already use, a way to record client meetings, and access to your email and phone records. I run Todoist, Day.ai, Google Workspace, and Quo. The specific tools matter far less than whether your decisions are getting captured at all.

Can a project manager do this instead of automation? Yes, and for many shops that is the better answer. A PM who writes clear updates solves the same problem. I built the system because I am a solo operator and a $65K operations manager was not on the table. If you can hire, hire.

How is this different from an automated weekly email? A weekly email is a broadcast on a schedule. This is a project the client opens whenever they want and reads the current state from. The difference shows up on the Tuesday a client wonders and the email does not come until Friday.

What if my client never opens it? Then you learned something. A client who never opens the project either does not need that detail or does not trust it yet. Both are useful. Both are invisible when you write updates by hand.

Will this work if my job data is a mess? It will produce a wrong picture faster than you would produce one by hand. Clean the inputs first. This is the same reason a messy cost catalog produces bad AI estimates. The system amplifies what you feed it, in both directions.

Is it legal to point this at recorded client calls? Tell your clients you record meetings, and get it in writing at the start of the engagement. Recording consent laws vary by state, and two-party-consent states require everyone on the call to agree. Have the conversation once, up front.

What's Next

Pick one client. One project. Make it readable by someone who was not in the meeting.

Do not automate anything yet. You will know inside a week whether the rest is worth building, and you will have already done the part that carries most of the value.

Then two things.

Run the AI-First Assessment to see where your operation actually stands. Five minutes, free, no email.
Watch the full build on Beyond the Bid. I walk all three parts on screen with a real client project, including the part where the system got something wrong.

Sources

Change-order figures ($2,847 per project, 40% never invoiced, 3 disputes per project) are segment pattern figures from GO First's approved reference set, not audited client results.