AI Chore Chart: How AI Decides a Chore Is Actually Done
An AI chore chart is a chore chart where an AI does the first pass on whether each chore was really finished, instead of a parent spot-checking every one. The child submits a photo, the AI compares it against what the chore asked for, and it either clears the chore or hands it to a parent with an explanation. This guide covers exactly how that decision gets made, what the AI can and cannot see, and what happens when someone disagrees with it.
What is an AI chore chart?
An AI chore chart is a chore chart where the "is it done?" step is handled by an AI reviewing a photo, rather than by a parent walking into the room. The chart itself is ordinary — chores, assignments, points, a schedule. The AI only changes one thing: what has to happen before a chore counts.
That one change is the whole point. Most chore charts fail at the same place, and it is not the chart. It is the argument that happens after a child says they are finished and a parent looks at the room and disagrees. A chore chart with no verification step makes that argument a daily event. A chore chart that requires a photo moves the argument to something concrete. A chore chart with AI verification means most of those photos never need a parent at all.
It is worth being clear about what this is not. An AI chore chart does not decide what chores your child should have, does not punish anyone, and does not replace you. It is a first-pass filter on evidence you were already asking for.
How does AI photo verification decide a chore is done?
The AI is given the photo plus the chore's title and description, and it returns three things: a yes-or-no decision, a confidence score from 0 to 100, and a short written explanation. A chore clears automatically only when both conditions hold — the AI says yes and its confidence is at or above the threshold the parent set. Everything else goes to a human.
In ChoreSplit that model is Google's Gemini, the default threshold is 80, and parents can move it anywhere between 50 and 95. If the chore carries a custom instruction — "the counter is wiped and nothing is left in the sink" — that instruction replaces the generic prompt, and the AI is checking against your words instead of its own idea of clean.
The decision, in order
- Child finishes the chore and submits a photo in the app.
- The photo, the chore's wording, and any reference photos you have previously approved for that chore go to the model together.
- The model returns
approved,confidence(0–100), and written feedback for the child. - Approved and confidence ≥ your threshold → the chore clears and the points post.
- Anything else → the chore sits in your approval queue with the photo and the AI's reasoning attached, and you get an email.
The AI can approve. It can never reject.
This is the most important design decision in the whole system and it is worth stating flatly: there is no code path in which software tells a child their chore does not count. The AI either clears a chore or escalates it to a parent. A child can be told no — but only by you, and only with a written reason.
What can AI photo verification actually detect — and what can't it?
AI photo verification is good at judging whether an image shows the visible end state a chore described, and it is bad at everything that is not visible in that image. It cannot tell who did the work, when the photo was taken, or what is happening outside the frame. Anyone selling you infallible chore verification is selling you something that does not exist.
| Reliable | Not reliable |
|---|---|
|
|
One more caveat about the confidence score, because it is easy to misread: a confidence of 80 is not a claim that the model is right 80% of the time. It is the model's own report of how sure it is, and models are sometimes confidently wrong. Treat the threshold as a dial for how much you want to review, not as a measured accuracy rate.
What happens on a disputed chore?
Every dispute resolves with a parent, because the AI has no authority to close one. A chore the AI did not clear is pending, not failed — it is waiting in your queue with the photo and the AI's written reasoning, and you either approve it or send it back. A chore the AI did clear that you disagree with can be overridden after the fact.
The AI said no, the child says yes
Open the photo yourself and approve it. The AI's opinion carries no weight against yours, and approving it teaches the system your standard for next time.
The AI said yes, you disagree
Override it. ChoreSplit requires you to write a reason before it will let you, and that reason is what the child receives as a redo request — so no child is ever told no without being told why.
It keeps happening
That is a threshold problem, not a chore problem. ChoreSplit counts your confirmations against your overrides and suggests moving the threshold once it has seen enough of your decisions to justify it.
If the AI is unavailable, or the family is out of AI credits, or anything else goes wrong on our end, submissions simply route to your approval queue the way they would if you had never turned AI on. The child's photo is never lost and their work never disappears.
How do you set up an AI chore chart?
Turn verification on for one or two chores — the ones that actually cause arguments — write down what "done" means for each, and leave everything else alone. Six steps, and the first one is the one most families get wrong by doing too much.
- 1
Turn verification on for one chore, not all of them
Open the chore and switch on 'requires verification', then switch on AI review. Start with the one or two chores that actually cause arguments. Verification is off by default for a reason — putting a camera on every chore turns a motivation system into a surveillance system.
- 2
Write the standard down
Each chore can carry its own verification instruction, and that instruction replaces the generic one. 'The bed is made with the pillows on top and nothing on the floor' gets checked far more consistently than 'clean your room', because the AI is matching the photo against words you chose.
- 3
Set your confidence threshold
The default is 80 out of 100. Anything the AI approves at or above that clears automatically; everything else comes to you. The range is 50 to 95 — 50 means you want the AI to clear most things and only escalate the obvious misses, 95 means you want to see almost everything yourself.
- 4
Let the child submit the photo
The child finishes the chore, takes the photo in the app, and submits. The AI evaluates it in a few seconds and returns an approve/not-approved decision, a confidence score, and a short written explanation the child can read.
- 5
Review what did not clear
Anything the AI did not confidently approve lands in your approvals queue and you get an email. You approve it, or you send a redo request with a written note. The note is required — the child always gets a reason, never just a rejection.
- 6
Let it calibrate to your family
When you confirm an AI approval, that photo becomes a reference example for that chore, so later submissions are judged against what you have already accepted. Over time the AI's idea of 'clean enough' converges on yours rather than on a generic standard.
Should every family use an AI chore chart?
No. AI verification earns its place when the same chore causes the same argument every week — a bedroom, a bathroom, a kitchen counter — and it is dead weight everywhere else. If your chore chart is working, a camera will not improve it, and photographing a five-year-old's every task will make chores feel like an inspection.
The families it helps most are the ones where a parent is spending real time re-checking work, or where two parents in two households need to agree on whether something got done without either of them standing in the room. Photo evidence plus a written standard settles that. The AI just means nobody has to look at the easy 90%.