# Homework AI-Use Decision Tree

Companion to [AI homework help vs. cheating: a decision tree by learning objective](/articles/ai-homework-help-vs-cheating).

## Step 0: check the actual rule first

- [ ] Checked the syllabus, assignment sheet, or teacher's stated AI-use policy for this task.
- [ ] If unclear, asked the teacher directly with a specific example rather than guessing.
- [ ] The school's or teacher's rule is the one that controls — not household preference.

No stated policy does not mean "assume yes." Treat silence as "ask," not "default to permitted."

## The four questions — fill in per assignment

| Assignment: | ________________________________________________ |
| --- | --- |

**1. What skill is this assignment actually measuring?** (Not the subject — the skill.)

______________________________________________________________

**2. Does the intended AI use replace that skill or support it?**

- [ ] Supports it (child still performs the graded skill afterward)
- [ ] Replaces it (the AI performed the graded skill)
- [ ] Genuinely ambiguous — go to question 4 for the deciding test

**3. Does the assignment require disclosure of AI use?**

- [ ] Yes — disclosed as: ________________________________________
- [ ] No requirement stated
- [ ] Unclear — asked the teacher

**4. Can the child independently demonstrate the skill without the tool?**

Test it directly, not hypothetically — have them redo a similar problem or explain their reasoning
unaided.

- [ ] Yes, demonstrated independently
- [ ] Not yet — this is a learning gap worth addressing regardless of any rule

## Verdict

- [ ] **Practice help** — supports the skill, disclosure not required or completed, independent
      demonstration confirmed.
- [ ] **Disclosed collaboration** — some AI contribution to the output, disclosed per the
      assignment's rules, independent demonstration confirmed.
- [ ] **Substitution** — the graded skill was performed by the AI, disclosure was skipped where
      required, or independent demonstration failed. Address before submitting.

## Three worked examples for reference

| Task | Skill measured | Use | Verdict |
| --- | --- | --- | --- |
| Vocabulary practice | Recall and usage of assigned words | AI generates example sentences to study from | Practice help, if recall later confirmed unaided |
| History essay | Constructing an argument from evidence | AI writes the full essay from an outline | Substitution — the graded work happened in the AI's process |
| Coding debug task | Locating and fixing a bug class | AI explains the bug, child fixes it after understanding | Ambiguous — resolved by testing independent debugging afterward |

## A note on AI detection tools

Automated AI-text detectors are not a reliable way to enforce any of the above, on either side. OpenAI
discontinued its own detector in 2023 over accuracy problems and has stated no tool reliably tells
AI-generated text apart from human writing. Peer-reviewed work has shown higher false-positive
rates for non-native English writing; claims about neurodivergent writers are less settled. Use
the four questions above and an
actual independent-demonstration check — not a detector score — to decide.

## Log across the term

| Date | Assignment | Verdict | Independent demonstration confirmed? |
| --- | --- | --- | --- |
| | | | |
| | | | |
| | | | |

## Related reading

- [The family AI agreement](/articles/family-ai-agreement) — for setting the household's disclosure
  norm before it comes up at school.
- [A child's first AI conversation](/articles/child-first-ai-conversation) — the verification habit
  that applies whether or not AI use is permitted on a given task.