0.7.12 Accept, Revise, Verify, or Reject

Every AI Output Requires a Decision

After receiving an AI response, you need to decide what to do with it.

A useful four-part framework is:

  • Accept
  • Revise
  • Verify
  • Reject

These choices are not grades for the AI.

They describe your next action.

Accept

Choose Accept when the output is appropriate for the task and does not require meaningful correction or additional verification before use.

This is most appropriate when:

  • the task is low risk;
  • the content is within the allowed use;
  • the response meets the requested purpose;
  • important factual claims are already supported or not material to the task.

Example:

If one heading fits the purpose, accepting it may require little additional work.

Revise

Choose Revise when the response is useful but needs changes.

Reasons can include:

  • the tone is wrong;
  • the wording is too long;
  • the audience is different;
  • part of the content is inaccurate;
  • the response includes unnecessary information;
  • important context is missing.

Example:

An AI drafts a professional message correctly but includes a detail you do not want to share.

You can remove the unnecessary detail and revise the message.

Verify

Choose Verify when an important claim needs evidence before you can rely on it.

Examples include claims about:

  • current course requirements;
  • software configuration;
  • current schedules;
  • policies;
  • technical specifications;
  • other facts that affect a decision.

Verification means finding appropriate evidence and comparing it with the claim.

Do not treat "Verify" as a permanent final state.

After checking the evidence, you may:

  • accept;
  • revise;
  • reject.
Reject

Choose Reject when the output should not be used.

Reasons can include:

  • the answer is substantially wrong;
  • the response does not address the task;
  • the content violates the rules of the activity;
  • it requires sharing information you should not provide;
  • the output is based on a false assumption that makes the whole result unusable.

Rejection can be the most responsible choice when revision would require rebuilding nearly everything.

One Response Can Require More Than One Action

Suppose an AI-generated technical explanation contains:

  • a strong introductory paragraph;
  • one current version claim;
  • an irrelevant final section.

You might:

  • accept the introduction;
  • verify the version claim;
  • reject the irrelevant section.

The framework can be applied to parts of a response rather than only to the response as one block.

Risk Changes the Decision

A creative wording suggestion and a factual technology recommendation do not carry the same risk.

For low-risk creative work, acceptance may be reasonable quickly.

For a claim that could affect:

  • security;
  • privacy;
  • grades;
  • production systems;
  • important professional actions;

verification may be necessary even when the answer sounds convincing.

Evidence Should Drive the Decision

Weak justification:

Stronger justification:

Weak justification:

Stronger justification:

A Simple Decision Flow

Conceptually:

Activity diagram showing Receive output, then Does it fit the task?, then no Reject or substantially revise, then Does an important claim need evidence?, then yes Verify, then Does the output need meaningful changes?.
ReceiveOutput.
  • New
  • Changed
  • Removed

Real decisions are not always perfectly linear.

The model helps make your reasoning visible.

The Framework Keeps the Human in Control

AI produces the output.

You decide whether the output is usable.

The important skill is not choosing the same action every time.

It is choosing an action that matches:

  • the task;
  • the evidence;
  • the risk;
  • the rules;
  • the quality of the response.