When to Reject vs Edit
Session 11.7 · ~5 min read
The Sunk Cost of Tokens
You generated a 1,500-word article. It cost $0.03 in API tokens and took 8 seconds. You read it and something is wrong. Not the grammar, not a single factual error, but the whole thing. The structure misses the point. The angle is wrong. The argument does not go where it needs to go.
Your instinct: fix it. You already paid for it. You already have text on the screen. Editing feels like progress. Regenerating feels like admitting failure.
That instinct is wrong. The $0.03 is gone either way. The question is whether the next 45 minutes of editing will produce better output than a revised prompt and 8 more seconds of generation.
The Edit/Regenerate Decision: Edit when the structure is sound and the problems are surface-level (voice, specificity, markers). Regenerate when the structure is wrong, the angle is off, or the foundation is built on hallucinated claims. Editing a bad foundation is more expensive than building a new one.
The Decision Framework
Three questions determine whether to edit or regenerate. Answer them in order.
(Right sections, right order,
right argument flow)"} B -->|No| C["REGENERATE
Structure problems cascade.
Fixing them means rewriting."] B -->|Yes| D{"Q2: Is the factual foundation sound?
(Key claims verified,
no hallucinated premises)"} D -->|No| E["REGENERATE
Content built on false claims
cannot be patched."] D -->|Yes| F{"Q3: Is the angle/framing correct?
(Right perspective for
target audience)"} F -->|No| G["REGENERATE
with revised prompt specifying
the correct angle."] F -->|Yes| H["EDIT
Surface problems only.
Voice, specificity, markers."] style C fill:#c47a5a,color:#111 style E fill:#c47a5a,color:#111 style G fill:#c47a5a,color:#111 style H fill:#6b8f71,color:#111
If any of the three questions returns "no," regenerate. If all three return "yes," edit. The logic is simple: structure, facts, and angle are the foundation. Everything else is finish work. You can repaint a wall. You cannot repaint a wall that is leaning.
The Cost Comparison
The real cost is not tokens. It is your time.
| Scenario | Token Cost | Your Time | Total Cost (at $50/hr) | Quality Outcome |
|---|---|---|---|---|
| Edit surface problems (3-5 markers, voice fixes) | $0.00 | 20 min | $16.67 | High. Good foundation + human polish. |
| Edit structural problems | $0.00 | 60+ min | $50.00+ | Variable. Often still feels patched. |
| Regenerate with revised prompt | $0.03-0.10 | 10 min (prompt revision) + 20 min (review) | $25.10 | High if prompt is improved. |
| Regenerate 3 times, pick best | $0.09-0.30 | 10 min (prompt) + 15 min (compare) | $20.97 | Very high. Selection from pool. |
Editing surface problems is the cheapest path. Editing structural problems is the most expensive path. Regeneration sits in the middle for cost but often produces the best outcome because the prompt gets improved in the process.
The Attachment Trap
Writers get attached to text they have read. Even AI text. You read 1,500 words, your brain processed them, and now they feel like "your" words even though they are not. This attachment biases you toward editing when regeneration would be better.
Counter this bias with a rule: make your edit/regenerate decision within the first 90 seconds of reading. If the structure, facts, and angle pass the three-question test in 90 seconds, commit to editing. If any question fails, close the file and revise the prompt. Do not read the whole piece before deciding. The more you read, the more attached you become.
The Prompt Improvement Loop
Every regeneration should improve the prompt. If you are regenerating because the structure was wrong, add explicit structural constraints to the prompt. If the angle was off, specify the angle explicitly. If the facts were hallucinated, add a constraint like "only include claims you can verify" or "do not cite specific studies unless I provide them."
Track your regeneration reasons in your production log. After 20 regenerations, you will see which prompt weaknesses recur. Fix those weaknesses and your edit/regenerate ratio shifts toward editing, which means your prompt engineering is maturing.
The Decision Rule
Write this down and tape it to your monitor:
Regenerate when: structure is wrong, angle is wrong, or the core argument rests on unverified claims.
Edit when: structure is sound, angle is right, and the problems are voice, specificity, AI markers, or minor factual corrections.
Never: spend more than 45 minutes editing a single AI-generated piece. If you hit 45 minutes and you are not done, regenerate. The output was not worth saving.
Further Reading
- AI Editor vs Human Editor (2025): Comparison, Scenarios & Decision Guide, SkyWork
- AI vs Human Editors: Which Is Better in 2025?, Leyline Pro
- The Difference Between Editing Human vs AI Writing, Your AI Copy Editor
- Human vs AI Writing Examples Side-by-Side: Style Differences, Hastewire
Assignment
Collect 5 AI outputs of varying quality. For each, apply the three-question framework and make a snap judgment: edit or regenerate? Then spend exactly 15 minutes attempting to edit each one (set a timer). After 15 minutes, evaluate: was your snap judgment correct? Which pieces improved with editing? Which ones were a waste of 15 minutes? Write a personal decision rule based on your findings.