GPT-5 Prompt Optimizer: test the revision rather than trust its label

Save the original request and its intended result

The 4 October 2025 video tests an official GPT-5 Prompt Optimizer with coding tasks, written summaries and Luke’s YouTube strategy. The description reports that some answers reverse from yes to no while others barely improve. A changed answer is therefore a result to investigate, not proof that the optimised version is more correct.

1. Follow the explanation at 0:37 and identify the original task before reviewing any expanded prompt.
2. Keep the raw prompt and the optimiser’s revision together so added requirements or assumptions remain visible.
3. Write the evidence or behaviour that would make the answer useful before deciding which version you prefer.

Inspect coding and writing by different criteria

The coding comparison begins at 1:54; the writing test at 3:32 uses a digital-IDs example. A coding response needs a working result, while a written explanation needs accurate claims and appropriate framing. Extra structure can improve readability without fixing the underlying facts, and an elaborate instruction can change the question being answered.

1. For code, run the intended task and inspect any change against the original requirement rather than judge the response length.
2. For the writing example, compare claims, qualifications and the requested perspective between the two outputs.
3. If a conclusion reverses, identify what changed in the prompt and check whether the evidence actually supports the new position.

Use the strategy comparison to inspect assumptions

The YouTube strategy test and verdict start at 5:25. Advice about a channel depends on its actual audience, goals and constraints. An optimised prompt should preserve those inputs rather than produce a more confident plan for a different situation. The source does not provide the complete raw or optimised prompts or a direct optimiser setup link.

1. Compare the advice with the same channel brief and record any invented audience, objective or constraint.
2. Keep useful clarifications while rejecting additions that silently redefine the job.
3. Repeat one bounded comparison before making optimisation a default step, recording both genuine improvements and cases where it adds little.

The takeaway

An optimiser is useful when it preserves the task and improves a checked result, not simply when it produces a longer prompt or a different answer.