When revising an AI creator search, restate the requirements that still matter along with the change. A short follow-up such as “cheaper” leaves too much implicit: the price period, topic, offer type, and treatment of unknown information may all become unclear.

Treat the revised message as a complete request that you could read on its own.

Keep a small requirements card

Before the first search, write four lines:

  1. The public interest or topic you want.
  2. Any essential offer or budget condition.
  3. Preferences you can relax.
  4. Details that must remain unknown unless publicly supported.

This card is for your own use. It is not a request to collect private characteristics or investigate someone outside their public creator identity.

OnlyDetect's methodology limits discovery to public profile information.

Change one condition and preserve the rest

A fictional initial request might be:

Find adult creators whose public profiles discuss photography. Prefer a paid subscription with a recorded price of $15 or less. Show the source and leave unsupported details unknown.

A weak revision is “make it cheaper.” A clearer revision is:

Keep photography as the topic. Change the recorded subscription-price preference to $10 or less. Keep the same public-source requirement and label unknown information.

The second request explains the change and the conditions that remain. It does not rely on an assumption that every interface carries forward every earlier detail.

Check the interpreted request

In OnlyDetect AI discovery, the reviewed interface displays interpreted preference chips. Use visible summaries as a check: do the budget, topic, and offer type still reflect what you asked?

A chip is a summary of an interpreted request, not proof that every result satisfies it. If a condition is absent or wrong, rewrite the whole request rather than layering several vague corrections.

This guide does not claim that a follow-up is persisted across every model or session. Product behavior and available context can differ.

Keep results and revisions paired

If you compare two runs, save the complete request used for each and the date. Otherwise, you may mistake a changed input for inconsistent results.

For example, changing both the topic and budget at once makes it harder to understand which change affected the shortlist. Revise one major condition when that distinction matters.

Stop when the evidence is missing

Repeating a request more forcefully cannot create a reliable current price or a public disclosure that the source does not contain. OpenAI's accuracy guidance warns that confident answers can still be incorrect.

If an essential condition remains unsupported, keep it unresolved. Use the unknown-fields case study to decide when a candidate remains a lead and when it is sufficiently supported for your next step.