A Pre-Send Safety Check for ChatGPT Prompts
Screen prompts for actionable harm, licensed professional decisions, and identifiable or confidential data before sending.
2 min read
There is no need to treat every question as dangerous. Pause before sending content that would enable actionable harm, a final decision that requires a licensed professional, or disclosure of identifiable or confidential data.
Three pre-send categories
| Category | Do not send | Safer transformation |
|---|---|---|
| Harmful or illegal action | Operational instructions for intrusion, fraud, or weapons | Ask about defense, prevention, ethics, and authorized training |
| Licensed final decision | A diagnosis, medication dose, or predicted court result | Organize facts and questions for a qualified professional |
| Personal or confidential data | Credentials, private contracts, identifying records | Remove unnecessary identifiers and use a fictional structure |
OpenAI's Usage Policies restrict malicious cyber activity, fraud, weapons-related uses, and tailored advice requiring a license without appropriate professional involvement. OpenAI Usage Policies
Turn a final decision into consultation preparation
OpenAI's Terms say output may not always be accurate and should not be the sole source of truth or a substitute for professional advice. Users must evaluate accuracy and appropriateness, including human review where appropriate. OpenAI Terms of Use
Instead of asking for a diagnosis, use AI to:
- put symptoms and timing into a clear chronology;
- list medication and medical-history details to bring to an appointment; and
- draft questions for a qualified clinician.
Urgent symptoms require local emergency guidance, not an AI exchange. Legal, tax, insurance, and investment conclusions likewise depend on jurisdiction, contracts, deadlines, and the person involved.
Removing a name may not be enough
An employer, role, rare event, and precise date can identify someone even after a name is removed. Check:
- Is the detail necessary for the answer?
- Do you have permission to provide a third party's information?
- Does it comply with your organization's rules?
- Could exposure affect authentication, a contract, health, or legal rights?
Consumer ChatGPT users can turn off “Improve the model for everyone” in Data Controls. Temporary Chat has different history and model-improvement handling, but information sent through a GPT's external action is governed by the recipient's privacy policy. OpenAI Data Controls FAQ
Settings do not replace data minimization. Do not enter passwords, API keys, identity numbers, or unpublished data merely because a control is enabled.
Takeaway
Before sending, ask whether the prompt enables harm, delegates a licensed final decision, or exposes identifiable or confidential information. If so, transform it into prevention, consultation preparation, or a fictional structure—and omit data that is not needed.
Primary sources checked
Important claims should also link to the relevant source in the article body.
- Data Controls FAQOpenAI · official-help · Checked: 2026-07-26
- Terms of UseOpenAI · provider-primary-source · Checked: 2026-07-26
- Usage policiesOpenAI · provider-primary-source · Checked: 2026-07-26