Your first ChatGPT prompt for lab and office work

A quick primer before the walkthrough: A concrete first prompt structure you can adapt for routine workplace drafting tasks.

In plain English: Most beginner prompts fail because they are too short and too vague. This tutorial gives you a 4-line template that fixes both.

Key terms:

• Prompt — what you type into ChatGPT.
• Role — telling the model who to act as ("You are a process development scientist drafting a non-confidential briefing").
• Constraints — telling the model what not to do ("Do not invent citations").

Comfort level: No coding required. Standard chat interfaces are sufficient, provided usage aligns with your employer's AI policy.

Data privacy and policy check

Before pasting any workplace text into an AI tool:

• Redact sensitive information: Always remove proprietary, confidential, trade secret, patient (PHI), or personally identifiable information (PII). Ensuring compliance with internal data classification standards is your responsibility under your organization's policies.
• Verify vendor terms: Check your organization's current agreement and account settings with the vendor, because enterprise features, data retention policies, and model-training terms vary and change over time.

The template

Copy this template into ChatGPT and fill in the bracketed fields:

Role: You are a [WHO — e.g., senior process development scientist].
Task: [WHAT — e.g., draft an illustrative 1-page summary of the provided text].
Audience: [WHO WILL READ IT — e.g., manufacturing site leadership, non-technical].
Rules:
- Only use facts from the material I provide.
- If something is missing or unclear, list it under "Open questions" instead of guessing.
- Keep it under 400 words.

[PASTE REDACTED / NON-CONFIDENTIAL SOURCE TEXT HERE]

Every part matters, but remember that all generated text must undergo qualified human review before use in any operational or governed context.

Why each line matters

• Role — tells the model what perspective to adopt. "Senior PD scientist" produces different terminology than "science writer." (Note: Role definitions are illustrative prompts and do not replace professional credentials or SOP-defined responsibilities.)
• Task — one clear verb (draft, summarize, extract, critique). Vague tasks produce vague answers.
• Audience — controls tone and depth. "For executive leadership" requires a different structure than "for bench scientists."
• Rules — helps limit unsubstantiated claims. Instructing the tool to "Only use facts from the material I provide" is an effective way to keep outputs grounded in your provided text.

Four small tweaks that make a big difference

1. Give it examples. "Here are two summaries that match our team's preferred structure — follow this format." One clear example is often more effective than lengthy written instructions.
2. Ask for a format. "Return as a Markdown table with columns: Identified Observation, Impact Level, Suggested Focus." Structured output simplifies review.
3. Ask for the reasoning. "Explain why you flagged each item based solely on the text." Requesting explicit reasoning makes the output easier for a subject-matter expert to fact-check.
4. Iterate, don't restart. If the initial response is partially helpful but needs refining, reply with specific edits (e.g., "Rewrite section 2 as bullet points") rather than starting a new conversation.

A worked example (Quality Assurance)

Regulatory and QMS notice: AI outputs may support, but do not satisfy, your validated-records, Quality Management System (QMS), or GxP requirements. AI-generated risk ideas or CAPA suggestions are strictly preliminary drafts for discussion and must never replace official deviation investigations, formal risk assessments, or approved quality procedures executed by qualified personnel.

Prompt:

Role: You are a QA specialist drafting preliminary review notes.
Task: Identify up to 3 potential operational risks mentioned in the text below and suggest 1 illustrative CAPA idea per risk for expert evaluation.
Audience: QA manager, technical background.
Rules:
- Only use facts explicitly stated in the excerpt below.
- If a root cause is not stated, explicitly state "Not provided in source text" — do not infer or speculate.
- Return as a numbered list.

[PASTE REDACTED / NON-CONFIDENTIAL DEVIATION EXCERPT HERE]

Using specific constraints yields a far more structured drafting starting point than a generic prompt like "read this report and tell me what's wrong."

Common beginner mistakes

• Pasting unvetted data. Uploading unredacted proprietary or regulated data risks violating internal data governance rules. Always check your organization's current vendor settings and policy guidelines.
• Asking without providing source text. If you do not supply context, the model relies on general training data, which is unsuitable for specific operational or scientific evaluation.
• Conflating drafting with execution. AI models generate text patterns; they do not perform validated scientific or regulatory work. Treat output as a rough first draft created by a junior assistant that requires thorough review by a qualified professional against your site's specific SOPs.

What to try next

• Prompting ChatGPT for high-quality scientific paper summaries
• Safer AI use at work: what to share and what to keep private

If something goes wrong

• The answer is generic. You probably left out the role, the context, and the format. Add all three and try again.
• It is too long. Ask for a specific length and shape — "five bullets, one line each".
• It confidently states something wrong. That is expected behaviour, not a malfunction. Verify anything that would change a decision.
• It forgets what you told it earlier. Long conversations drift; start a fresh chat and paste the essentials back in.
• You are unsure whether you may paste this. If it is unpublished, identifiable, or contractual, do not — see Safer AI use at work: what to share and what to keep private.