Production-ready in weeksTalk to a Quality Expert

Pharma AI glossary

Prompt / prompt engineering

The instruction given to a generative model, and the craft of writing it well.

Definition

The instruction given to a generative model, and the craft of writing it well. The reliable skeleton: **context** (who this is for, what it concerns), **task** (exactly what to produce), **format** (length, structure), **guardrails** ("if information is missing, say so; don't invent it"). Prompts are not incantations; they're specifications, and the discipline transfers directly from writing good work orders. In regulated settings, standardized prompts for recurring tasks (with the guardrail clause built in) are an underused, nearly-free control.

Instructions as a control

Prompt templates are one layer of the agent harness, alongside tool allowlists, output schemas, and review gates. In enterprise pharma, ad hoc user prompts are not a control strategy; approved templates with version control, logged runs, and workflow-defined sign-off are.

Literacy for teams

Quality staff should understand what prompts can bound (format, tone, scope) and what they cannot (factual guarantee, QMS write authority). GMP workflows using agents should spell out where the tool stops and where human approval is required, whether for investigations, batch documentation, CAPA, or change control, with those steps in SOPs and audit logs.

Frequently asked questions

Should SOPs define AI-assisted GMP steps?

Yes. SOPs or work instructions should name the approved agent or application, allowed actions (read vs draft vs write), required reviewers, and what the audit trail must capture for each workflow (investigations, CAPA, batch records, change control, training). That is governance and workflow design, not prompt trivia.

Is prompt engineering a technical or quality skill?

Both. Quality defines acceptable outputs and review; IT/data science hardens templates, logging, and testing. Literacy bridges the two.

Benchmark your AI literacy

Free 20-question assessment, eight modules, and instant scoring, built for pharma quality and manufacturing teams.

Open the Literacy Center

Browse all terms

See the full 71-term glossary index or search interactively in the Literacy Center.

Back to glossary hub →

Educational content only, not legal or regulatory advice. Regulatory guidance cited here includes drafts (FDA AI credibility guidance; EU GMP Annex 22) as of August 2026; verify against final texts before relying on them in submissions. Company-reported figures (Merck CSR timings, Sanofi results) are labeled where used. MIT's ~95% pilot figure carries its own caveat (~150 interviews, contested definitions, not peer-reviewed). Re-check sources on module finalization.