3 Generative AI Prompts To Strengthen Your SOPs Without Sacrificing Compliance
By Valerie Przekop, MS, fractional head of operations, PDC Pharma Strategy
For the past five years, the most common FDA observation across biologic, device, and drug programs was related to procedures — they either didn’t exist, were written poorly, and/or weren’t being followed properly [1]. This multi-program trend highlights a widespread challenge: Many companies continue to struggle with effectively developing, refining, and executing standard operating procedures (SOPs).
As AI continues to weave its way into everything we do, it will inevitably reshape the industry standard approach to documentation and compliance. Rather than resist change, let’s embrace AI as a tool to tackle one of the FDA’s most persistent issues: poorly developed, incomplete, or unenforced procedures. By leveraging AI, companies can enhance SOP development, clarity, and training, ultimately improving their compliance program with minimal resources. The question is: How?
The answer begins with understanding the type of AI being used and its limitations.
Generative AI is a subset of AI that includes models such as ChatGPT, DeepSeek, and Claude. When prompted, these models create new content by drawing from patterns in existing data. Crafting targeted prompts is critical to optimizing a model’s output.
With that said, generative AI isn’t perfect. Understanding its limitations is as important as knowing how to craft the input. To evaluate AI-generated content effectively, consider two key questions:
- Does the response make sense?
- Even polished responses may fall short without proper context. Ensure the output is relevant and applicable to the given situation. Improve the relevance of responses by including essential background details in each prompt.
- Is the response factually correct?
- Generative AI generates content using patterns learned from training data. When the training data is low quality, the output may follow suit. While generative AI’s ability to discriminate between true and false data has and will continue to improve over time, it’s unlikely that any model will ever be perfect 100% of the time. Further, the output may sound legitimate but be outdated, biased, or factually incorrect. To reduce the risk of factual inaccuracies, ask the model to cite its sources in its response.
Now that the basics of generative AI have been established, the next step is understanding how to apply it to SOP development in a practical, low-risk manner.
3 Low-Risk Generative AI Prompts For SOP Development And Compliance
The following three prompts will support SOP development, refinement, and training — areas where AI can add immediate value with minimal downside.
Prompt 1: SOP Ideation
Prompt: “Generate a list of common SOPs for a [program type] company.”
Expected results: A list of SOP titles commonly found in the specified program type (e.g., biologic, device, drug).
Why this works:
- Provides a timesaving starting point for SOP brainstorming and planning.
- Ensures core SOPs aren’t overlooked.
- Provides industry-standard SOP title terminology.
How to improve it further:
- Provide more background information, such as:
- company type (e.g., sponsor, contract research organization (CRO), pharmacovigilance vendor),
- therapeutic area (e.g., oncology, immunology, cardiology),
- indication (e.g., influenza, diabetes, diarrhea),
- development phase (e.g., Phase 1, Phase 2, Phase 3, post-market),
- regulatory region (e.g., FDA, EMA, MHRA).
- State the scope of the SOPs (e.g., companywide, function-specific, in-house, outsourced).
- Request a list prioritized by regulatory importance.
Improved prompt: “Generate a list of common SOPs for a sponsor company conducting Phase 1 non-small cell lung cancer drug trials in the U.S. The list should include both in-house and outsourced activities and be prioritized by regulatory importance.”
Prompt 2: SOP Readability
Prompt: “Rewrite this SOP for better clarity, consistency, and readability without changing its meaning: [paste SOP text].”
Expected results: An enhanced version of the submitted SOP that maintains the original meaning, including the process, responsibilities, and intent, while improving language and structure to improve staff comprehension, adoption, and retention.
Why this works:
- Ensures consistency of terminology, roles, and responsibilities.
- Refines grammar, style, and logic.
- Removes duplications.
How to improve it further:
- Ask for the identification of conflicting information.
- Request an alignment of the SOP with regulatory expectations.
- Upload a style guide for comparison.
Improved prompt: “Rewrite this SOP for improved clarity, consistency, and readability without changing its meaning. Identify any conflicting information, align the content with FDA expectations, and refer to the attached style guide for formatting and tone: [paste SOP text].”
Prompt 3: SOP Training
Prompt: “Create a quick reference summary for this SOP: [paste text].”
Expected results: A summarized version of the submitted SOP that highlights key points from each section, serving as a clear and accessible quick reference guide for staff.
Why this works:
- Simplifies lengthy SOPs for practical use.
- Improves retention by making SOP information more digestible.
- Provides a solid base to build more extensive training.
How to improve it further:
- Prompt the model to “pretend it’s a professional trainer.”
- Further specify the form of the output (e.g., list, deck outline, infographic text).
- Ask for a section covering common mistakes and/or frequently asked questions.
Improved prompt: “Pretend you’re a professional GCP trainer. Create a quick reference summary for this SOP in the form of a checklist or infographic text. Highlight key steps, include a section on common mistakes, and add a brief FAQ to support staff understanding: [paste SOP text].”
Conclusion
As the renowned computer scientist and AI pioneer Fei-Fei Li said, “Artificial intelligence is not a substitute for human intelligence; it is a tool to amplify human creativity and ingenuity.”
Further, this article is not meant to encourage readers to blindly copy and paste the output of generative AI. Rather, it’s intended to promote the thoughtful use of generative AI as a low-cost, low-risk tool to support SOP development and training — all critical components of compliance.
Collaborate with the model to iteratively improve the quality and relevance of AI-generated content. Build a dialogue that starts broad and homes in on the desired output, or simply ask, “What information do you need from me to optimize your output?”
With that said, due diligence is essential, even when results seem legitimate. Always review content carefully for factual accuracy, contextual relevance, and alignment with regulatory and operational expectations. Generative AI is a powerful tool, but it requires thoughtful human oversight throughout its implementation.
Editor’s note: This topic will be explored in greater detail during the 2025 Society of Quality Assurance (SQA) 41st Annual Meeting & Quality College (Session V-3), where Valerie Przekop will be presenting “Leveraging Artificial Intelligence to Streamline Standard Operating Procedure Development." The presentation will provide an overview of the process used to develop a custom AI model trained on high-quality SOPs, industry regulations, and best-practice writing standards.
About The Author:
Valerie Przekop is a quality systems and data science professional with a strong background in operations, compliance, and AI-driven process improvement. She currently serves as fractional head of operations at PDC Pharma Strategy, where she provides GxP quality assurance and compliance consulting services. Valerie works full-time as a portfolio manager at Elevance Health, where she leads the Carelon Digital Payment Integrity Center of Excellence. Her experience includes supporting successful FDA inspections, developing foundational quality systems, and leveraging AI to streamline SOP development. She has worked with organizations such as Boehringer Ingelheim, ILiAD Biotechnologies, and Lucille Packard Children’s Hospital. Valerie holds a master's degree in biomedical informatics from the Stanford University School of Medicine and dual bachelor's degrees in management science & engineering and psychology from Stanford University. She is Lean Six Sigma Yellow Belt Certified and has been published in Life Science Leader magazine.