Important Considerations When Using AI (4/4)
1. Set Up Project-Based Instructions and Custom System Prompts
Don't just use AI tools with default settings. Configure them properly for proposal work:
Create a master system prompt for your proposal work:
You are an expert assistant helping with clinical research RFP responses. Before providing any answer:
1. Ask clarifying questions until you're 100% certain you understand the context
2. Never fabricate informationβif you don't have enough information, explicitly say so
3. When making recommendations, provide your reasoning
4. If I provide conflicting information, point out the inconsistency and ask for clarification
5. Use British English spelling and grammar conventions [or specify American English, or other preferences]
6. When citing information from documents I've uploaded, always provide specific references
7. If you're uncertain about any aspect of clinical research or CRO operations, ask me rather than making assumptions
My priorities in order are: accuracy, compliance with RFP requirements, differentiation from competitors, persuasive narrative flow.
Before answering any question, confirm you understand what I'm asking by briefly restating it in your own words.
Why this matters: This prompt structure actively prevents the most common AI pitfalls:
Hallucination: By instructing the AI to ask questions rather than guess, you reduce false information
Assumption errors: By requiring clarification of inconsistencies, you catch potential mistakes early
Style consistency: By specifying language preferences upfront, all generated content maintains the same conventions
Where to implement this:
Claude Projects: Create a project specifically for RFP work and add this as your custom instructions
ChatGPT: Use Custom Instructions in settings
Gemini: Add to the conversation at the start of each proposal
Perplexity: Include as preamble to your first query in a thread
2. Maintain Confidentiality and Data Security
Never paste:
Actual sponsor names (use "Sponsor X" or generic descriptors)
Proprietary methodologies or trade secrets
Specific pricing details that could constitute competitive intelligence
Patient data or any PHI (though this shouldn't be in proposals anyway)
Non-public information about your CRO's pipeline or financials
Do create sanitised versions:
Replace sponsor names: "large biotech sponsor" or "Sponsor A"
Generalize therapeutic areas if needed: "rare genetic disorder" instead of specific diagnosis
Round budget numbers to ranges: "$2-3M" instead of "$2,347,392"
Use role titles instead of names: "Senior Clinical Operations Lead" rather than actual person's name
Exception: Tools like NotebookLM that process data locally or in your private workspace are safer for confidential documents, but still verify your organization's data policies before uploading sensitive materials.
3. Verify Everything AI Generates
AI can be persuasively wrong. Always verify:
Citations and references: If AI mentions a regulation, guideline, or industry standard, verify it exists and says what the AI claims
Numerical calculations: Double-check any budget calculations, timeline math, or statistical claims
Technical accuracy: Have SMEs review AI-generated technical content in their domain Logical consistency: Read the full output to ensure different sections don't contradict each other
Practical tip: Use the "fact-checking buddy system". Have AI generate content, then paste that content back into a different AI tool (e.g., generate in Claude, verify in Perplexity) and ask: "Review this content for factual accuracy. Flag any claims that need verification."
4. Maintain Your Authentic Voice and Strategic Judgment
Over-reliance on AI can make your proposal sound generic. AI should accelerate your work, not replace your expertise.
Use AI for:
Initial drafts that you'll substantially revise
Structural suggestions and organization
Identifying gaps in your logic or content
Generating multiple options for you to choose among
Handling routine formatting and consistency tasks
Don't use AI for:
Final strategic decisions (positioning, differentiation strategy, pricing approach)
Content that requires deep institutional knowledge
Nuanced judgments about client relationships or politics
Final sign-offβalways have human experts review
5. Create Feedback Loops for Continuous Improvement
After each proposal, document what worked and what didn't with your AI tools:
Which prompts generated the most useful outputs?
Where did AI lead you astray?
What custom instructions would have prevented issues?
Which tools were most valuable for which tasks?
Add these learnings to your Notebook LM "lessons learned" notebook so future proposals benefit from your accumulated AI expertise.
