Want Better Results from ChatGPT (or Any LLM)? Try Contextual Prompts
- Paul Rex
- Sep 5
- 3 min read
Most of us have had this experience: you ask ChatGPT (or any large language model) a question, and the answer comes back flat. You think to yourself, “Really? That’s it?”
Instead of moving forward with useful insights, you spend more time re-asking the question, tweaking your words, or trying again from a different angle. Frustrating, right?
When I first started experimenting with prompts, the results improved. But I’ve since learned there’s an even better way: Contextual Prompts.
Why Context Matters
The results of contextual prompting are striking: sharper, more accurate, and highly usable answers. And it makes sense.
Think about onboarding a new hire. If you gave them a task with no background, would you expect a great result? Of course not. You’d provide context, guidance, and examples. AI works the same way.
Yes, it takes more effort upfront. But like any skill, you get faster with practice — and the quality of the output makes it worth it.
From Stanford: Contextual Engineering
This approach has been articulated well by Jeremy Utley, Adjunct Professor at Stanford and co-author of Ideaflow. He calls it Contextual Engineering: treating AI like a teammate, not just a tool. Instead of short, search-like prompts, Utley recommends giving detailed background (brand, objectives, audience, tone, constraints). In his words:
“I would recommend never prompting a model without at least 400 words of context.”
Utley’s research shows that when you treat AI like a new colleague , giving it direction, assigning roles, asking it to explain its reasoning , the quality of the results improves dramatically.
Utley’s Key Principles
Provide Rich Context (≥400 words)
Most people use prompts like a Google search: 5–10 words.
Utley recommends giving detailed background (brand, objectives, audience, constraints).
“I would recommend never prompting a model without at least 400 words of context.” – Capgemini
Treat AI as a Teammate, Not a Tool
Anthropomorphizing AI improves collaboration and output quality.
“If you want to work with an LLM, treat it like you would treat a new team member.” – Capgemini
“Most surprising discovery… LLMs respond to human-like treatment in remarkably human-like ways.” – Jeremy Utley
Use Chain of Thought & Clarifying Questions
Encourage step-by-step reasoning: “Walk me through your thought process.”
Instruct the AI to ask clarifying questions before answering — shifting from directive to collaborative dialogue.
Set Explicit Roles & Personas
Define the perspective: e.g., “You are a McKinsey partner in medtech commercialization.”
Personas anchor the model’s framing and style.
Feedback Loops
Share edits and instruct the AI to learn from them.
“Here’s how I modified your output. What can you learn from these changes?”
Cross-Model Collaboration
Utley often dictates to one model (ChatGPT), then feeds it into another (Claude, Gemini, Perplexity) for critique and iteration.
This multi-model interplay improves rigor and reduces blind spots.
Prompt Engineering vs. Contextual Engineering
Executive Actions & Recommendations
Pilot a Contextual Engineering Sprint
Pick a high-stakes deliverable (e.g., payer-access slide deck).
Draft a 400–600-word brief (objectives, audience, constraints).
Have one model generate, another critique, and iterate.
Embed Persona-Driven Prompts in Workflows
Standardize prompts: “You are a regulatory strategist with 15 years in oncology.”
Store reusable prompt templates in a knowledge hub for your team.
Mandate Chain-of-Thought
Require models to explain their reasoning, not just give answers.
Always ask: “What assumptions are you making?”
Establish Feedback Loops
After revising AI output, feed changes back and ask: “How would you adapt next time?”
Build organizational memory with iterative refinement.
Train Teams
Run workshops showing side-by-side results of short prompts vs. contextual prompts.
Track how quality improves when context is added.
Measure Impact
KPIs: time saved, revision rates, peer-review quality scores, adoption across teams.
Use these as performance indicators for AI-enabled workflows.
The Payoff
By applying Contextual Engineering, you transform AI from a basic Q&A machine into a collaborative partner. The more context you provide, the more decision-ready the outputs become.
For business leaders, that means sharper decks, clearer strategy briefs, stronger regulatory submissions, and faster execution.
Utley’s framework reinforces what many of us have experienced firsthand: AI is at its best when we give it the kind of direction we’d give a junior colleague. The upfront effort pays back in speed, quality, and confidence in the results.
Final Thought
Treat AI like an eager assistant you’re coaching. Provide it with context, give it a role, ask for reasoning, and close the loop with feedback. The extra work up front turns into sharper, faster, more actionable insights.
Or, as Jeremy Utley puts it, think of this as moving beyond “prompt engineering” into contextual engineering ,
where AI is not just a tool, but a true teammate.
Comments