Skip to content

Context Matters

Key Takeaways

  • Context is everything: All information you provide helps LLMs understand your request (role, purpose, audience, requirements)
  • Be specific, not vague: Clear context gets relevant responses; vague prompts get generic ones
  • Refine iteratively: Start basic, review output, add details, repeat until satisfied
  • Never share sensitive data: No personal records, unpublished research, or confidential information

Why Context Matters for LLMs

Context is everything you provide to an LLM to help it understand your request and generate appropriate responses during your chatbot session.

Without proper context, LLMs often produce generic, inaccurate, or inappropriate outputs.

The Core Problem

LLMs have no memory between separate sessions. They have no inherent knowledge about:

  • Who you are
  • Your organisation's policies
  • The purpose of your request
  • Your audience
  • Technical constraints

Large Language Models work by navigating through vast embedding spaces—multidimensional representations of knowledge and concepts.

Vague or poorly defined context → model explores irrelevant areas → generic/off-target responses
Well-crafted context → precise navigation instructions → relevant, targeted outputs

Poor vs. Better Context

❌ Poor Context Example

Write me a report about productivity

Problems: - No audience specified - No length guidance - No context about purpose - No domain focus

Result: Generic business report that could apply to any industry.

✅ Better Context Example

I am a senior R&D manager at AmaDema, a nanotechnology company 
specialising in non-oxide ceramics. I need to write a 2-page 
executive summary about remote work productivity for our R&D 
department heads, focusing on evidence-based strategies specific 
to laboratory-based research teams, including practical 
implementation steps for hybrid lab schedules.

Improvements: - ✓ Role identified (senior R&D manager) - ✓ Organisation context (nanotechnology, non-oxide ceramics) - ✓ Length specified (2 pages) - ✓ Audience defined (R&D department heads) - ✓ Domain focus (laboratory-based research) - ✓ Output requirements (evidence-based, practical steps)

Result: Targeted, relevant content suitable for your specific needs.

Types of Context

1. Explicit Context

Information you directly provide to the LLM:

  • Your role and organisation
  • The purpose of the task
  • Target audience
  • Output format (length, structure, tone)
  • Specific constraints

2. Implicit Context

Assumptions the LLM makes based on your prompt:

  • Cultural assumptions (Western vs. global perspective)
  • Educational level expectations
  • Language formality
  • Technical depth

Making Implicit Context Explicit

Instead of assuming the LLM will understand your context, state it clearly.

Example Transformation

Vague (relies on implicit context): 

Help me write a proposal on sustainable materials

Explicit (states all relevant context): 

I'm a researcher at AmaDema, a nanomaterial company. Help me 
write a 3-page research funding proposal targeting EPSRC for a 
project on sustainable non-oxide ceramic materials in 
engineering applications. The audience is technical reviewers 
with expertise in materials science. Include: research gap, 
methodology, expected outcomes, and alignment with UK net-zero 
goals.

Never Share Sensitive Data

Red List – Never Share These

🚫 Personal data: Student records, staff information, health data
🚫 Confidential research: Unpublished findings, grant applications under review
🚫 Commercial sensitive: Partnership agreements, financial information, exact synthesis ratios
🚫 Legal privileged: Legal advice, disciplinary proceedings
🚫 Security sensitive: Passwords, system configurations, access credentials
🚫 Intellectual property: Unpublished molecular structures, novel formulations, proprietary processes

Why This Matters

While LLMs have no memory between separate chat sessions and don't retain information from previous conversations, the data you share within each individual session may still be:

  • Stored by the service provider
  • Used for training future models (depending on terms)
  • Subject to different legal jurisdictions
  • Potentially accessed by third parties

Always follow AmaDema's data protection policies when sharing any information.

Working Within Context Limits

LLMs have context windows—limits on how much text they can process at once or over different iterations.

Model Approximate Context Window
GPT-4 ~8,000-32,000 tokens
Claude 3 ~200,000 tokens
Llama 3.3 8B ~128,000 tokens

Rule of thumb: 1 token ≈ 4 characters

Strategies to Overcome Limits

✅ Summarise lengthy background information and prioritise the most important context
✅ Break complex tasks into smaller parts
✅ Use previous outputs as context for follow-up requests
✅ Extract only relevant sections from long documents

Iterative Context Building

You won't always get what you want on the first attempt. A useful strategy is to start with basic context and refine:

The Refinement Loop

  1. Initial request: Provide core context
  2. Review output: Identify what's missing or wrong
  3. Refine context: Add specific details or corrections
  4. Iterate: Repeat until satisfactory

Iterative Refinement in Action

Round 1: 

Summarise the key findings from this polymer synthesis paper.
Output: Generic summary of paper structure

Round 2: 

Focus specifically on the tensile strength results for PLA/PCL 
blends at different ratios. I need this for a technical review.
Output: Focused on tensile data, but missing statistical significance

Round 3: 

Include the statistical significance (p-values) for each blend 
ratio and highlight which formulations showed >20% improvement 
over pure PLA.
Output: Precise, relevant summary with quantitative comparisons

Exercise: Context Writing

Challenge

Transform this vague prompt into an effective, contextualised request:

Vague prompt: 

Write about nanotechnology applications

Your task:

  1. Add explicit context (role, audience, purpose)
  2. Define output requirements (length, format, depth)
  3. Specify constraints (focus areas, excluded topics)
  4. Test both versions in the sandbox
  5. Compare the quality difference

What to include:

  • Your role at AmaDema
  • The intended audience
  • Specific application area (e.g., biomedical, aerospace, energy)
  • Length and format
  • Any required sections

Next: Prompting Frameworks: Learn structured approaches to prompt engineering →