AI Dictionary › Prompting

Context Stuffing

Context stuffing means cramming into the prompt all potentially relevant material, whole documents, many instructions, numerous examples, filling the context window in the hope the model will find what it needs on its own. Sometimes it works, but it often hurts results: the model gets distracted, loses the key instructions, or suffers the "lost in the middle" effect, ignoring information at the centre of the context. Example: pasting 40 pages of a manual for a single question instead of extracting and passing only the 2 relevant paragraphs.

Definition

Avoid it as a default strategy: prefer selected, relevant context (via retrieval or compression) over indiscriminate filling. Stuffing makes sense only when you truly cannot know in advance which part matters and high token cost is acceptable; even then, order the information well.

How to use it

The anti-pattern of pasting everything you have into the prompt: whole documents, full histories, every detail. More context is not more accuracy: past a point, the model loses the middle of long inputs and costs grow linearly.

A concrete example

Instead of pasting a 40-page policy to answer one question, extract the relevant section first: "Based ONLY on this excerpt (below), answer: does the policy cover remote contractors? Quote the sentence that decides it."

When to use it (and what to watch)

Recognize it in yourself when the prompt is mostly paste and the question is one line. The fix: select before you send: the relevant excerpt, a summary of the history, the three fields that matter. Precision of context beats volume of context.

More prompting techniques

Multi-turn prompting · JSON mode · Prompt scaffolding · Contrastive prompting · Retry with backoff · Self-ask · Chain-of-thought · Role prompting · Prompt injection · Temperature · System prompt

Related terms

• Lost in the Middle
• Context Window
• Prompt Compression
• Grounding
• Context (in prompting)

More in Prompting

• Prompt
• Prompting
• Role Assignment
• Zero-shot
• Few-shot
• Chain of Thought
• Prompt Engineering
• Output Format

Put it into practice

• Real scenario: apply this concept with an instant AI verdict.
• Daily gym: short drills to train dictionary concepts.
• Playoff game: challenge other players, elimination-style.

📱 Download the Android app (beta) iOS coming soon

Say what you mean. Get what you need.

Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.