Using LLMs for Academic Writing

A visual summary of effective strategies and common pitfalls based on researcher experiences.

The Spectrum of Prompting Effectiveness

Least Effective

Single-Shot Prompting

Providing minimal context and expecting a finished product in one go.

Flawed Drafts Generic Content Ineffective Unbalanced

Most Effective

Human-in-the-Loop Collaboration

Treating the LLM as a partner in an iterative, guided process.

High-Quality Output Nuanced & Grounded Iterative Context-Rich

Key Strategies for Success

1. Iterative & Interactive Refinement

Engage in a multi-turn conversation. Start with a broad request and progressively narrow it down with specific feedback.

2. Structured Prompting

  • Outline-First: Have the LLM create a "funnel structure" outline before writing prose.
  • Socratic/Q&A Method: Ask the LLM to pose clarifying questions to focus the core message.

3. Advanced Strategies

  • Creative Personas: Use prompts like "act as a research storyteller" for narrative flow.
  • 'Superprompt' Design: Create a detailed, rule-based prompt with a specific persona and constraints.

Common Pitfalls to Avoid

Inappropriate Tone

Overly verbose, generic, or containing "AI slop".

Generic Content

Lacks specificity, especially when discussing prior work.

Factual Hallucinations

Inventing claims when not grounded by provided text.

'Number Dumping'

Including data without necessary context or interpretation.