The Protocol

A Practical Workflow to Use AI in Academic Writing Without Hallucinations

Telling researchers "don't use AI" is unrealistic.

Telling them "be careful" is useless.

What's missing is a provable workflow.

Core Principle — Evidence-Binding:
Every AI output must be tied to a real, verifiable source.
Step 01

Retrieval First, Generation Later

Never ask AI to "find references". You retrieve papers yourself.

  • Google Scholar
  • PubMed
  • ArXiv
  • Library databases
Key Principle: AI never invents sources. It only works on materials you provide.
Step 02

Source Injection

Paste abstracts, excerpts, or DOIs directly into the AI prompt.

  • AI is restricted to only these materials
  • No external knowledge is used
  • Every claim must come from your sources
Key Principle: If you didn't provide it, AI can't use it.
Step 03

Evidence-Bound Writing

Ask AI to write based only on provided sources.

  • Every claim must point back to one of them
  • Request inline citations for each statement
  • Reject any content without source attribution
Key Principle: No source, no claim.
Step 04

Citation Mapping

After writing, map every citation back to its source.

  • Each sentence → exact paper → exact paragraph
  • If you cannot map it, delete it
  • No exceptions
Key Principle: If it can't be mapped, it doesn't exist.
Step 05

Manual Verification Pass

Open every cited paper. Confirm the claim truly exists.

  • Read the original text
  • Verify the claim matches what's written
  • Check page numbers and sections
Key Principle: This removes 100% of hallucinated references.
Step 06

AI Usage Log (Compliance Layer)

Record everything for auditability.

  • What AI was used
  • What sources were provided
  • What prompts were used
  • What sections were AI-assisted
Key Principle: Now your AI usage is auditable.

AI is not the problem.

Untraceable AI usage is.

This Protocol makes AI writing provable.

Get the Templates