Home Gallery
Standards Paper GitHub

Bringing Transparency and Accountability to
AI System Prompts

AISPA (Artificial Intelligence System Prompt Assurance) is the first open standard for evaluating system prompts in commercial AI applications.

All system prompts were collected from publicly available GitHub repositories.1

See the results Try your own prompt
Research by leading institutions
Stanford University Stanford HAI Carnegie Mellon University MIT
scroll

What is a system prompt?

A system prompt is a hidden instruction sent alongside every user message. It is invisible to the user but sets rules, tone, and priorities for every reply.

Check the demo below to see how the system prompt affects the model's response.

User messages

Choose a scenario to load its user message, system prompts, and sample reply.

System prompts & response

Response

Audit Results

Company-Level Results

Average problematic and protective instruction annotations per prompt across all audited products.

Full gallery →

Protective Instructions Over Time

Average number of +1 (protective instruction) entries per product

Average Prompt Length

Average system prompt length in characters

Problematic Instruction Percentage

Percentage of products containing at least one problematic instruction

Dimension Coverage Across Products

Percentage of the 88 audited prompts that contain at least one annotation per dimension

User Protection Evolution
Across Representative Series

Number of protective and problematic instruction entries per product version, grouped by organization

All Products

Sorted alphabetically · click a column to re-sort
Product Problematic Risk Protective Ranking

All system prompts were collected from publicly available GitHub repositories.1

Interactive Tool

Try it yourself

Paste any system prompt. We'll score it across all 8 dimensions.

System Prompt