Prompt management

​
Overview

The Prompts tab in Settings allows administrators to customise the system and user prompts used for each AI analysis type. Custom prompts override the built-in defaults for your organisation only — other organisations are not affected.

​
How prompts work

Each analysis type (kind) has two prompt components:
ComponentPurpose
System promptSets the AI’s role, constraints, and output style. Applies to all analyses.
User prompt templateThe specific instruction for this analysis type. Supports template variables.

​
Template variables

Use these placeholders in the user prompt template:
VariableReplaced with
{{context}}The EPD data passed to the analysis
{{topic}}The analysis topic or specific question
{{content}}Additional content (e.g. pasted LCA report text)

​
Editing a prompt

1

Go to Settings → Prompts

Navigate to /settings and click the Prompts tab.
2

Select an analysis kind

Click on the analysis type you want to customise (e.g. Gap Analysis, Conformance, Draft Narrative).
3

Edit the system prompt

Modify the system prompt to change the AI’s persona, constraints, or output format.
4

Edit the user prompt template

Modify the instruction template. Ensure you keep the {{context}} variable so EPD data is included.
5

Save

Click Save prompt. The updated prompt is versioned — the version number increments automatically.

​
Version history

Each save creates a new version. The current active version is shown next to the prompt name. You can:
  • Revert — go back to the previous version
  • Reset to default — discard all customisations and restore the built-in prompt

​
Best practices

Always keep {{context}} in the user prompt template. Without it, the AI will not receive any EPD data and will produce generic, unhelpful output.
The system prompt includes mandatory conservative reasoning instructions (“never invent numeric results”). If you modify it, ensure these constraints are retained for EPD compliance.
  • Keep system prompts concise — focus on role, tone, and output format
  • Test changes on a low-stakes EPD before applying to production declarations
  • Use versioning to roll back if a prompt change produces worse results