consistency-terminology — Terminology Consistency

Severity: INFO · Auto-fix: No · Category: 🔄 Consistency

What It Does

Detects when the same concept is referred to by two different names within the same prompt. When both terms from a known synonym pair appear together, the model may treat them as distinct entities — causing duplicated logic or ignored instructions.

One finding is emitted per conflicting pair found.

Built-in Term Pairs

The rule checks 12 built-in synonym pairs. A finding fires when both terms from a pair appear anywhere in the prompt:

Term A	Term B
user	customer
function	method
error	exception
AI	model
LLM	model
query	request
response	reply
output	result
prompt	message
system prompt	instruction
task	goal
agent	assistant

Pair-based, not cluster-based

The rule only compares terms within the same predefined pair. It will not flag "user" vs "client" or "person" vs "customer" — those aren't built-in pairs. Use custom_term_pairs to add your own.

Examples

Mixed terminology

Extract all user records from the database.
Each customer entry must include a name, email, and account status.
Return the user list as JSON.

Findings:

[ INFO ] consistency-terminology (line -)
  Mixed terminology: 'user' and 'customer'. Use one term consistently.

Consistent terminology

Extract all user records from the database.
Each user entry must include a name, email, and account status.
Return the user list as JSON.

Multiple conflicting pairs

Call the LLM to generate a response.
The model will return a reply in under 200 tokens.

Findings:

[ INFO ] consistency-terminology (line -)
  Mixed terminology: 'LLM' and 'model'. Use one term consistently.
[ INFO ] consistency-terminology (line -)
  Mixed terminology: 'response' and 'reply'. Use one term consistently.

Custom Term Groups

Add project-specific synonym groups in .promptlintrc. Each group is a list of terms that should not be mixed — every pair within the group is checked:

rules:
  consistency_terminology:
    enabled: true
    custom_term_pairs:
      - ["config", "configuration", "settings"]    # flags config+settings, config+configuration, etc.
      - ["endpoint", "route", "path"]
      - ["payload", "body", "data"]

False Positives

Intentional distinction — "An AI model differs from a traditional rule-based model in…" uses both AI and model but refers to different things. The rule can't distinguish intent. Disable it or use custom_term_pairs with terms that are genuinely equivalent in your domain.

Technical writing conventions — API documentation often uses request and query with distinct meanings (HTTP request vs. database query). In that context, the pair query/request may flag legitimately different concepts.

Configuration

Enable/disable:

rules:
  consistency_terminology: true   # default: true

Disable:

rules:
  consistency_terminology: false

No Auto-Fix

Resolving a terminology conflict requires knowing which term is canonical — PromptLint reports the ambiguity but leaves the choice to you.