Prompt Audit

# Prompt audit Published date: 2026 02 11Last reviewed date: 2026 02 11. Reviewed for accuracy and scope alignment.Layer: 4 Primary question: What do we analyse in a prompt audit, and what do you get back? A prompt audit tests inclusion patterns across prompt types, maps citations and mentions where they appear, and identifies structural gaps that can reduce confidence. You get a simple output: what shows, what does not show, why it likely behaves that way, and what to fix first. It does not guarantee inclusion or citations. ## What we analyse - Comparative prompts that trigger best and top logic - Navigational prompts that trigger list and retrieval behaviour - Question prompts that trigger explanation and validation - Problem based prompts that trigger matching and suitability ## How we test prompts We run a structured set of prompt variants so we can see what is stable versus what is prompt specific. We record the output type, the entities named, and whether citations appear. Where citations do not appear, we still record mention behaviour and repeated reasoning criteria. ## Citation mapping and mention analysis Citation mapping focuses on what sources are cited and when. Mention analysis focuses on whether an entity appears without a cite, and what context it appears in. The goal is not to chase one interface. It is to reduce ambiguity and increase stability across the surfaces that matter. ## Confidence scoring We score the clarity conditions that often affect inclusion: naming consistency, category fit, service boundaries, and reinforcement patterns. This is a practical score used to order fixes. It is not a promise and it is not a ranking score. ## Competitive comparison Where it makes sense, we compare prompt behaviour against a small set of competitors or alternatives, using the same prompt set. This helps explain whether your issue is unique, or whether the system is cautious across the whole category. ## What you receive | Audit component |