Evidence6 min readUpdated April 12, 2026

Prompts

Understand how system-generated prompts are organized, evaluated, and scored into prompt evidence rows.

01

Overview

The platform uses a curated prompt library generated from your brand and category. You cannot edit prompts yet — this keeps scores comparable across accounts and over time. Each prompt is tagged with a category so you can slice evidence by intent.

02

Prompt categories

Prompts are grouped into five functional buckets. The mix is tuned per category at generation time so every run has balanced coverage.

Comparison
"X vs Y" and shortlist-style queries.
Recommendation
"Best tool for…" discovery queries.
Definition
"What is…" knowledge queries.
Use case
Job-to-be-done framing.
Troubleshooting
Post-purchase and support framing.
03

Prompt evidence table

Every row in the evidence table is a single prompt × model pair. Each row records the mention type, position, and the raw excerpt used by the parser. Click the excerpt to see the full response captured during the run.

04

Mention type definitions

The parser classifies every brand reference into one of four mention types. The classification is deterministic and runs on the captured response text only — no live LLM calls at scoring time.

Direct
The brand name appears in the response.
Linked
The brand's domain appears as a citation.
Implicit
A product or tagline uniquely tied to the brand appears.
Missing
No form of mention was detected.
Still need help?

Didn't find what you're looking for?

Reach out and we'll point you to the right guide — or fast-track a new one.