Short definitions of the terms used throughout KnitKnot and this documentation. Each links to a fuller explanation where one exists.

Core objects

Subject — the thing a prompt benchmarks: either your brand or one of its products. One company has one brand subject and zero or more product subjects. The canonical way everything is scoped per-product. See Subjects.

Brand — your company as a whole; the top-level subject. A recommendation of any of your products counts as a brand win.

Product — a subject under your brand, with its own prompts, competitors, and report slice. See Benchmark products separately.

Prompt — one buyer-style evaluation question. Prompts live in a persistent prompt library that stays stable across runs, so your score stays comparable. See Prompts and Manage your prompt library.

Run — one full benchmark: your active prompt library executed across the selected engines and scored. Runs are the unit of measurement over time. See Runs and scoring.

Evaluation — one scored AI response: a single prompt run against a single engine. The atom every metric is built from.

Engine — one of the four AI systems a prompt runs through: ChatGPT, Claude, Perplexity, Gemini. See Engines.

Competitor — a vendor AI compares you against; head-to-head results are scoped to your competitor set. See Manage competitors.

Metrics

AI Presence Score — the 0–100 composite headline number. See Metrics reference.

Coverage — the per-response label for how prominently you appear: primary, substantial, peripheral, incidental, or absent.

Visibility rate — the share of scored responses where you appear at all, computed organic-only (excluding prompts that name you).

Win rate (W-L-T) — of decided head-to-head comparisons, the share you won; ties count against you.

Sentiment — how favorably you’re framed when you appear, on a 0–100 scale.

Full definitions: Metrics reference.

The fix loop

Issue — a tracked gap a benchmark surfaced (factual error, outdated claim, missed feature, losing comparison, visibility gap), with its evidence attached. See Issues and playbooks.

Playbook — an evidence-grounded create, revise, or repair brief linked to one or more issues and a measurable hypothesis.

Source — a web page present in a captured AI response’s citations. A claim is attributed to that page only when a reliable response-native signal can be bound to the claim.

Head-to-head (H2H) — a comparison prompt that names you against a competitor. Excluded from organic visibility, counted in win rate.

Access

Workspace — your company’s shared data in KnitKnot. Teammates on the same email domain join the same account. See Getting started.

Public report — a published, read-only report at {workspace}.knitknot.io/reports/{slug} that anyone with the link can open. See Share your report.

MCP — the Model Context Protocol connection that lets AI assistants query and act on your workspace. See Connect to your AI tools.