ai-visibility-competitor-comparison

Comparing a target site against competitors

This skill governs how to interpret comparison.json produced by /aiv-compare and how to talk about competitor deltas without overclaiming.

What "comparison" means here

• We compare only pages that were actually analyzed on both sides.
• Per-AIVB-xxx FAIL rates are the unit of comparison.
• Delta = target FAIL rate − competitor FAIL rate.
  • delta ≥ +0.10 ⇒ target worse than competitor on that check;
  • delta ≤ −0.10 ⇒ target better than competitor on that check;
  • between −0.10 and +0.10 ⇒ effectively tied.
• Confidence is low whenever either side has fewer than 10 analyzed pages.

Workflow

1. Open the latest comparison.json under .ai-visibility/comparisons/.
2. Surface coverage warnings first; never present deltas without coverage context if the sample is small.
3. List target-worse checks ordered by delta, with AIVB-xxx, severity, delta, and minimal fix family.
4. List target-better checks ordered by delta, with AIVB-xxx, severity, and a one-line note of what the competitor is doing differently.
5. Recommend the next 3–5 minimal technical fixes, scoped to the highest-severity target-worse checks.

Cautious framing

• Do not infer total competitor superiority from one or two checks.
• Do not equate "smaller delta on AIVB-019" with "we will be cited more often". The 36 checks measure crawler-readiness/extractability, not citation outcomes.
• Note when a competitor is "better" purely because of low signal density (e.g. fewer images naturally yields fewer alt-text failures). The source research explicitly flags this for IBM in docs/source-research.md.

Red flags

• Treating competitor wins as proof of strategy correctness without accounting for sample size or signal density.
• Comparing pages that were not analyzed on both sides.
• Using Sxxx as a primary public ID in tables.
• Promising ranking or AI citation outcomes from closing a delta.
• Reducing the runtime scope to "25 checks" anywhere in comparison output.