Original analysis
Freeze the AI Visibility Baseline Before Optimizing
A small prompt-panel protocol that records the starting distribution before content, technical, entity, and authority work begins.
What matters
- Keep non-brand, brand, comparison, local, and factual prompts separate.
- Record absence and incorrect answers as carefully as citation wins.
- Pair AI trials with search, crawler, referral, and conversion baselines.
The minimum viable baseline
Start with 30 to 60 prompts representing actual discovery and buying questions. Classify them before running them. Write a short scoring guide for mentions, citations, correctness, and recommendations. Then run more than one trial per prompt and preserve the complete output.
The result may be zero owned citations. That is useful. A baseline is a comparison point, not marketing collateral.
What to freeze
- Exact prompt wording and ID.
- Included products and modes.
- Intended geography and account state.
- Number and schedule of repeated runs.
- Brand mention and citation scoring rules.
- The canonical fact register used for correctness.
- The change window and the next measurement date.
Do not rewrite prompts midstream because another wording performs better. Put newly discovered questions into the next frozen panel.
Use the complete AI visibility measurement guide to build the run ledger.
Evidence & maintenance
How this page is maintained
- Content basis
- Original analysis
- Evidence grade
- Expert analysis
- Next review
- Dec 10, 2026
Material errors can be reported through the public corrections process.
Sources
- Towards a Measurement-Based Audit of Generative AI Citation Behavior — Proceedings of Machine Learning Research