Remember when rank tracking felt exact? Position 4.2 on Tuesday, 4.1 on Wednesday—numbers with false confidence, but comforting. AI visibility breaks that habit. Models resample, providers refresh weights, and user context changes answers. The micro metric (one run, one screenshot) is gone; the macro trend (distributions over time) is what remains.
The micro–macro shift
| Micro (fragile) | Macro (durable) |
|---|---|
| Single answer screenshot | Mention rate over N runs |
| “We were cited once” | Citation rate per prompt set |
| Subjective “sounds positive” | Tagged descriptor buckets (leader, budget, risky) |
| Daily prompt edits | Versioned prompt list with changelog |
Executives still want a number. Give them ranges and direction with explicit methodology.
Build a measurement contract
Document these before you report externally:
- Prompt registry — text, intent tag (branded / category / comparison), owner.
- Model list — which providers you include each cycle.
- Run count — minimum three per prompt; more for volatile categories.
- Scoring rubric — binary mention, optional position band, incorrect-fact flag.
- Cycle length — weekly operational, monthly executive.
When someone asks “why did we drop 12 points?”, you can answer whether the prompt set, model version, or site release changed.
Qualitative review is not optional
Numeric mention rate hides wrong facts. Schedule a 30-minute review each cycle:
- Pricing or tier names accurate?
- Compliance claims present or absent as required?
- Competitors named fairly or omitted?
Tag errors and route to legal or product marketing—not only SEO.
When precision returns (partially)
Some teams maintain Recall Test suites with explicit ground-truth tables for regulated claims. Use them sparingly where you can afford curated facts; for most marketing narratives, prioritize schema + cite-ready copy first.
theaivis for macro monitoring without spreadsheet chaos
theaivis centralizes GEO audits, Audit comparisons, and scheduled visibility prompts so macro trends live in one workspace instead of scattered folders. Monitor your brand across providers on a cadence your leadership approves—then attach module scores (citability, schema, platforms) to explain why mention rate moved. Precision may be gone; disciplined repetition is not.