Valoh
Research Notes & Methodology Updates
What we publish.
Methodology versioning notes. Aggregate, anonymized category benchmarks. Reading on the history of independent measurement. We don't publish marketing posts, vendor comparisons we have a stake in, or thought leadership about how AI is changing everything. The point of these notes is to make Valoh's measurement work auditable in public.
Most Recent
Methodology v1.2 release notes: prompt-set review cadence and a new sentiment rubric.
We've moved prompt-set review from quarterly to biannual, expanded the sentiment rubric to better capture cautionary framing, and published a new validation κ for recommendation strength. This note documents what changed, why, and how prior windows compare.
Read the note| Prompt review | Quarterly → Biannual |
| Sentiment rubric | 3-tier → 5-tier |
| Validation κ | 0.81 / 0.76 |
| Models | 4 (no change) |
| Backward-comparable | Yes, with footnote |
Recent Notes
All published research notes, newest first.
Prompt-set review cadence moves to biannual, sentiment rubric expanded to five tiers, new validation κ scores published.
Aggregate, anonymized mention rates, rank positions, and recommendation tiers across the running-shoe category for Q1 2026, by model.
Why Valoh reports five named signals at stable units rather than rolling them into a proprietary aggregate. The case for transparency over benchmarkability.
Coverage decisions are real. We currently track ChatGPT, Gemini, Claude, and Perplexity. Here's the framework we use to decide when a fifth gets added.
A short reading list of public material on the path from "third-party measurement vendor" to "MRC-accredited methodology." Useful context for anyone scoping AI brand measurement.
The validation procedure for sentiment, including the published rubric, the stratified sample protocol, and the inter-rater agreement scores from the last four quarters.
Aggregate, anonymized signal values for the DTC skincare category, by model, for Q4 2025. Includes notes on category breadth and a known limitation around private-label brands.
Editorial Standards
What we will and won't publish here.
- —
- Methodology updates. Every change to prompt sets, model coverage, sampling, or validation gets a note before it lands. Versioned and dated.
- —
- Aggregate category benchmarks. Quarterly. Anonymized. Categories with sufficient measurement coverage only. Individual brand values are never published without explicit client permission.
- —
- Positioning notes. Why we made a measurement-firm choice that distinguishes Valoh from optimization tooling. Useful for buyers; useful as our own thinking-out-loud.
- —
- Reading lists. Public material we found useful for understanding measurement-firm history, MRC processes, or the AI category.
- —
- No vendor takedowns. We don't write competitive comparisons that single out specific AEO/GEO vendors by name. Category-level positioning, yes; named takedowns, no.
- —
- No marketing posts. No "Top 10 ways AI is changing X." If a note doesn't update methodology, publish data, or sharpen positioning, it doesn't go up.
Get Notes by Email
Methodology updates and quarterly benchmarks, sent when published.
Low frequency. Roughly one email a month. No marketing.