Open Standard · v0.9 RFC · April 2026
The Elephant Visibility Index.
An open standard for measuring LLM visibility.
EVI is a 0–100 score that tells you how often and how prominently AI assistants recommend a brand. The formula, the corpus, and the reference code are public. Anyone can compute it. Anyone can fork it. We publish it so the industry has a common number.
Creative Commons BY 4.0. Public comment period open until 2026-06-22.
The formula
How EVI is computed
EVI = 0.40 × Coverage + 0.30 × Prominence + 0.30 × Consistency
0 – 100 · weighted composite · higher is better
Coverage
40%
The percentage of category queries where your brand is named at least once. If 5 of your 25 category queries return your name, Coverage = 20.
Prominence
30%
Average rank position when named, scored on a 0–5 rubric. Named first in a response scores 5; mentioned incidentally at the end scores 1.
Consistency
30%
Stability of your visibility across all five LLMs in the panel. Low variance across models earns a higher Consistency score.
Letter grades
EVI scores map to letter grades
Letter grades give you a quick read. The underlying 0–100 score is the authoritative number for tracking and comparison.
| Grade | EVI range | What it means |
|---|---|---|
| A+ | 95 – 100 | Dominant across all five LLMs in your category |
| A | 85 – 94 | Strong, consistent presence. Category leader. |
| A− | 80 – 84 | Above average. Gaps on 1–2 LLMs or query types. |
| B | 70 – 79 | Named often, but not prominently. Ranking room exists. |
| C | 55 – 69 | Inconsistent. LLMs know you but don't reach for you. |
| D | 40 – 54 | Below average. Competitors have meaningful advantage. |
| F | 25 – 39 | Rarely or never named. Category-invisible to AI. |
| F− | 0 – 24 | Not found across the panel. Complete LLM invisibility. |
Why it exists
A common number the industry can use
Closed proprietary scores already exist. DerivateX AVS, iPullRank's AI visibility scoring, and Search Atlas's LLM visibility metric are all real products. The problem is that you cannot compare them, reproduce them from first principles, or audit how they are computed. Each one is a black box tied to a vendor relationship.
The internet's most enduring infrastructure is open. PageRank's original paper is public. DKIM is an RFC. SOC 2 is a published framework. Schema.org is a community-maintained vocabulary. Every measurement standard that became a standard was open. Closed standards don't become standards — they become footnotes.
We are an LLM SEO firm. We have a commercial interest in more organizations measuring LLM visibility. The fastest way to grow that measurement practice is to publish the methodology and let everyone use it for free. If agencies, platforms, and competitors all use EVI, the category grows. That is good for us.
Scope
What EVI measures — and what it doesn't
EVI is deliberately narrow. It measures one thing precisely rather than many things loosely.
What EVI measures
- How often AI assistants name your brand in category queries
- How prominently they name you when they do
- How stable your visibility is across different AI tools
What EVI does not measure
- Sentiment — whether mentions are positive or negative
- Click-through rate or referral traffic
- Revenue attribution from AI-driven discovery
- Hallucinated facts or incorrect claims about your brand
The corpus
A public, maintained query registry
EVI scores are computed against a defined corpus of category queries. The corpus is published and versioned on GitHub. Anyone can submit a pull request to add or modify queries. Each category has a named maintainer committee.
Categories launched in v0.9
- 25 category queries per category, published in full
- Maintained on GitHub as an open registry under CC-BY 4.0
- Pull requests accepted to add or modify queries
- Each category has a named maintainer committee
- New categories require a 30-day public comment period before launch
Versioning
All versions are frozen and reproducible
A core requirement of EVI is that any score computed at any point in time must be
reproducible by any party with access to the same raw LLM responses, the same
corpus version, and the same formula. v1.0 computed today must match
v1.0 computed in 2030.
Version changes are not patches. Adding a model to the panel, adjusting a weight, or changing query phrasing requires a version bump and a full re-run of the panel. Scores from different EVI versions are not directly comparable. Every EVI report carries its version number alongside the score.
The reference Python implementation is frozen per version. evi-python
ships one module per major version, and older modules are never modified.
How we use it
Every report we sell returns your EVI
We built EVI because we needed a rigorous, reproducible output for our own client work. All three of our service tiers are built on top of it.
-
$49 Visibility ReportYour EVI — with Coverage, Prominence, and Consistency broken out — plus a ranked list of what's pulling it down. Delivered in five minutes as a PDF.$49
-
$999 Starter AuditFull EVI panel across all 25 category queries, a 30-day EVI-lift projection, and a prioritized implementation roadmap.$999
-
$15K Done-For-You EngagementA contractual EVI improvement target. We implement the changes, re-run the panel monthly, and report against the baseline.$15K
FAQ
Six questions about EVI
Who owns EVI?
Elephant Accountability LLC is the current editor and maintainer of the EVI specification. The content of the spec is published under Creative Commons BY 4.0. Anyone can implement, fork, or sell services on top of EVI with attribution. The reference Python code is MIT-licensed. Neither the spec nor the code is proprietary.
Can my agency use EVI in our own reports?
Yes, that is the point. Attribution requirement: "Computed per EVI v0.9 methodology, eaccountability.org/evi." No license fee. No approval required. If you make improvements or corrections to the methodology, we encourage submitting them as a pull request to the spec repository.
How is this different from DerivateX AVS or Search Atlas's visibility score?
Those are closed proprietary scores from commercial vendors. You cannot reproduce an AVS number from raw LLM responses without purchasing the platform. You cannot inspect the formula. You cannot fork it. EVI is the opposite: the formula, the corpus, and the reference Python implementation are all public. Any party with the same LLM responses, the same corpus version, and the same formula version should arrive at the same EVI score independently.
What about model drift?
Every EVI report carries its version. The v0.9 panel is:
- ChatGPT-4o
- Claude Sonnet 4.6
- Perplexity Sonar
- Gemini 2.5 Pro
- Copilot
When model versions change meaningfully enough to shift scoring behavior, or when a new model warrants inclusion in the panel, the EVI version number bumps. Prior-version scores are not retroactively modified. This is by design.
Is EVI a replacement for SEO?
No. Classic SEO measures Google rankings and the traffic that follows. EVI measures whether AI assistants cite your brand when a buyer asks for a category recommendation. Both surfaces matter. They have limited overlap in what they optimize for. A strong Google presence does not guarantee a strong EVI, and vice versa.
What happens after the RFC period?
The public comment period runs through 2026-06-22. We incorporate substantive feedback, resolve open issues in the spec repository, and freeze v1.0 on 2026-08-01. That version persists indefinitely — no modifications, no retroactive changes. Future improvements require v1.1 or v2.0 and a new panel run.
Governance and license
Open. Versioned. Attributed.
The EVI specification is published under Creative Commons BY 4.0. The reference Python library is MIT-licensed. Elephant Accountability LLC acts as editor, not as sole authority. The governance model is documented in the spec repository.
Compute your EVI — $49 Visibility Report
The $49 Visibility Report runs your brand through the full v0.9 panel and returns your EVI with Coverage, Prominence, and Consistency scores, plus a ranked list of what's holding you back. No call required. Five-minute delivery.