AI Visibility Engine

V0.1 / METHODOLOGY

How we measure brand visibility in AI

We probe the major LLMs with niche-specific prompts and translate their answers into one number you can act on — the ECHO.

ECHO = 50 × 30 × 20

01 — 03

The ECHO Index is a weighted blend of three things every brand needs from AI search:

50%

Citation

How often the LLM actually names your brand in its answer. Zero mentions = zero ECHO no matter what else you do.

30%

Sentiment

Whether the mention is positive, neutral, or negative. Being known and hated isn't a win.

20%

Authority

Whether the LLM cites your own domain as a source. If yes — you control the narrative.

What we scan

01 — 06

Every audit fans out across all active providers in parallel. Every prompt × every model = one row in the result matrix.

  • 01

    OpenAI

    GPT-5

    EN / RU

  • 02

    Anthropic

    Claude Opus / Haiku

    EN / RU

  • 03

    Google

    Gemini 2.5

    EN / RU

  • 04

    Perplexity

    Sonar Pro

    EN

  • 05

    GigaChat

    Sber

    RU

  • 06

    YandexGPT

    Yandex Cloud

    RU

Russian engines (GigaChat, YandexGPT) are first-class — most Western tools skip them.

“If AI doesn’t cite you, you don’t exist in the answer.”

What happens during an audit

01 — 05
  1. 01

    Intake

    We fetch your homepage and parse <title> + meta description, so the LLMs are graded against your real positioning — not their hallucination of the domain name.

  2. 02

    Multi-provider fan-out

    We send the same prompts to every active provider in parallel — full matrix, no cherry-picking.

  3. 03

    Extraction

    We extract brand mentions, sentiment heuristics, and cited source URLs from each raw response.

  4. 04

    ECHO computation

    Citation, sentiment, and authority are aggregated and weighted 50/30/20 into one number from 0 to 100.

  5. 05

    Snapshot

    Each ECHO computation is persisted as a time-series snapshot so you can see trends after weekly re-scans (Phase 3).

Try it

Drop your domain — first audit takes about 30 seconds.