AI Recommendation Intelligence for B2B teams

AI recommends your competitors. We measure why, and what to change.

An independent U.S. firm for the teams whose reputation AI now decides. We measure how models recommend, cite and represent you, across competitors, media, vendors and investor diligence, on global and sovereign models, and turn it into decisions you can act on.

models 7  ·  lenses 2  ·  runs/probe K≥5
signals/probe ~40  ·  infra dedicated GPU
measured acrossChatGPTClaudeGeminiPerplexityFalconJaisQwensovereign · no public API
Built for CMOs, heads of communications, founders raising capital, and the agencies serving them.
Engagements under NDA · we work with a limited number of organizations at a time.
Who we are

A measurement instrument, not an agency, not a dashboard.

An instrument with its own methodology and index, not a content shop. We never execute the work we recommend, which is what keeps the diagnosis honest.

Who we are

An independent AI recommendation intelligence firm. U.S. company, MENA reach.

What we are

A measurement instrument with a proprietary methodology and index, not a content shop.

What we do

Measure and explain why AI recommends one entity over another, across global and sovereign models, and identify what to change.

What we don’t do

We don’t write content, run SEO, build websites or place PR. That independence is what keeps the read objective.

Not SEONot a visibility dashboardNot an agencyProprietary indexMulti-model + sovereignIndependent by design

Monitoring tools tell you that AI mentioned you. None tell you why it recommended someone else, or what to change.

What we deliver

Decisions, not dashboards.

Each engagement answers a specific question a leader is already asking. These are drawn from real work: the same engine, pointed at a different decision.

Media & PR allocation

Which journalists and outlets to back, and where to publish

We run your media list through the engine and rank who the models actually recognize and cite, mapped to your goals, regions and languages. You stop spending PR budget on coverage AI ignores.

from a real engagement · journalists scored global + regional
Independent vendor review

Is this agency proposal any good, measured against your goals?

We pressure-test an incoming proposal or compare bidders: what each covers, what it misses, and whether its headline claims hold up in the models and regions it never tested. We have no stake in who wins the work, which keeps the read honest.

from a real engagement · goal-by-goal coverage map
Competitive recommendation gap

Why AI names your competitor as the answer, even when you are the official one

Prompt by prompt, we measure where a rival owns the recommendation and decompose why: comparative framing, fact distortion, the sources AI trusts. We have seen an official category partner stay invisible to AI while two rivals were named in every answer.

live competitive case · named-competitor head-to-head
AI investor-lens audit

What AI tells investors about you, in every region you are raising in

Before you raise in the US, the Gulf or Asia, we measure how the AI tools investors use for first-pass diligence describe and rank you, region by region, model by model, language by language. You fix the gaps before the roadshow, not after.

measures the AI-mediated layer of perception, not human sentiment
Result · selected engagement

A fintech moved from 7th to 2nd in comparison-intent AI answers across three models in 60 days.

+31
Recommendation share gain, commerce intent
7 → 2
Rank in “best platform for…” answers
3 / 3
Global models now recommending it
Sample result, pending a real client-approved case. Replace before publishing.
U.S. company · GEONYX LLCDedicated GPU via CrusoeNDA-first engagementsIndependent · no stake in execution
Our own methodology, index and metrics

A measurement discipline of our own.

We built a proprietary methodology, our own scoring index, and a multi-model dataset that grows with every scan. The metrics below are ours: named, defined and reported consistently. The exact prompts, weights and calibration are a trade secret and stay with GEONYX LLC.

Per probe

7 models × 5+ runs × 2 lenses. 35+ measurements behind one score.

7models queried
2lenses · global + sovereign
5+runs per model
~40signals per probe
Confabulated mentions and fabricated citations are dropped before scoring. The result is decomposed by intent, verified against a client-approved truth-set, and returned as a 0–100 index with its components and the exact prompts where you lose.
Stability Scoreproprietary · headline
How reproducible your recommendation is across repeated runs of the same prompt. LLMs are non-deterministic, so a single answer is noise. This is the number no monitoring tool reports.
GEONYX Indexcomposite · 0–100
A single executive read, always shown with its components beneath it. The composite sells; the components defend.
Intent-weighted SOMproprietary
Recommendation share weighted toward commerce intent, where revenue happens, not a flat average across every prompt.
Authority Surfaceproprietary filter
Share of citations that come from real, verified sources, after fabricated citations are filtered out by validation.
Confabulation controlengine layer
Separates genuine recognition from confident invention, so a fabricated answer is never counted as visibility.
Fig.01 · pipeline
The system behind the score

A measurement engine, not a survey.

Each score is the output of a versioned pipeline. The stages below show the data flow. The prompts, scoring rules and calibrated weights are proprietary and stay with GEONYX LLC.

01
Prompt universe
in: intents, EN/AR
02
Multi-run exec
M models × K runs
03
Parser
LLM → JSON
04
Signal extraction
~40 signals/probe
05
Confabulation filter
real vs fabricated
06
Authority validation
DNS / known-works
07
Composite index
out: 0–100 + components
metrics · selected§ scoring
vp(probe) = rule(recognition, accuracy, citation) ∈ {0,1,2,3} Visibility_L = 100/3 · mean_k( vp ) per lens L Stability = 100 · e^(−λ · CV) CV = σ_rank / μ_rank Gap = Visibility_global − Visibility_regional Index = Σ wᵢ · normᵢ // weights proprietary
λ, weights and rules calibrated on real scans. Not disclosed.
parser output · one probe§ schema
{ "entity": "Acme Compliance", "model": "gpt-4o-2024-08-06", "lens": "global", "run": 4, "signals": { "recognized": true, "outlet_ok": true, "articles_offered": 3, "articles_verified": 2, "fabrication": false }, "vp": 3 }
Every response is reduced to structured signals before scoring.
Coverage by region

Global tools reach 4 models. We measure 7, including 3 with no public API.

ChatGPT, Gemini, Perplexity and Claude are one slice. The sovereign and Arabic models buyers trust in the Gulf, Falcon, Jais and Qwen, sit behind no public API. We run them on dedicated GPU, in-language, so a gap that is invisible to every API-only tool comes back as a measured number.

USA
English
ChatGPT
Claude
Gemini
Perplexity
Global frontier, where most funds start. Strong here can hide weakness everywhere else.
China
Mandarin · behind the firewall
ERNIE
Qwen
DeepSeek
Hunyuan
A separate world. Visibility in global models means nothing here.
South Korea
Korean
HyperCLOVA X
EXAONE
Solar
A.X
Sovereign models, measured in Korean.
India
Hindi · English
Sarvam
Krutrim
BharatGen
+ global
Sovereign stack exists, but global models still dominate usage.
MENA
Arabic · English
Falcon
Jais
Qwen
On dedicated GPU. Coverage no API-only tool can offer.
Why this is not a $200 tool

A monitoring tool counts mentions. We measure recommendation, and explain it.

The price gap is a capability gap. Here is what a dashboard physically cannot do.

A monitoring dashboard
GEONYX
Tells you if AI mentions you.
Tells you why AI picks someone else, and what to change.
Takes a single answer as fact.
Runs many times per model and reports how stable the answer is.
Global API models only.
Global plus sovereign models, Falcon, Jais and Qwen, on dedicated GPU.
Counts any mention, even invented ones.
Filters fabricated claims and citations before scoring.
A login you check once.
An analyst read tied to your goals, with the next move spelled out.
What you receive

Analyst-grade intelligence with clear instructions, not a login.

A report your team reads and acts on carries more weight than a dashboard visited once and forgotten. Every report tells you what to do next, why, and how we will verify it.

Executive Snapshot · sample layoutformat only · not a client result
63
GEONYX Index
Global vis71
Regional vis48
Stability66
Win rate2 / 5
Gap to leader−23
Authority surface54%
Monthly

Executive Snapshot

One page for leadership: the read at a glance, plus the single most important move.

  • Visibility change, global and regional
  • Win rate and Stability Score
  • Gap to the category leader
  • The month’s risk flag and next action
Quarterly

Deep Diagnostic

Ten to fifteen pages of structural decomposition and a prioritized plan.

  • Decomposition by discovery / comparison / commerce
  • Head-to-head against named competitors
  • Fact and authority-source analysis
  • Interventions, each with expected impact
How to start

Start small. Scale into a program.

Begin with a scoped diagnostic on one category. It is a real read, not a demo, and it tells you whether a deeper program is worth it before you commit.

Start here

Scoped Diagnostic

from $8,500 · about 2 weeks

One category, the core models, your top competitors. A real read on where you stand and why, with the single most important move to make.

Foundational Audit

Custom scope

The full baseline: every intent, global and regional lenses, sovereign models, competitor decomposition and a prioritized plan you can act on.

Ongoing Program

Monthly + quarterly

Executive Snapshot each month, Deep Diagnostic each quarter. We diagnose, you act, we re-measure and confirm what moved.

DiagnoseYou actWe re-measureConfirm what moved
The disciplines behind the measurement

Senior specialists who lived inside the systems that now decide recommendations.

We don’t sell these disciplines as services. We turned them into a measurement discipline of our own. Knowing how these systems work from the inside is exactly what lets us diagnose why a model recommends your competitor.

Creative direction & brand standard
15+ years · 2 Red Dot · 3 Pentawards

Why our reports read like a scientific instrument

The same restraint that earned Red Dot and Pentawards recognition is what makes a diagnostic legible to a CMO and defensible to a CTO. We design the result to be acted on, not decorated.

Search & intent
7+ years · organic search, semantic research

Why we measure intent, not a flat prompt list

Years inside commercial search intent are the reason our methodology separates discovery, comparison and commerce, and finds where in the buying journey a brand actually loses.

AI discovery & recommendation signals
7+ years · inside the U.S. AI ecosystem

Why we can diagnose the signals behind a recommendation

We understand entity signals, source authority and content structure deeply, which is what lets us isolate which factors move a model and which do not. We measure the signal; acting on it is your team’s work, never ours.

Editorial & factual accuracy
7+ years · editorial judgment

Why we can tell when a model gets your facts wrong

Editorial expertise powers our truth-set and Fact Precision layer, comparing what a model claims about you against an approved set of facts, and flagging distortion before it costs you.

Engineering & infrastructure
7+ years · dedicated GPU via Crusoe

Why we can measure models no API can reach

Our engineers built the multi-model, multi-run engine and operate it on dedicated GPU, the reason we can diagnose sovereign models like Falcon, Jais and Qwen that API-only tools cannot access.

How you work with us
Senior at every stage · no junior layers

One senior team, on your side of the table

You work directly with the specialists who run your measurement. And because we have no stake in who executes the work we recommend, the read stays independent.

One senior team across brand, search, content, AI and engineering, pointed at a single question. We measure why your recommendation moves, and prove what changed it. We don’t execute the work.

Who we work with

Organizations whose reputation is now shaped by AI.

A focused set of B2B organizations in competitive or regulated categories, plus founders raising capital where AI now shapes the first impression.

Brands & enterprises

Companies in competitive categories

See why AI recommends a competitor, where you disappear in purchase-intent queries, and what specifically moves the ranking.

Communications & PR

In-house teams and their agencies

Validate media plans, publications and vendor proposals against the sources AI actually cites, before you commit budget.

Founders raising capital

Teams heading into a round

Know what AI tells investors about you, region by region, before the roadshow, and close the gaps that show up in first-pass diligence.

Gulf & MENA

Regional and sovereign-AI users

Diagnostics across sovereign and Arabic models such as Falcon, Jais and Qwen, on dedicated GPU. Coverage no API-only tool can offer.

Live probe

Watch one probe run across the model set.

Type an entity and run a single probe. Each model is queried, parsed and scored, then reconciled into a visibility read. This is an illustrative simulation of the pipeline. Live engagements run real models on dedicated GPU infrastructure, under NDA.

probe · entity
ChatGPTglobalqueued
Geminiglobalqueued
Perplexityglobalqueued
Claudeglobalqueued
Qwenregionalqueued
Falcon-Arabicregional · no-searchqueued
Jaisregional · no-searchqueued
Global vis··
Regional vis··
Stability Score··
GEONYX Index··
single probe · live engagements use K≥5 runs per model on dedicated GPU, under NDA
Cooperation by request

We work with a few organizations at a time.

Every engagement begins with a request and a scoped diagnostic. We review each one personally. If it is a fit, we move to a full assessment, under NDA.

  • 01Send a request with your company and competitive field.
  • 02We run a scoped scan and share what we find.
  • 03If it is a fit, we agree on scope and begin, under NDA.