Arbitr
Use cases▾
Rules
European Union
Generic environmental claimsAI interaction disclosureAccessibility statement and conformity
Pay range in job postings
Reference price and promotion display
United States
Auto-renewal and cancellation disclosure
Prop 65 warning text
Made in USA claims
Endorsement and advertising disclosure
Do not sell or share link
United Kingdom
Prescribed risk warning in promotions
Gambling advertising warnings
Age verification notices
Modern slavery statement
Gender pay gap reporting
Japan
Japanese and English disclosure parity
Governance code comply or explain
Australia, New Zealand and Canada
All markets and rules
Tasks
Content localisation
ProductsDocumentsSpecificationsBrandCampaigns
Content compliance
Terminology
Approved claims
Pricing and commercial terms
Disclaimers and notices
Brand and voice
AI discoverability
Multilingual human experts
NAATIISO 17100ECFMG
PlatformPricing
Resources▾
Explore
About▾
The company
Book a demo
Home > Use cases > AI discoverability

Your buyers stopped searching. They started asking.

ChatGPT, Perplexity, Gemini and Claude now answer the questions your customers used to type into Google - and most brands have no idea what those answers say about them. Arbitr measures it, then fixes it.

Trusted by
Oracle
Entrata
Trane Technologies
Ingersoll Rand
IBM
Spirax Sarco
SEAT
Varian
Mitutoyo
Panasonic
Oracle
Entrata
Trane Technologies
Ingersoll Rand
IBM
Spirax Sarco
SEAT
Varian
Mitutoyo
Panasonic
The challenge

If the model cannot cite you, it cannot recommend you.

An assistant answers from what it has indexed, trusts and can defend. Where your information is thin, unclear or out of date, it does not leave a gap. It fills the gap with someone else. There are four ways a buying question goes wrong.

Your name never surfaces

+↑

Alternatives are recommended and the question closes without you in the room. Nothing about it reaches you.

Present here, absent there

+↑

The same question in another market returns a different answer, because that is where the least approved evidence about you exists.

Mentioned, but not chosen

+↑

You appear in the list and nothing commits - no first recommendation, and no clear reason to pick you over the others beside you.

Mentioned, and wrong

+↑

An expired accreditation, an old price, a discontinued feature - stated with confidence, carried by a source still online, and costing you trust.

The solution

Measure it. Fix it at the source. Prove it moved.

Plenty of tools will score you. Scoring is the easy half. The fix needs to know what you have already approved - Arbitr already holds it.
01
Measure
ChatGPT78%
Perplexity83%
Gemini71%
Claude52%

Where you surface, assistant by assistant, on a benchmark that holds still.

02
Correct
Online MBA durationmajor
Programme guide says 18 months; website says 24 months.
Sources: programme-guide.pdf · southernstar.edu.au/online-mba · status: open

Contradictions surfaced with both sources named, then corrected from the approved claim.

03
Prove
ChatGPT78%+12
Perplexity83%+5
Gemini71%+9
Claude52%+2
Marker = baseline · holdout maintained

Measured against a baseline and a holdout set before the change shipped.

Methodology

One score. Seven weighted signals.

The index is rebuilt from real buying questions run against real assistants, not a one-time crawl. Repeated observations across providers and locales smooth out the variability inherent to generative answers, so a trend line means something.
The same questions, locked for the period
Written for each market, not translated into it
Real answers, sampled repeatedly, not a crawl
Current index
0/100

500 buying-question prompts, 4 providers, 4 locales, 2 repeats - 4,000 observations per scan, scored against the same benchmark hash so trends stay comparable.

MethodologyEAVI_v1
Sample500 prompts · 4,000 obs
Window90 days, rolling
Coverage25%0
Prominence20%0
Citation quality15%0
Knowledge accuracy15%0
Consistency10%0
Multilingual performance10%0
Freshness5%0
Learn more about the Arbitr Platform →
Inside Arbitr

See exactly what the models see.

Each score in the index traces back to a real, inspectable observation. This is the platform underneath it.
New scan

Nothing changes but the answer

Set the assistants, the markets and the prompt set once. Re-runs use the same set and the same weights, so a movement in the score is attributable rather than noise.

New visibility scanScopeReviewRun
Assistants
ChatGPTPerplexityGeminiClaude
Markets
English · refSpanishJapaneseVietnamese
500 prompts×4 assistants×4 markets×2 repeats=0observations
Visibility Explorer

Every observation, laid bare

The exact response to each benchmark prompt - where you placed, in which language, and whether the claim made about you matches what you have approved. Inspect it, correct it, or challenge it.

Visibility Explorer5 observations
PromptLocaleProminenceAccuracy
What are the best online MBA programmes in Australia?ENTop‑threeMatches approved claim
¿Cuáles son los mejores programas de MBA en línea?ESAbsentNot mentioned
Is Southern Star Business School accredited?ENMentioned onlyStale claim
Competitors

Your share of the answer

Not market share - share of recommendation inside the benchmark. Who wins the citation when a buying question comes up, and which domains are feeding the answer on their behalf.

CompetitorsWithin this benchmark
Harbour School of Management36%
Southern Star Business Schoolyou24%
Atlas Business Institute19%
Other21%
harbourmgmt.edu.au · competitor first‑party · 33 uses
Languages

Parity, market by market

Coverage, prominence and accuracy for each market you operate in, measured against your reference language. The gap is usually wider than expected, and it is the one most tools do not look at.

LanguagesParity vs reference
Englishref100%
Vietnamese26%
Spanish18%
Japanese11%
Knowledge Layer

Your approved facts, and what they rest on

Claims, evidence and lineage in one layer - each carrying its state, its language and the source supporting it. Only claims that are approved and public are eligible to ground generated content.

Knowledge LayerClaims
ClaimLocaleState
Online MBA offers flexible entry pathwaysENapproved
Apoyo internacional para estudiantesESneeds review
Accreditation (2022 note)ENexpired
Online MBA duration 18 monthsENdisputed
Experiments

Prove the fix worked, honestly

An intervention is tied to a baseline and a holdout before it ships, and results are labelled observed, associated, or statistically distinguishable. "Caused" is reserved for designs that can support the claim.

Experiments2 running
A reviewed Spanish support page will raise ES coverage for the Online MBA clusterAssociated improvement
Baseline 2026‑04 → 2026‑06 · Holdout ES accreditation prompts
Fixing the accreditation staleness will improve EN knowledge‑accuracyInconclusive
Baseline 2026‑05 → 2026‑06 · Holdout EN pricing prompts
Knowledge and guardrails

Nothing publishes without knowing where the line is.

Your approved messaging, product and pricing detail, approved claims and the evidence behind them, and the legal, privacy and regulatory constraints that apply - held in one layer inside Arbitr.

It is also where you set the guardrails: what Arbitr must never claim, what needs marketing, legal or executive sign-off before it goes near a live channel, and who is brought in when something is ambiguous. Generated content is checked against it before it moves.

And when the evidence is not there, the draft says so. It returns insufficient approved evidence rather than writing a plausible sentence to fill the gap.

GreenApproved

Within approved knowledge and guardrails. Proceeds under your existing publishing permissions.

AmberHuman review required

Uncertainty, conflicting information, or sensitive material. Routed to a person before it proceeds.

RedBlock or escalate

Conflicts with approved information, or breaches a guardrail outright. Held and escalated, never published.

Why us

Twenty-five years experience in getting this right.

Measuring how assistants describe you is new. Knowing what an organisation is allowed to say, market by market and language by language, is not. We have been helping companies keep what they publish right in every market since long before a model was answering for them.
37,000
Expert reviewers, across 120+ languages, inside the workflow rather than outside it
25+
Years checking what regulated companies publish, in the markets they publish it in
IBM Platinum Partner
Technology partner
Your next customer already asked the question.
Book a demoExplore the platform