Platform overview
What GEO measures
GEO measures whether generative systems can identify, understand, describe, cite, compare, and safely recommend an entity. The measurement target can be an organization, brand, merchant, website, page, content item, product line, product, service, campaign, message, claim, person, topic, market, audience, or channel asset.
Every measurement is scoped to a workspace and carries its metric version, numerator, denominator, sample size, confidence, low-sample state, dimensions, timestamp, and supporting evidence. A score without this context must not be treated as a business conclusion.
Platform layers
| Layer | Responsibility |
|---|---|
| Control plane | Authentication, tenant isolation, workspace configuration, approvals, policies, audit, API and dashboard orchestration |
| Observation | Executes approved prompts against configured providers and stores immutable answer/evidence snapshots |
| Intelligence | Resolves entities, citations, demand and narratives; calculates versioned metrics and gaps |
| Optimization | Converts findings into recommendations, roadmap items, governed actions and verification plans |
| Activation | Creates evidence-grounded content/artifacts and publishes through approved channel connections |
| Measurement | Tracks AI referrals and outcomes, attributes impact and feeds observed results back into planning |
Reference topology
Independent service boundaries
The platform has independent runtime boundaries for core graph, AI components, observation, intelligence, brand knowledge, content, channels, simulation, actions, reporting, attribution, tracking, entity resolution, citation graph, prompt demand, narrative analysis, experiments, and controlled autopilot. Each service publishes health, readiness, capability, and domain contracts. The authenticated web application remains the BFF/control plane.
See Architecture and service topology for the control-plane/data-plane model, all 18 services, worker topology and deployment patterns. See Domain service reference for the contract and failure behavior of each runtime.
Observed versus synthetic data
- Observed: produced by a configured provider or tracking collector and stored with provenance.
- Derived: calculated from observed records using a versioned metric definition.
- Synthetic: created by simulation and shown with assumptions and prediction intervals.
- Planned: proposed recommendation or action that has not executed.
- Verified: an executed change with post-action evidence.
Synthetic data never contributes to observed share of voice, citation presence, or attribution.
Default governance posture
The default automation level is Suggest. The system may create a plan, but publishing or external mutation remains non-executable until policy and approval requirements are satisfied. Dry-run, idempotency, audit events, verification, rollback, safe stop, and emergency stop are part of the action contract.