AI Verified endpoints are designed for both human developers and AI systems.
They are easier to understand, faster to integrate, and more reliable in automated workflows.
🧠AI Verification Methodology
What "AI Verified" Means
An endpoint marked as AI Verified has been tested with multiple AI systems and achieved a high score in our AI-readability evaluation.
This means the endpoint is:
- clearly understood by AI models
- usable for generating valid requests
- correctly interpreted within typical workflows
- supported by sufficient examples and error handling
Our goal is to ensure that both developers and AI systems can integrate reliably with minimal friction.
Why We Validate APIs for AI
Modern integrations increasingly involve AI systems that:
- read API documentation
- generate requests
- automate workflows
Traditional documentation is written for humans. We go one step further and ensure it is also:
- machine-understandable and integration-ready
How We Evaluate Endpoints
Each endpoint is tested using a standardized evaluation process across multiple AI systems.
AI systems used
We validate endpoints using a diverse set of models, including:
- OpenAI models
- Anthropic Claude
- Google Gemini
- Open-weight models (e.g., Llama family)
This ensures that the documentation is not optimized for a single AI system.
Evaluation Criteria
Each endpoint is scored across five categories:
1. Endpoint Clarity
Can the AI correctly understand:
- what the endpoint does
- when to use it
- what inputs and outputs represent
2. Schema Completeness
Can the AI:
- generate valid request payloads
- correctly identify required and optional fields
- respect data types and constraints
3. Example Quality
Do the provided examples:
- reflect real-world usage
- help the AI generate correct requests
- reduce ambiguity
4. Flow Usability
Can the AI:
- place the endpoint within a broader workflow
- correctly sequence calls
- choose the right endpoint for a given task
5. Error Explanation
Can the AI:
- understand possible errors
- interpret error responses
- determine how to recover from failures
Scoring System
Each category is scored from 1 to 5 across multiple AI systems. The final score is calculated as a weighted total:
| Criterion | Weight |
|---|
| Endpoint clarity | 25% |
| Schema completeness | 25% |
| Example quality | 20% |
| Flow usability | 15% |
| Error explanation | 15% |
What Qualifies as "AI Verified"
An endpoint receives the AI Verified badge if:
- it achieves a total score of 90 or higher, and
- it passes all critical thresholds:
- Endpoint clarity ≥ 3.5
- Schema completeness ≥ 3.5
- Error explanation ≥ 3.5
In addition:
- multiple AI systems must agree on the endpoint's purpose
- AI-generated requests must be valid in the majority of cases
What "AI Verified" Does Not Mean
AI Verified does not guarantee:
- perfect AI behavior in all cases
- fully autonomous integrations without supervision
- absence of edge cases or business complexity
Instead, it indicates:
- high reliability and low friction when used by AI systems
Continuous Improvement
We continuously:
- re-evaluate endpoints
- improve documentation based on AI feedback
- expand test coverage with new use cases
As AI systems evolve, our evaluation process evolves with them.
Feedback
If you encounter issues when using an AI Verified endpoint:
- please contact us
- or report the issue via our support channels
Your feedback helps us improve both the API and its documentation.