Observational LLM & Generative Engine Research
ASTONTO studies outputs that users can observe. We do not claim access to proprietary model internals or unreleased vendor algorithms.
Black-Box Evaluation Framework
Because commercial AI platform vendors do not publish real-time internal weights, empirical evaluation relies on controlled black-box testing. We isolate observable input signals, prompt variations, and output consistency across ChatGPT, Perplexity, Gemini, and Google AI Overviews.
Published Studies & Methodology Notes
Black-Box Evaluation of Observable AI Recommendation Behaviours
An overview of ASTONTO’s methodology for measuring how large language models and AI-enabled search systems represent, compare and recommend organisations through observable public outputs.
In methodology review: a study of how AI assistants recommend managed IT providers in Greater Manchester. Studies are published only after methodology and technical review.
Ethics & independence
- Every published finding is backed by preserved, verbatim AI outputs and a documented prompt set.
- We test only publicly available interfaces and never claim knowledge of model internals.
- We accept no commercial arrangement that would restrict or pre-determine research results.
- Client names, metrics and results are never published without written consent.
- We do not promise rankings, citations or recommendations from any AI platform.
Citing ASTONTO research
You are welcome to quote or reference our published research. Please cite the organisation, study title, methodology version and URL, and quote findings together with their testing period and reliability rating.
ASTONTO (2026). Study title. PULSE Method v1.0. https://astonto.com/research/…