How OneGlanse collects and analyzes AI answers

1. Collect from the product interface

Provider collection uses browser automation to submit configured prompts and capture the rendered response and citations available in that interaction. It does not use the provider's model API for collection.

2. Analyze the captured response separately

A separately configured OpenAI, Anthropic, or compatible model endpoint analyzes captured text. The resulting visibility, rank, sentiment, and recommendation fields are model-backed interpretations. They are not scores returned by the provider interface.

3. Interpret each result as a sample

An answer is an observation from one prompt run, account, product interface, and point in time. Results can change with prompt wording, account state, location, product updates, and time. OneGlanse reports the captured sample; it does not claim that one run represents every user or every answer from that product.

A repeatable UI-versus-API comparison

A product interface and an API are separate collection surfaces. To compare them, run the same prompt on both and record the conditions that could affect each result:

• Exact prompt text and any system or follow-up context
• Product interface or API, model/version when available, and run time
• Account state, language, location, and enabled search or grounding options
• Rendered answer text, named brands, recommendation order, and citation URLs
• Number of attempts, incomplete runs, and any exclusions

Repeat trials across prompts and dates. Compare answer text, brand mentions, recommendation order, and cited URLs as separate outcomes. Report the sample size and incomplete runs. Do not combine UI and API results into one metric unless the purpose and limitations of that combined metric are stated.