How OneGlanse collects and analyzes AI answers

1. Open the real website

The Agent opens each website in a browser, runs your prompt, and saves the answer and citations it can find on the page. It does not call those products' model APIs to get the answer.

The Agent uses Camoufox for that browser layer. Read why OneGlanse moved from Chromium to Camoufox for the design history and its limits.

2. Analyze the answer afterward

An OpenAI, Anthropic, or compatible model you connect analyzes the saved answer. It produces the visibility, rank, sentiment, and recommendation fields in the dashboard. The AI website does not provide those scores.

3. Don't treat one answer as universal

One run is just one run. Change the prompt, account, location, or date and you may get a different answer. A saved answer does not tell you what every user will see.

Want to compare the website with the API?

Run the same prompt on the website and API. 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.