Results are estimates: the apps can answer differently from their APIs.
Why answers differ
- Answers vary every time. The same question asked twice can list different brands.
- Apps know you. The ChatGPT app uses your chat history, memory and location; the API does not.
- Models differ. The plugin uses low-cost models with web search by default; the app may use a larger model.
How the plugin handles it
- Web search is on for every service, so answers are close to what people see.
- The Dashboard and Wins show trends across runs, not one answer.
- Suggestions are judged against untouched prompts, two runs in a row.
- Pro can ask each prompt up to 3 times per run for steadier numbers.
Compare the same answer over time with Compare with last run on the Results page.