Every value Osello's AI writes gets a confidence score from 0 to 100 — the AI's honest grade of how sure it is about its own answer. Values at or above the field's threshold are applied to your listing; anything below is held back for you to review.
How the score reads
Scores map to plain-English labels wherever they appear:
| Score | Label |
|---|---|
| 95-100 | Very High |
| 80-94 | High |
| 60-79 | Medium |
| 40-59 | Low |
| 0-39 | Very Low |
Alongside the number, the AI writes a short explanation of why it's confident (or isn't) — for example, that the fabric content came straight from your product description versus being inferred from a photo.
The threshold
By default, a value needs a score of 80 or higher to be applied to the listing. Some fields carry their own stricter or looser threshold, depending on how costly a wrong answer would be for that field.
When an answer scores below the threshold, Osello does not write it to your listing. The field keeps whatever value it had, and the AI's answer stays visible in the prompt's results, marked as unused, so you can judge it yourself.
Review a prompt's scores
- In the sidebar under AI, click Field Prompts and open a prompt.
- The line items table shows each field with its before value, after value, and a confidence badge. Hover the badge to read the AI's explanation.
- Click a field to open its full detail: the score, the explanation, the before and after value for every variant, and the exact instructions the AI followed. An answer that scored too low is labeled "Value was not used (confidence too low)."
What to do with a low-confidence answer
- If the AI's answer is right, or you know the correct one: enter it yourself on the listing's Fields tab. Your value is treated as final.
- If the AI keeps missing for a reason you can name: add custom instructions to that field, then run the prompt again — from the field's row menu choose Update with AI, or use Retry on the prompt.
- If the source data was the problem (a thin description, a missing spec), enrich the product in your store; Osello imports the change and the next run has more to work with.
Low scores are the system working as designed: you spend review time only where the AI itself says a human should look.