Read the bands
What each Sherlock Search confidence band means
The score is a comparison signal produced from the submitted face and a candidate image. It does not measure honesty, account ownership, consent, or legal identity. Use the label to decide review order, then open the linked source.
- 85–100% · Strong
- Review first. Check facial details and require consistent source context before acting.
- 50–84% · Likely
- Treat as a plausible lead. Look closely for age, angle, lighting, and feature differences.
- 0–49% · Unlikely
- Low-priority visual similarity. Keep only when strong independent context justifies review.
- No useful match
- The comparison completed without a useful candidate; it does not prove online absence.
Verification method
How to review face match confidence scores
Work from the strongest band downward, but apply the same source test to every result. A lower score with a consistent name and publication history may be more useful than a high score attached to an unrelated page.
- 01
Inspect the candidate image
Compare stable facial features while accounting for age, expression, facial hair, makeup, angle, and image quality.
- 02
Open the public source
Do not make a decision from the result card or cropped thumbnail alone.
- 03
Check independent context
Compare names, dates, locations, captions, biography details, and account history.
- 04
Consider an alternative explanation
Ask whether a lookalike, repost, filter, stale page, or poor input could explain the result.
- 05
Choose a proportionate action
Dismiss, keep as an unresolved lead, or use an official report process when the source supports it.
Ready to apply the method? Review the photo before checkout and keep each possible source in context.
Prepare a photoMatch variables
Why confidence can change between two photos
The same person can produce different scores when one image is blurry, filtered, taken years earlier, captured from the side, or partly obstructed. Two different people can also share enough visible characteristics to produce a convincing candidate.
For repeatable testing, record the input file, crop, date, and outcome. Change one major variable at a time. Otherwise, a different result set tells you little about what improved or degraded the comparison.
- Input variables: angle, light, sharpness, resolution, crop, obstruction, and filters.
- Candidate variables: age, expression, makeup, facial hair, compression, and source crop.
- Coverage variables: public availability, indexing, page removal, and provider access.
Common mistakes
Do not average scores or convert them into certainty claims
Two scores are not probabilities that can be averaged. A 90% result does not mean there is a 90% chance that the account belongs to a specific person. It means the candidate ranked with strong visual similarity under the service’s current comparison method.
Avoid publishing match cards as accusations. Share source evidence only with the platform, adviser, or authority that needs it, and describe the result as a possible visual match.
If confidence is inconclusive
Prefer an unresolved result to an unsupported conclusion
When the image and source context disagree, label the lead unresolved. Try a clearer input only if one is available and lawful to use. More searches do not turn weak evidence into strong evidence.
If the source is inaccessible, save the public URL and revisit it later rather than relying on a cached thumbnail. If the decision is high stakes, use independent verification appropriate to that context.
Questions, answered
Practical questions about this guide
01Is 85% a confirmed identity match?
No. It enters the Strong Match review band, but you must still inspect the public source and independent context.
02Why can the same person receive different scores?
Age, angle, lighting, expression, image quality, obstruction, cropping, and filters can change the available facial detail.
03Should I ignore every result below 50%?
Treat it as low priority. Review it only when independent source context makes it relevant, and do not treat it as identity evidence.
04Does no match mean the person is not online?
No. Private, unindexed, removed, blocked, or visually different images may not be available to the comparison.