Start with the comparison being performed
AI photo matching uses a reference image to help find or compare candidate images in available dating-profile evidence. Its useful output is a set of candidates and comparison information for review. It does not identify every account belonging to a person, establish who uploaded a photograph, or determine whether someone is currently using an app.
The word “AI” also does not establish that a service is more accurate than a person reviewing the same evidence. That claim would require a relevant evaluation with known answers and measured errors. OopsBusted’s comparison bands are not an independently validated accuracy percentage. This guide explains the workflow and its boundaries rather than promising superiority over another method.
If you are choosing between methods, use the manual and AI comparison guide. If you already have a report, the facial-match reading guide explains its labels.
Prepare the search details and reference image
A standard OopsBusted search uses a name, adult age range, gender, and location. A reference photo is optional for that path. Enabling AI photo matching requires an uploaded photo. With AI matching enabled and a photo supplied, the photo-only path can omit the name; the other required details still apply.
Choose a reference that shows the intended person clearly. The upload form accepts up to five photos, with a maximum of 1,000 KB per image. The upload API accepts JPEG, PNG, and WebP image types. A file being accepted is different from a successful comparison, so review the feedback instead of assuming the upload completed every stage of the search.
The photo preparation guide covers format, framing, and rejection messages. Do not obtain private account access or unrelated personal material to fill gaps in the information you have.
Understand what the input check establishes
The photo check looks for a usable face in the uploaded image. It can reject an unreadable file, an image without a usable detected face, a face that occupies too little of the frame, or an image with multiple detected faces. Those messages concern suitability of the input. They are not search results about the subject.
For example, a group image may show a person you recognize, but the system still needs an unambiguous reference. Use a suitable individual image or crop out unrelated people while preserving the subject’s head and useful detail. If the file cannot be read, retrying or choosing a different supported file addresses that input problem; it does not change search coverage.
A usable-face result does not confirm identity. It indicates that the input passed this stage. Coverage, comparison, attribution, and interpretation still require their own checks afterward.
Keep available evidence separate from app selection
The initial search checks available Tinder, Bumble, and Hinge data. App selections are editable clues in the shared intake, not a guarantee that only the selected app is examined. A later paid unlock determines which app reports you can access. Access to more reports does not create missing profiles or fresh captures.
Photo matching operates on the evidence available to the service. A good reference cannot compensate for an account that is absent from that evidence. Likewise, a comparison cannot reveal private messages or a password. A service should explain these boundaries before you use a visual match as the basis for a broader conclusion.
Review the search-input and report-scope guide if you are deciding whether the workflow fits your question. A subject’s phone number or email address is not a supported lookup input.
Read the comparison as a fallible estimate
Face detection and face comparison answer different technical questions. Detection concerns whether a face appears in an image. Comparison concerns resemblance between faces. Amazon’s comparison documentation describes probabilistic behavior, possible false negatives, and the importance of human review before consequential decisions. This provider guidance does not validate OopsBusted’s end-to-end results.
OopsBusted uses bands such as “Possible face match” and “Strong face match” for review. A band does not establish ownership of a dating account, current activity, or relationship behavior. It also is not a calibrated probability that the candidate is the person you are looking for.
Keep a performed comparison with no match separate from a comparison that failed or could not be completed. An unavailable result is not evidence that the person has no profile. The model-evaluation guide explains what to ask about accuracy claims and missing coverage.
Check each candidate against its context
Review one candidate at a time. Identify the app and the evidence shown, then check any capture date and surrounding profile details. Preserve discrepancies. If the photo resembles the reference but the available context conflicts, the resemblance does not erase that conflict. Several copies of one image also do not provide several independent confirmations of ownership.
A capture date records collection of the evidence. It does not establish that the account owner was using the app when you opened the report. Treat absent dates as absent rather than substituting the date of purchase or the time a screenshot was forwarded to you.
Use sample reports to understand the presentation of evidence and limitations. Samples illustrate format; they do not promise that a particular search will return equivalent material or a match.
Stop at the boundary of the evidence
Before another upload or search, identify the new information it would add. A corrected input or a different usable reference may justify revisiting an input issue. Repeating the same comparison against unchanged evidence is not independent corroboration, and extra spending does not turn a tentative attribution into certainty.
Review AI privacy questions before sharing more images. You may decide to pause, ask a bounded question, or leave the result inconclusive. Photo matching can help organize a review, but the limits of the underlying evidence remain part of every outcome.