Compare the two reports claim by claim, not paragraph by paragraph. For each claim, record the original input, the accessible public sources named or findable, and whether the claim stays supportable when you hold those sources steady. Treat a reworded sentence and a newly supported sentence as different events: the first is a wording change, the second is a change in evidence. Keep one control source unchanged between versions so you can tell what the reports actually added from what they only repeated.
Find public social profile clues with our TraceSeek app.
Sort claims before you compare reports
Start by breaking each permitted report into individual claims you can check: identity or role, stated work, stated location, stated relationships, and any interpretation about income or dating. Give each claim a short identifier of your own, so the same claim can be followed across both versions. This is editorial bookkeeping you own, not a feature of any app, and it keeps you from drifting into a paragraph-level impression that one report sounds stronger than the other.
- Copy only what you are permitted to keep; if you may not retain the report text, summarise each claim in your own words and note which report version it came from.
Record the original input exactly
For each version, write down what was actually submitted: the public username or the photo, plus a plain note about any OCR text you expect it to carry. The product owner says username input searches matching publicly accessible social profiles with non-exhaustive coverage, and photo input uses visible OCR text and image similarity to look for public social pages and image appearances. Those input differences matter because a changed photo or username can change the source set, so a changed claim may reflect a changed input rather than better evidence. [TraceSeek: product overview and operator-confirmed capabilities]
Use one stable source as your control
Pick one public artifact that is genuinely accessible, relevant, and unchanged between the two reports, such as a specific public profile page or a single public post. Mark it as your control. When a claim appears in both reports, check whether your control source alone supports it. If it does, the second version has not added confirmation, only repetition. Generative AI tools can present content confidently even when it is inconsistent with inputs, so verification of sources rather than the fluency of the report is the discipline that matters here. [NIST AI 600-1: generative AI confabulation and source verification]
Separate a wording change from a source change
For each claim, record two separate fields: what the report says, and which public source appears to support it. A reworded claim with the same single source is a wording change. A claim is a source change only when a distinct public artifact, one you can still reach and inspect, supports it. AI analysis of found public profile content, including interpretations about work, income and relationships, is described by the operator as unverified inference rather than established fact, so a sharper sentence is not automatically a better-supported one. [TraceSeek: product overview and operator-confirmed capabilities]
| observation | limit | next step |
|---|---|---|
| The claim appears in both versions with new phrasing | Same single source; repetition is not independent confirmation | Keep it labelled one-source and do not raise confidence |
| The claim newly names a second public artifact | Source may be new, but reachability and relevance still need checking | Open the artifact; if it is accessible and distinct, record it as a second source |
| The claim mentions income, relationships or dating | Operator describes these as uncertain interpretations of public content | Treat as an AI guess unless an explicit public statement says it |
| The photo input changed between versions | Different OCR text or image-similarity matches may produce different claims | Rebuild the source set for the new input; do not compare across it as if unchanged |
Keep the limits visible in the same note
Next to each claim, write the precise limit that applies. Useful limits include: the source was not reachable; the profile matched by username but ownership is unverified; the photo matched by similarity with no identity verification; the source is private and therefore outside scope; or the claim is an AI interpretation with no explicit public statement behind it. Deep search widens scope across accessible public sources but does not bypass private accounts or guarantee completeness, so an absent result stays an absence, not a contradiction. [TraceSeek: product overview and operator-confirmed capabilities]
Stop where the evidence stops
If a source is missing, private, mismatched or ambiguous, record that exact condition and stop the inference for that claim. Do not try to work around a private account, do not treat a matching photo as identification, and do not treat a badge, status or silence as proof of intent or safety. Where the report offers interpretations about money, relationships or private interactions, mark them as unsupported guesses unless a public statement by the person makes the claim explicit. This keeps the comparison honest without turning it into a judgement about anyone. [NIST AI 600-1: generative AI confabulation and source verification]
A couple more questions
The second report phrases the same claim more confidently. Is that new evidence?
No. Confidence in wording is not source support. Record the claim, check which accessible public artifact backs it, and if that artifact is unchanged, keep the claim labelled one-source regardless of how assertively it is written.
The second report names a public profile the first did not. Does that confirm the claim?
Only after you check it. Open the named public artifact, confirm you can reach it, and judge whether it actually supports the claim rather than merely sharing a name or image. Username matches have non-exhaustive coverage and do not verify identity or current activity, so a new name is a lead to inspect, not confirmation.
What do I do when a claim rests on income or relationship interpretation?
Treat it as an AI inference, not a personal fact. The operator describes income, social relationship, dating and platform-interaction interpretations as uncertain readings of public content. Record it as a guess, look for an explicit public statement if one exists, and otherwise leave the claim unsupported.
How should I record a claim whose source I cannot reach?
Write the precise limit, such as private, removed, ambiguous or mismatched, and stop. Do not attempt private access and do not infer the claim from the report alone; an inaccessible source leaves the claim unconfirmed in that version.
Sources & editorial notes
Official references checked on October 24, 2026. Checklists, comparison tables and example wording are DateClues editorial tools, not official tests or validated risk scores.
- NIST AI 600-1: generative AI confabulation and source verification ↗
- TraceSeek: product overview and operator-confirmed capabilities ↗
General information. No guide or tool can certify a person’s identity, intentions or safety. Read our editorial standards.