Travel-scam evidence image review

Business value

• Protects editorial credibility on the most-shared content type in travel publishing: scam exposés and victim stories.
• Catches the user-submitted 'photo of the scam' that's actually a generated image submitted to bait a publication into citing it.
• Documents the verification step for editorial liability — when a publisher cites visual evidence, the verification trail matters.

Agent job to be done

Be an editorial fact-checker for visual evidence. Allow photos with clear provenance and authentic-photography signals. Hold ambiguous submissions for source verification. Reject when synthetic cues make the image unusable as evidence.

format: articleintended_use: citedomain: travel scam visual evidence

When to call VeracityAPI

On any reader-submitted scam photo, social-media-sourced grab, or user-generated visual that will be cited in published editorial.

What image URL to submit

Public HTTPS image URL for the photo as it will appear in the published guide. For screenshots, the resolution and platform-chrome details matter — submit the version closest to the original capture.

Decision policy

• allow: low risk AND identifiable source (named submitter, named location, traceable context).
• revise: medium risk OR anonymized source — request a higher-resolution version, request the original submission email/upload metadata, or use a verified replacement photo.
• human_review: high risk OR photos used to support specific accusations against a named business/individual.
• reject: visible heavy manipulation, OR photos that would expose the publication to defamation claims if they turn out to be generated.

Automation recipe

• Editorial worker receives the photo from the submission pipeline.
• Score with intended_use=cite.
• Allow-tagged photos go to layout with the source caption.
• Held photos route to the editor with the evidence categories pinned and a 'request source verification' button.
• Maintain a verified-source registry: photos with completed verification get a notation in the publication record.

Evidence spans agents should inspect

• 'synthetic_signage' — generated street signs, storefronts, or printed materials with characteristic generation artifacts
• 'composite_screenshot' — UI elements that don't match the platform's actual rendering
• 'document_inconsistency' — ticket stubs, receipts, or document photos with composition cues that don't match real-world printing
• 'recycled_imagery' — photos that appear in unrelated articles or stock libraries

Policy pseudocode

if (result.recommended_action === "allow") continueWorkflow();
if (result.recommended_action === "revise") rewriteWith(result.evidence, result.recommended_fixes);
if (result.recommended_action === "human_review") queueForHumanReview(result);
if (result.recommended_action === "reject") discardOrRebuild();

KPIs to track

• % of submitted scam photos held for verification before publication
• post-publication corrections involving image provenance (drive to zero)
• verification-completion rate within editorial deadline
• editor confidence on a periodic survey of fact-check workflow
• median time from submission to published-ready

What can go wrong

• Travel-safety content is high-engagement but high-litigation. The gate is one input to the verification step; legal review remains essential for content naming specific businesses or individuals.
• Screenshots of platforms (Booking.com, Airbnb, Uber confirmations) re-shared on social often pass through multiple compression cycles. 'Low confidence' is normal on heavily-recompressed screenshots.
• Submitter anonymity is sometimes the right editorial call (victim privacy, source protection). Anonymous submissions raise the verification bar; they don't necessarily mean reject.

Cost and latency notes

Image analysis is a flat $0.02 per image. The endpoint accepts HTTPS image URLs, stores no image bytes, and logs only a URL hash plus hostname. Current v0.1 latency is vision-model-bound, so preflight balance and retry carefully.