couldthisbetrue.com

What It Is

Could This Be True? is a forensic detection tool for verifying AI-generated content across text, images, audio, and video. Its core selling point is not a single black-box score like “87% AI,” but a breakdown of each judgment into auditable signals, residual maps, confidence intervals, and literature references. The page is labeled Live v0.1, indicating that it is still at an early stage.

Core Capabilities

On the image side, it offers ELA, FFT spectrum analysis, RGB/YCbCr channel decomposition, PRNU noise residuals, Grad-CAM attention, and C2PA manifest reading. For text, it analyzes sentence-length burstiness, lexical habits, n-gram repetition, sentence-opening variation, punctuation patterns, and word-by-word predictability heatmaps. For audio, it emphasizes watermark-first detection, spectral forensics, and codec-aware analysis. For video, it covers frame sampling, motion residuals, and keyframe analysis. Its methods are tied to academic papers or open standards, making it suitable for media verification, education, moderation, and investigative scenarios where an explainable evidence chain is required.

Pricing and Trial

The free tier is permanently available: anonymous users get 3 checks per day, while logged-in users get 10 checks per day, with no credit card required. Pro costs $19/month and includes unlimited web verification, API access, JSON export, watermark-free reports, history, files up to 100MB, and priority email support; the API is marked as capped at 5,000 requests per day. Lifetime costs a one-time $199. The Enterprise plan supports self-hosting, SOC 2/GDPR documentation, and custom integrations, but pricing is not disclosed.

Pros and Cons

Its strengths are browser-side processing by default, zero uploads, and a clear privacy commitment. Results are visualized and citable, which makes them useful for review and communication. It also supports C2PA provenance, helping distinguish “who generated/edited this” from “whether it appears synthetic.” The limitations are also explicit: detection is statistical inference, not proof. Short text, heavily edited content, outputs from the latest models, heavy compression, blur, or adversarial perturbations can all reduce reliability. Some audio/video capabilities are described on the page as operational, while the About section lists them as coming soon, so their maturity needs to be verified in practice.

Who It’s For and Access from China

It is suitable for journalists, teachers, trust and safety teams, fraud investigators, and content moderation pipelines—especially for users who cannot rely on black-box scores and need to present evidence. Individual users can start with the free tier; organizations needing API access or batch workflows may consider Pro or Enterprise. Payments are handled by Stripe, with no mention of Alipay, WeChat Pay, or RMB billing. The scraped text does not state whether the service is accessible from mainland China, so network reachability and payment success rates need real-world testing. Alternatives to consider include GPTZero, Originality, Hive, Google SynthID Detector, Resemble, and C2PA/Content Credentials tools.