AI detection tools and the Seal of Humanity solve the same problem — verifying whether text was written by a human — but they take fundamentally opposite approaches. AI detectors analyze finished text and estimate the probability it was AI-generated. The Seal of Humanity captures evidence of human writing at the moment of creation and produces a cryptographic proof that anyone can verify. One guesses. The other proves.
The distinction matters because AI detection has a documented accuracy problem. Research shows false positive rates of 20-60% depending on the tool and the writer's background. The Seal of Humanity uses SHA-256 cryptographic hashing with near-zero false positive rates because verification is deterministic, not probabilistic.
How AI Detection Works
AI detectors like GPTZero, Turnitin's AI Writing Detection, Originality.ai, and ZeroGPT analyze text using two primary metrics:
Perplexity measures how predictable the text is. AI-generated text tends to choose statistically likely word sequences, resulting in lower perplexity. Human writing is more varied and unpredictable.
Burstiness measures variation in sentence structure. Humans naturally alternate between short and long sentences, simple and complex structures. AI output tends to be more uniform.
Some tools add additional signals — vocabulary diversity, sentence-level analysis, or pattern matching against known AI model outputs. But all share the same fundamental approach: statistical analysis of the finished text.
The False Positive Problem
The statistical approach has a well-documented weakness: it misidentifies human writing as AI-generated at alarming rates.
- A 2023 Stanford study tested AI detectors on TOEFL essays written by real international students. Over 60% were incorrectly flagged as AI-generated.
- GPTZero, one of the most widely used detectors, has acknowledged that its results should not be used as the sole basis for academic integrity decisions.
- Turnitin's AI detection system assigns percentage scores that can fluctuate between scans of the same text.
- Research from the University of Maryland found that AI detectors consistently score non-native English writing as more likely to be AI-generated, regardless of actual authorship.
The false positive problem is not a bug that will be fixed. It is inherent to the approach. As AI models improve and produce more human-like text, the statistical boundary between human and AI writing narrows. Detection becomes harder, not easier.
For a detailed analysis of specific tools and their limitations, see AI Content Detection Tools Compared.
How the Seal of Humanity Works
The Seal of Humanity takes the opposite approach. Instead of analyzing finished text and guessing its origin, it captures evidence of human writing at the source.
For handwritten and typewritten pages: When you photograph a handwritten or typewritten page with LyteWriter, the system analyzes the physical evidence — ink irregularities, pressure variations, paper texture, handwriting characteristics, and typewriter mechanical patterns. This physical evidence, combined with the extracted text, is processed through SHA-256 cryptographic hashing to produce a unique verification code.
For digitally typed text: When you write in the LyteWriter editor, the system captures keystroke dynamics — typing speed, rhythm, pause patterns between words and sentences, correction behaviors, and revision sequences. These biometric signals distinguish human composition from automated text generation.
Verification: Anyone can verify a sealed document at lytewriter.com/verify by entering the verification code and pasting the text. The system compares the text against the stored cryptographic hash. No account is needed to verify. The original text is never stored on LyteWriter's servers.
Direct Comparison
| Aspect | AI Detection | Seal of Humanity |
|---|---|---|
| Approach | Analyzes text, guesses origin | Captures evidence, proves origin |
| Method | Statistical analysis (perplexity, burstiness) | Physical evidence + cryptographic hashing (SHA-256) |
| False positive rate | 20-60% (documented research) | Near zero (deterministic verification) |
| Non-native speaker bias | Yes (higher false positive rates) | No (based on physical/behavioral evidence, not language patterns) |
| Verification type | Probabilistic score (e.g., "87% likely AI") | Deterministic (match or no match) |
| Who can verify | The person running the tool | Anyone, publicly, without an account |
| Privacy | Text uploaded to third-party servers | Text never stored; hash-based verification |
| Forgery resistance | AI can be tuned to evade detectors | SHA-256 is computationally infeasible to forge |
| Consistency | Results can vary between scans | Same result every time |
| Cost | $10-100+/month for reliable tools | Free tier available (10 scans/month) |
When AI Detection Makes Sense
AI detection is not useless. It has legitimate applications as a screening tool:
- Initial triage. In environments processing large volumes of submissions (admissions offices, content platforms), detection scores can flag submissions for human review. The key is that flagging triggers review, not punishment.
- Self-checking. Writers can use detectors to see whether their text resembles AI output and adjust their style if desired.
- Trend monitoring. Organizations can use aggregate detection data to understand how much AI-generated content they are receiving over time.
The problem arises when detection scores are treated as verdicts. A detection score is an estimate, not evidence. Using it to accuse someone of cheating is like using a lie detector result in court — suggestive, but not proof.
When Proof-of-Humanity Is Better
Cryptographic verification is the stronger choice when:
- Someone's reputation is at stake. Academic integrity cases, publishing disputes, or legal proceedings require verifiable evidence, not probability estimates.
- The writer is a non-native English speaker. Detection tools are biased against non-native speakers. Cryptographic verification has no language bias.
- Independent verification is needed. Detection requires running the tool yourself. The Seal of Humanity can be verified by anyone with the code.
- The text will be scrutinized multiple times. Detection scores can vary between scans. Cryptographic verification produces the same result every time.
- Long-term proof is required. A detection score from today may not match a scan tomorrow as tools update their models. A cryptographic hash is permanent.
The Arms Race Problem
AI detection and AI generation are locked in an arms race. As detection tools improve, AI models are fine-tuned to evade detection. As evasion techniques advance, detectors add new signals. This cycle ensures that detection accuracy will fluctuate and never reach certainty.
The Seal of Humanity sidesteps this arms race entirely. It does not try to distinguish AI text from human text by analyzing the text itself. Instead, it documents the physical or behavioral evidence of human writing — evidence that exists independently of what the text looks like. A handwritten page photographed and sealed today will remain verifiable regardless of how AI text generation evolves.
For a broader analysis of why authentication will replace detection, see AI Detection Is Broken and Not by AI vs Seal of Humanity.
Getting Started
If you currently rely on AI detection to verify human authorship, consider adding cryptographic verification as a complement or replacement:
- For handwritten or typewritten work, photograph your pages with LyteWriter
- For digital writing, use the LyteWriter editor with keystroke dynamics capture
- Share verification codes with anyone who needs to confirm your authorship
- Download proof files for formal submissions
All features, including the Seal of Humanity, are available on the free plan — 10 scans per month, no credit card required.