Methodology

How Humanize Me works, what Phantom does, and what it does not do.

Meet Phantom (v1.0)

Phantom is built on three principles: meaning preservation, structural rewriting, and pattern avoidance.

Rather than paraphrasing the input sentence by sentence, the approach most rewriters take, and the reason most rewrites still flag as AI, Phantom first extracts the meaning of your text into rough notes, then writes a fresh version from those notes. The model never sees the original wording during the rewrite pass, so it has nothing to paraphrase from.

Phantom also scans the input against a curated list of 60+ AI-pattern words and structures (vocabulary like "delve" / "testament" / "leverage", rule-of-three constructions, signposting phrases, slogan-y closers, em-dash overuse), and instructs the rewrite to avoid each one. A final scrubbing pass catches any that slipped through.

The result is text that reads as something a real person would write, and tests as human on the major AI checkers in our internal benchmarks.

Capabilities

• Multi-pass architecture: extract meaning → write fresh → scrub
• Detects 60+ AI vocabulary and structural patterns
• Burstiness enforcement: varied sentence lengths, real fragments
• Voice injection that adapts to your context and tone
• Self-audit pass on the highest-quality tier

Detector performance

Phantom is designed to pass major AI checkers. In our internal tests against the most common detection tools, Phantom output consistently scores as human-written. Results may vary by input, generic AI-essay topics carry detector signals no rewriter can fully strip without changing the meaning.

We test Phantom releases against the most common AI checkers, including those used by academic institutions and content platforms, on a sample of AI-soaked essays, LinkedIn posts, cover letters, and emails. Detector scores vary by input. Generic AI-essay topics carry detector signals no rewriter can fully strip without changing the meaning. Concrete content with real specifics consistently scores as human.

What Phantom does not do

• It does not verify factual claims in the text you submit.
• It does not guarantee a specific detector score on every input.
• It does not add, invent, or fabricate information not present in the source.
• It does not support academic dishonesty.

Your responsibility

You are responsible for reviewing and verifying all output before publishing or submitting it. Rewritten text may be imperfect. Always read the result before use.