Can Claude Humanize Text? We Tested Anthropic's AI Against 3 Detectors

Claude writes better than most AI models. It still cannot humanize text past detectors.

Claude has a reputation for producing more natural-sounding text than ChatGPT. That reputation is deserved — Claude's output has more sentence variation and fewer obvious AI tells like "Furthermore" and "It's important to note."

But "more natural sounding" is not the same as "undetectable." We tested whether Claude can humanize text to pass AI detectors, and the answer is no — though it gets closer than ChatGPT does.

---

Raw detectability: Claude vs ChatGPT

Before testing humanization, we compared how detectable each model's raw output is.

We generated 2,000 words on the same topic (the history of urban planning) using both models, then ran each through three detectors.

Metric	Claude 3.5 Sonnet	ChatGPT-4o
GPTZero score	82% AI	96% AI
Turnitin score	79% AI	94% AI
Originality.ai score	77% AI	92% AI
Average sentence length variation	38%	12%
Transition word density	2.1%	4.8%

Claude is harder to detect. Its sentence lengths vary more, and it uses fewer formulaic transitions. But 82% on GPTZero is still a failing score.

---

Can Claude humanize its own text?

We asked Claude to rewrite its own output using 5 different prompts:

Prompt	GPTZero	Turnitin	Originality.ai
"Rewrite to sound human"	58% AI	55% AI	52% AI
"Vary sentence structure, add personality"	52% AI	49% AI	47% AI
"Rewrite as informal first-person essay"	48% AI	46% AI	44% AI
"Make this undetectable"	54% AI	51% AI	49% AI
"Rewrite each paragraph with different style"	50% AI	48% AI	46% AI

Best result: 48% AI on GPTZero. Better than ChatGPT's best of 61%, but still a failing score.

---

Can Claude rewrite ChatGPT text?

A common trick: generate with ChatGPT, rewrite with Claude. We tested it.

Method	GPTZero	Turnitin
Raw ChatGPT text	96% AI	94% AI
ChatGPT rewrites ChatGPT	61% AI	58% AI
Claude rewrites ChatGPT	44% AI	41% AI
Claude rewrites Claude	48% AI	46% AI
Humanize AI Pro	2% AI	3% AI

Cross-model rewriting helps. Claude rewriting ChatGPT (44%) is better than ChatGPT rewriting itself (61%). But it is still not enough.

---

Why Claude is better but still fails

Claude's advantage comes from Anthropic's training approach. The model produces more sentence length variation (38% vs ChatGPT's 12%), fewer filler transitions, and better paragraph-level variety.

But the fundamental limitation is the same: Claude is still a language model predicting tokens. Its predictions are more varied, but they are still predictions. Detectors measure the statistical distribution of word choices across the entire document, and that distribution is still detectably non-human.

---

The humanization advantage with Claude text

Claude text humanizes exceptionally well because the starting quality is higher.

Source Text	Humanize AI Pro Output Score (GPTZero)	Quality Rating
ChatGPT-4o text, humanized	2% AI	Good
Claude 3.5 Sonnet text, humanized	3% AI	Excellent
Gemini 1.5 Pro text, humanized	4% AI	Good

The scores are similar after humanization, but the readability of humanized Claude text is consistently better. Claude's natural sentence variation gives the humanizer more to work with.

If you use Claude as your primary AI writing tool:

1. Write with Claude (better starting quality)
2. Humanize with Humanize AI Pro (3 seconds)
3. Verify with a detector (should be under 5%)
4. Light editing pass (usually minimal with Claude-origin text)

---

Bottom line

Claude writes more naturally than ChatGPT and is harder to detect in raw form. But it cannot humanize text past detection thresholds, even with careful prompting.

Use Claude for writing. Use Humanize AI Pro for humanization. Claude text plus humanization produces the highest quality undetectable output we have seen in our testing.