Ideas AI Agents AI Use Cases AI Tools Reports Library About Extract Transcript

The AI Analysis Paralysis

AI Detector Tools promise to be able to distinct AI generated texts. Turns out that they can report original work as AI! To make it worse the AI to human rewriters missed facts and hallucinated new ones. Optimizing content for passing detector tests is a wrong goal that might not get you engaging, factual content.

Results

ZeroGPT rated text generated by ChatGPT at 26.5% AI and declared it “likely human written.” Rewriting it with Both humanized versions of that same text scored higher.

Humanizing the text made it look more artificial, and the raw model output passed as mostly human. Whatever these tools do or fail to do, this detector is not the instrument that can tell you. The most reliable tools from the pack appear to be UndetectedGPT following closely with Originality.

The premise

Why does the AI generated text sound so unnatural?

The models are trained afterall on human text, why is it still easy to spot the difference between a human writer and LLM?

Language models predict new tokens and words based the training data producing of can be described as a least surprising possible sentence. The human brain and writing doesn’t work in the same way, the writing can contain a distinct style of the author, unpredictable, nuanced use of words and phrases. Or sometimes it could be just simpler than elaborate LLM creations.

Wikipedia's WikiProject AI Cleanup, which has spent years on removing AI Slop stated:

“LLMs use statistical algorithms to guess what should come next. The result tends toward the most statistically likely result that applies to the widest variety of cases.”

Why it matters

Why does it matter?

People are getting tired of ‘AI Slop’. Companies followed it, and rank AI content lower or explicitly prohibit it. Google new guidance seem to fight exactly against mass produced low engagement pages. It might be easier than ever produce content but it doesn’t mean that it will be engaging. That’s why there is entire suite of tools created to make the AI text look like a real person wrote it.

Mechanism

How the tools actually work

We tested two different tools and each of them has distinct approach. The first one Lynote, choose a very interesting destructive approach using translation passes to destroy AI fingerprint. The Humaniify uses prompt and tries to erase 35 LLM patterns recognized by Wikipedia

Approach one

Multiple Translations Chaining

AI text carries measurable fingerprints: predictable token distributions, uniform sentence lengths, repeating n-grams. Machine translation rewrites text at the grammar level, so word order, morphology and vocabulary all shift. Chain enough hops through distant languages and the fingerprint does not survive. The issue is if the sense and message can survive the passes.

Passes: 4 (2 LLM rewrites, 2 machine translations)
Cost: ~$0.0002 and 58 seconds per run
Needs: An LLM key plus two translation services
Sees source: Only at step one

Approach two

Pattern rewriting

Humanify is not a library but a Markdown file skill that can be loaded by any agent. Its trying to remove 35 named patterns drawn from Wikipedia's Signs of AI writing in one pass, with adding self critique in the same single pass that keeps the original in view the whole time.

Passes: 1, plus a self-critique before the final draft
Cost: One agent pass, no API key
Needs: Nothing. No network calls, no dependencies
Sees source: Throughout

The four passes, in order

The chain is genuinely sequential, and each hop is doing something different. Steps one and two are not translations at all, they are rewrites that happen to change language. Steps three and four are pure machine translation.

Standard Pipeline — Five steps: two LLM rewrite passes (the second carries the first as conversation history) followed by two NMT hops across translation engines. The translation chain routes through Chinese → Japanese → Finnish before returning to English, maximizing linguistic distance at each hop so that no single engine's structural fingerprint survives. The original library using Google Translate and Niutrans but we opt in for two Google Translate passes as it’s harder to for us to register to an Asian service.

STEP 01

EN → ZH

LLM rewrite

Instructed to strip the “AI flavor” while translating into Chinese. The last time the chain sees your original.

STEP 02

ZH → JA

LLM rewrite

Same instruction into Japanese, with step one's input and output fed back as few-shot history.

STEP 03

JA → FI

Google Translate

Finnish is agglutinative. Meaning rides in stacked suffixes, not word order, so the structure has to be rebuilt.

STEP 04

FI → EN

Niutrans by default but we changed it to Google Translate

A second vendor, deliberately. Reconstructing English from Finnish forces structural change rather than word swaps.

Tinted steps are LLM rewrites; plain steps are machine translation. German and Korean are the other tested intermediate languages.

What the rewrite prompt says

Both LLM steps share one system prompt and one instruction. Translated:

system → 你是一个专业的文案改写专家，精通多语言本地化。

“You are a professional copywriting rewrite expert, skilled in multilingual localization.”

user → 翻译为{语言}，去掉 AI 味道，拟人化改写，只输出结果：

“Translate to {language}, remove the AI flavor, rewrite to sound human, output only the result:”

Here is the problem, look what is missing. No instruction to preserve facts. No reference to the source after step one. No verification pass. The pipeline is built to defeat a classifier, and it is honest about that.

The test

Finding text that a detector agrees a human wrote

AI space has a lot of moving elements, one of the hardest one was establishing a baseline. We decided to use fragment from the book written by me, the author of this report, before LLM produced content become so popular. We were surprised to discover that it was marked as partially generated by the AI!

Three detectors, four conditions: the human baseline, raw ChatGPT output, and that same ChatGPT text after each humanizer.

Condition	ZeroGPT	Originality.ai	UndetectedGPT
A paragraph from the book	3% AI	under 15%	0% AI
ChatGPT generated text	26.5% called human	exceeds 15% flagged	Detected 100% flagged
After Humanify	29.1%	25% or less	0% credits out
After Lynote	31.7%	25% or less	0% partial

ZeroGPT score, by condition

Higher = more “AI”

Human baselineknown author

3.0%

Raw ChatGPTno humanizing

26.5%

After Humanifyhumanized

29.1%

After Lynotehumanized

31.7%

The ordering is backwards. If the score measured what it claims to, humanizing should move text down toward the baseline. Instead both tools moved it up, past the raw model output that ZeroGPT had already waved through as human.

Originality.ai was the only detector that behaved consistently: it flagged the raw ChatGPT text and reported both humanized versions inside its lower band. That is the result the tools are sold on. But its output is a coarse threshold, not a number, so a move from “exceeds 15%” to “25% or less” is a smaller claim than it appears.

What this test cannot support

One text per condition. These are readings, not a sample, and nothing here is statistically significant.
UndetectedGPT ran out of credits mid-test. Worth noting though it accurately flagged book chapter as human, and raw Chat GPT output as 100% AI.
Originality.ai reports bands rather than values, so its numbers cannot be compared with ZeroGPT's percentages.
Detector models change without notice. These readings describe one afternoon.
The tool labelled Humanify above is recorded as it was reported and has not been confirmed against the pattern-rewriting skill benchmarked in the next section.

The blind spot

What no detector was checking

While the detectors disagreed about classifiying the origin of the text a separate test ran on meaning. The same source passage, roughly 1,900 characters of financial writing, went through both tools and every checkable claim was compared against the original.

Fourteen facts: the return figure, the market capitalization, the price target, four named entities, three dates, two more numbers. The pattern-rewriting skill kept all fourteen. The translation chain kept thirteen, and changed three things about what the text asserts.

every millionaire in the world to own one full coin every billionaire in the world

Factual error. Moves the scarcity argument by three orders of magnitude.

long-term value could rise substantially long-term price increases are inevitable

Hedge deleted. A speculative claim became a certainty, in financial copy.

(nothing in the source) “It seems better to be prepared for twists and turns.”

Sentence invented outright.

This is structural, not bad luck. After step one the chain never sees the original again. Each hop rewrites the previous hop's output, so errors compound and no stage has the information needed to notice. By step four the English is four translations from the source.

The pattern-rewriting skill avoids this for one reason: the source stays in context, and its process explicitly asks, before the final draft, whether any fact, name, number, date, quote or citation was added or lost. That question is only answerable if the original is still there.

Both tools cleaned up the surface tells equally well. Both removed every em dash and every curly quote. The difference showed up only where no detector looks.

Conclusions

What holds up

Detector scores are not a measurement

ZeroGPT passed raw ChatGPT output and then rated both humanized versions as more artificial. A number that moves the wrong way under a known intervention is not tracking what it claims to track. Do not tune content against it.

The detectors disagree with each other

On identical input, one flagged it, one missed it, and one ran out of credits. Any single score is a reading from one instrument on one day, not a verdict.

Translation chaining is unsafe for factual copy

It works as advertised on style and it is cheap. It also silently changed a number, removed a hedge, and invented a sentence, in a passage about money. For finance, medicine, law or technical documentation, that cost is not recoverable by a proofread that does not compare against the source.

The cheapest tool was the most faithful

The pattern-rewriting approach needs no key, no network and no vendor. It kept every fact and cut nearly 300 characters of filler. Its ceiling is lower, because it edits rather than rebuilds, but it does not trade accuracy for a lower score.

Optimizing for the detector is the wrong target anyway

Search engines penalize low-quality, low-engagement content, not text that trips a classifier. Those are different problems. A page that reads as though someone thought about it solves the one that pays.

Full report, findings and the repo used to run all this test can be found at: github.com/gulfcoastdev/HumanizeIt