By Aryan Gupta · October 19, 2026
An AI humanizer is a website that rewrites text from ChatGPT or another chatbot so that AI detectors will think a person wrote it. Many students use these tools, and regular AI detectors often miss the text they produce, since that is what humanizers are designed to do.
However, humanizers usually leave signs behind. I made Optimal to look for these signs, and while I was testing it on real humanizers, I noticed the same ones appearing many times. This guide explains what they are and how to check for them.
Some humanizers place invisible characters, such as zero-width spaces, inside words. These characters do not show up on the screen, but they split the words apart so that a detector reads them incorrectly. Normal typing does not create them.
How to check: Use Ctrl+F to search for a word that you can see in the essay. If the search does not find the word, there is most likely a hidden character inside it. Optimal also shows each hidden character as a small diamond.
Another method is to replace English letters with letters from the Russian or Greek alphabet that look the same. For example, the Russian "а" and the English "a" look identical, but a computer treats them as different letters.
How to check: Spell check will often underline these words even though they appear to be spelled correctly, and Ctrl+F will not find them. There is almost no reason for these letters to appear in an English essay unless they were added on purpose.
Older humanizers and paraphrasing tools replace words with synonyms without understanding the sentence. This results in phrases such as "counterfeit consciousness" instead of "artificial intelligence" or "understudies" instead of "students." A student would not normally write these phrases, because they do not mean what the student intended to say.
Some newer humanizers do the opposite. Instead of making the writing more complex, they add small mistakes so the text looks like a student wrote it. These are some examples I found in real humanizer output:
The main sign is that the mistakes do not match the rest of the writing. Students make mistakes too, but their writing also includes their own ideas, specific details, and opinions. If an essay reads like a general textbook summary but has a few random grammar errors, those errors were likely added by a tool.
Some humanizers split sentences into very short pieces so the writing seems simpler. One humanizer I tested changed "You can play it alone or with a partner" into "You can play it alone. With a partner." The second part is no longer a complete sentence, and the word "or" was removed. When an entire essay is written this way but still covers the topic completely, it is a strong sign that a humanizer was used.
Humanizers also add short personal statements to make the writing sound more human, such as "I love how volleyball pushes you" or "I have seen many friends grow through this sport." These statements are usually vague. They do not describe anything specific, and the rest of the essay does not mention them again.
ChatGPT sometimes uses a special non-breaking hyphen in words like "self-expression." It looks the same as a normal hyphen, but it is a different character that is not on a keyboard. Some humanizers leave it in the text, so if it appears in an essay, the text was probably copied from a chatbot at some point. Optimal checks for this character as well.
None of these signs proves anything on its own, and I would not accuse a student of cheating based on one of them or on a score from Optimal or any other detector. A better next step is to look at the student's earlier drafts (Google Docs saves a version history), compare the essay to their other work, or ask them to explain what they wrote. A student who wrote the essay should be able to discuss it without much trouble.
To check an essay for all of these signs at once, you can paste it into Optimal for free. It highlights everything it finds so you can review the evidence yourself. You can also see how it performed against real humanizers on the Test Results page, or read Do AI Humanizers Actually Work?