What is an AI detector?
An AI detector is a tool that estimates whether a text was written by AI or by a person. You paste the text or upload a file. The tool gives a score, often a percentage, and it may mark the sentences it doubts. The score is an estimate, and it can be wrong.
By Miles FosterFigures updated

Key facts
- Catches AI text: 10 of the 10 detectors I tested got a high grade for catching AI writing.
- Can be wrong: 2 of the 10 got a low grade because they marked human writing as AI.
- Not proof: a score is an estimate, never proof against a person.
How does an AI detector work?
You give it a text, and it gives you a score that estimates if AI wrote it. It does not know who wrote the text.
It works in four steps.
You paste a text or upload a file
You copy the text into a box on the page. Some detectors also read files and web pages.
The detector reads the text
It looks at how the text is written and compares that with patterns that are common in AI writing. It cannot see how the text was made.
You get a score
Most detectors show a percentage and a label, such as AI or human. Some show how sure they are. Most of the ones I tested did not.
You read the marked sentences
Many detectors mark the sentences they doubt. This tells you more than the score alone.
How accurate is an AI detector?
Good at catching AI text, in my tests. But some detectors marked real human writing as AI.
I wanted to count the mistakes, not guess them. So I labelled every test text as AI or human before I opened any detector. I pasted each text once and kept the first verdict.
Catching AI text was the easy part. 10 of the 10 detectors got a high grade for it.
Human text was harder. I used passages from old books, written long before AI could write. If a detector marks one of them as AI, that can only be a mistake. 2 of the 10 detectors got a low grade here. One detector marked every human passage as AI.
One note on these numbers. I tested on free plans and trials. Some of them allowed the full set of texts, and some allowed only a few scans.
AI detector vs plagiarism checker
A plagiarism checker looks for copied text. An AI detector looks for text written by AI. They answer two different questions.
AI detector
- Asks: did AI write this?
- Gives an estimate, often a percentage
- Has no source to show you
- Can mark human writing by mistake
Plagiarism checker
- Asks: was this copied?
- Compares your text with other texts
- Can show the text that matches
- Does not tell you if AI wrote it
Some tools do both jobs in one account.
What is an AI detector good at, and bad at?
It is fast, and it catches plain AI text. It is weaker on mixed text, and it rarely explains its score.
I have tested 10 AI detectors, and the same story repeated.
The good part was speed. 9 of the 10 detectors got a high grade for it. A verdict often came in a few seconds.
Mixed text was harder. I joined a human paragraph to an AI paragraph in one text. Several detectors still called the whole text human in the main score. The marked sentences were often more correct than the score.
The score itself was the other weak point. 4 of the 10 detectors got a high grade for showing clear evidence. Most gave no reason for a flag and no margin of error. Other languages were mixed. Several detectors judged my Spanish, French and German texts correctly. One got both human texts in other languages wrong.
So the lesson is simple: use the score as a first look, and read the marked sentences yourself. You can see how I test on the AI detector methodology page.
Who should use an AI detector?
People who want a quick first look at whether a text reads as written by a machine: teachers, editors and writers who check their own work.
Use it as one sign among others. A high score is a reason to read the text closely and to talk to the writer. It is never proof against a person.
It is not the tool for finding copied text. For that, people use a plagiarism checker.
How much does an AI detector cost?
From $5 a month. A typical price is $10.50 a month. Many detectors also have a free plan.
I check the prices of 10 detectors every day. Here is the cheapest plan of each one.
| Tool | Cheapest price |
|---|---|
| Winston AIHas a free plan | $5 |
| AI or NotHas a free plan | $5 |
| Undetectable AI | $5 |
| Decopy AI DetectorHas a free plan | $6.99 |
| ZeroGPTHas a free plan | $8.99 |
| WriteHumanHas a free plan | $12 |
| Originality.aiHas a free plan | $12.95 |
| SolvelyHas a free plan | $12.99 |
| QuetextHas a free plan | $14.99 |
| SaplingHas a free plan | $25 |
Tool
- Winston AIHas a free plan
- Cheapest price
- $5
- AI or NotHas a free plan
- Cheapest price
- $5
- Undetectable AI
- Cheapest price
- $5
- Decopy AI DetectorHas a free plan
- Cheapest price
- $6.99
- ZeroGPTHas a free plan
- Cheapest price
- $8.99
- WriteHumanHas a free plan
- Cheapest price
- $12
- Originality.aiHas a free plan
- Cheapest price
- $12.95
- SolvelyHas a free plan
- Cheapest price
- $12.99
- QuetextHas a free plan
- Cheapest price
- $14.99
- SaplingHas a free plan
- Cheapest price
- $25
A listed price is the lowest price shown on the pricing page, which is often for a yearly or longer term. Each price links to its full history.
Free plans differ a lot. In my tests, some had no limit that I could reach, and some allowed only a few scans a month. Paid plans usually count words or credits.
Prices change often, so check the price again before you pay. You can see every change on the price tracker.
How popular are AI detectors?
Very popular: people search Google for "ai detector" 13,600,000 times a month.
I also count visits. The detector sites I track got 25.5 million visits from Google in October 2026. That is 31.3% more than a year earlier.
WorldVisits from Google to AI detectors, worldwide
Visits from Google search each month, to October 2026.
ZeroGPT, GPTZero and Clever AI Detector get the most visits. MyDetector went from 10,834 visits to 230,684 in one year.
I follow these numbers every month on the AI detector statistics page.
How to choose an AI detector
Do these five checks before you pay.
Test it on human text first
Paste something you know a person wrote. Some detectors marked human writing as AI.
Test it in your language
One detector got both human texts in other languages wrong.
Read the marked sentences
Do not stop at the percentage. The marked sentences show what the detector doubts.
Look for a reason behind the score
A good report explains a flag or gives a margin of error. Most did not.
Check the free plan
Find out how many scans or words you get before you plan work around it.
Which AI detectors have I tested?
10 detectors so far. Every one got the same kinds of tests.
You can see them in order, with my grades, on the best AI detectors page. Every tool I know is on the list of AI detectors.
AI detector FAQ
Can an AI detector be wrong?
Yes. In my tests, some detectors marked real human writing as AI. A score is an estimate.
Can an AI detector prove that someone used AI?
No. It gives an estimate, not proof. Use it as a first look, and never as the only reason to accuse a person.
Are AI detectors free?
Many have a free plan. 9 of the 10 detectors whose prices I track have one. You can find more on the list of free AI detectors.
Is an AI detector the same as a plagiarism checker?
No. A plagiarism checker looks for copied text. An AI detector estimates whether AI wrote the text.