Perplexity detector
What the Perplexity detector does
The Perplexity detector reads a passage and reports the patterns typical of Perplexity's answer engine, one at a time, with the reasoning attached. The output is designed to be worked through rather than glanced at: each named signal is something you can go and look at in the text.
The check runs on our AI server, so there is an allowance on it: 5 free checks every rolling 24 hours, no account, and 1 to 12,000 characters per check. Longer documents get checked in sections.
Which one you want
Perplexity the product, or perplexity the metric
Two different searches end up on this page. One is "did Perplexity AI write this", which is what the tool above answers. The other is "what is perplexity in AI detection", which is a measurement question.
As a measurement, perplexity is how predictable a passage is to a language model, and burstiness is how much its sentence lengths vary. Some detectors were built on that pair. If that is what brought you here, skip to the section on why those scores are weaker than they sound, then come back to the checklist.
How it works
How to run the Perplexity detector in three steps
Paste, analyze, work the list. There is nothing to configure.
- Step 1
- Paste a few hundred words. A single sentence carries too little pattern for the check to say anything useful about it.
- Step 2
- Press Analyze. The text goes once to our AI server and the breakdown comes back.
- Step 3
- Take each flagged item back to the passage and confirm it yourself before you act on it.
The checklist
The Perplexity signals worth checking by hand
If you would rather scan a passage yourself, this is the order to do it in. The first item settles more cases than the rest combined.
- Search for a bracket
- A stray [1] or [4] mid-sentence is citation residue from a search-grounded answer, and it is a fact about the text rather than a guess.
- Read the first sentence of each paragraph
- A reporting register announces itself there: neutral, third person, summarising rather than arguing.
- Count the attributions
- "According to", "reports suggest", "several sources note". Two or three in a short passage is a pattern.
- Look at the shape
- A digest that covers four angles evenly usually came from synthesis. A person writing to a point rarely stays that balanced.
- Check the ending
- A closing caveat about information possibly being out of date is a search-engine habit, not an essay habit.
- Measure the paragraphs
- Near-identical paragraph lengths across a whole piece is the quietest signal and the one that survives light editing.
Why it is harder
Why Perplexity text is harder to flag than ChatGPT text
Perplexity searches before it writes. Retrieved pages bring real names, dates, figures and quoted phrasing into the draft, and that borrowed specificity makes the text less uniform than a model writing from its own weights.
The practical consequence for your workflow is simple: trust the surface signals over any statistical impression. Citation residue and attribution phrasing hold up. A vague sense that the writing is too smooth does not.
The honest limit
Why perplexity and burstiness scores mislead
The metric penalises the wrong writing. Clear, plain, conventional prose is predictable by design, so tools built on the score flag well-known public-domain documents as machine-written. Writing by people working in a second language scores the same way, and that is where the false positives pile up.
There is also a mechanical problem: the major closed models do not expose the token probabilities a perplexity calculation needs, so a score computed on their output is not measuring what the name implies. This checker does not produce a perplexity number. It names patterns you can verify.
Straight answer
Can the detector prove who wrote something?
No. It reports signals, never proof of authorship. Models trained on overlapping material converge, and one instruction to answer directly without hedging strips most of the register this check reads.
So treat a strong match as a prompt to look closer. We have no affiliation with Perplexity AI and we make no prediction about what Turnitin, GPTZero, Copyleaks or Originality.ai would say about the same text.
In practice
When to reach for a Perplexity detector
Before publishing submitted copy, when a paragraph reads like a literature review nobody commissioned. Before sending your own draft, when Perplexity did the research and you want to know how much of its voice came with it.
The checklist works in the other direction too. If you want a draft that does not read as retrieved and summarised, the signals above are an editing pass with the order already decided.