Was This Written by AI? How to Tell in 5 Minutes
You are reading an essay, a job application or a product description — and something feels off. The sentences are smooth, the grammar is perfect, and yet nothing in it sounds like a person actually wrote it. Was this written by AI? You can answer that question properly in about five minutes, with no special software. This guide gives you the quick check: the manual signs to read for, a free detector to confirm what you found, and what to do with the result — because the worst mistake is not missing AI text, it is accusing a human of using it.
The short answer first: no single sign, tool or score can prove a text was written by AI. What works is combining several checks and treating the result as a signal, not a verdict. Here is how.
The five-minute check
Work through these three steps in order. Each one takes a minute or two, and together they are far more reliable than any one of them alone.
Step 1 — Read it yourself (2 minutes). Look for the five manual signs below. One sign means little; four or five together are worth a closer look.
Step 2 — Run it through a detector (2 minutes). Paste at least 300 words into a free checker such as GPTZero or ZeroGPT, then run the same text through a second, different tool. If the two disagree, the answer is “unknown”.
Step 3 — Check the facts and the context (1 minute). Do the claims have sources you can verify? Does the writing match the person’s other work? Is there any evidence about how it was written — drafts, version history, notes? A detector score plus a factual check beats a detector score on its own every time.
The five manual signs of AI writing

Before you open any tool, read the text yourself. Trained readers catch AI writing with signals no scanner reports:
1. Uniform rhythm. Sentence after sentence of the same length and cadence. Human writing breathes — a short sentence, then a long winding one, then a fragment. Generated writing hums along at one steady pace. If every paragraph is three sentences of near-identical shape, be suspicious.
2. Generic transitions and filler phrases. Phrases like “in today’s fast-paced world”, “it is important to note”, “plays a crucial role” and “when it comes to” appear again and again. So do overused words such as delve, leverage, landscape, multifaceted and furthermore. One is normal; a page full of them is a pattern.
3. Vague attribution. Claims with no source, no name and no detail: “experts say”, “studies show”, “many people believe” — with nothing you could actually check. AI text invents confident-sounding claims and rarely names where they came from.
4. Missing specifics. The text describes a topic perfectly and says nothing that could only come from someone who did the work. No names, places, dates, numbers or strange specific details — just smooth generalities. Real writing has texture: a price, a street name, a mistake that was made, a number that does not round neatly.
5. A flawless surface with an empty centre. No typos, no opinions, no risk. The prose is so safe it reads like no one wrote it, and after reading it you cannot remember a single specific idea. The conclusion often restates the introduction almost word for word.
None of these signs proves anything on its own. Plenty of careful human writers use some of them, especially in formal or technical writing. But a text that shows four or five at once deserves step 2.
Step 2: run it through a detector

Detector tools automate the statistical check. They measure how predictable the word choices are — a measure called perplexity — and how much sentence length and structure vary. AI models tend to pick the most likely next word, so AI-written text is often smoother and more predictable than human writing.
Two free tools are enough for a quick check:
GPTZero. The most widely used detector in education. It highlights sentences individually as human, mixed or AI, and its free tier covers 10,000 words a month. Paste your text, read the highlighted sentences as well as the headline score, and note which passages were flagged.
ZeroGPT. A free checker that needs no account for a basic scan and returns a simple percentage. It is a good second opinion, though independent testing has found it less accurate than the paid tools on edited text. Our ZeroGPT guide walks through it step by step.
For a fuller comparison of the free options, see our best free AI detectors guide and our general Detect AI guide.
Read the result properly. A high “AI” percentage on a long, unedited passage is a meaningful signal. The same score on 100 words is almost meaningless — detectors need at least 300 words to give a stable result. And if the two tools disagree, the honest answer is that you do not know. Do not pick the result you prefer; treat a disagreement as inconclusive.
Step 3: check the facts and the context
This is the step most people skip, and it is the one that catches what detectors miss. AI-written text often invents statistics, quotes and references that do not exist. Pick two or three factual claims from the text and check them: search for the study, the quote, the number. A made-up reference is stronger evidence of AI writing than any detector score.
Then look at the context. Does this piece match the person’s other writing — the same vocabulary, the same level of polish, the same structure? A sudden jump in fluency between assignments tells you more than any percentage. And is there any evidence about how it was written? Drafts, notes, browser history, or a version history in Google Docs can settle a question that no detector can.
Also ask what tool might have been used. A detector cannot tell you whether text came from ChatGPT, Claude or Gemini — it measures machine-like patterns, not a specific tool’s fingerprint. If the question is specifically “was this written by ChatGPT”, a detector alone cannot answer it.
Why the check sometimes fails
Honesty requires saying where this five-minute check breaks down. Paraphrasing tools and so-called humanizers rewrite AI text specifically to defeat detectors, and 2026 testing found they cut top detectors’ accuracy from around 98% to between 41% and 67% — barely better than a coin flip. Heavy human editing of an AI draft has the same effect. A clean detector score never proves a text is human-written.
The other direction is just as important: detectors wrongly flag human writing as AI. Research from Stanford’s Human-Centered AI lab found that leading tools wrongly flagged 61.3% of essays written by non-native English speakers as AI-generated. Formal writing, technical writing and short passages are the most likely to be flagged. And OpenAI itself withdrew its own AI text classifier in July 2023 after it caught only about 26% of AI-written text while wrongly flagging 9% of human writing — “due to its low rate of accuracy”.
What to do with the result

If you are a teacher or lecturer. Use the result as a first-pass signal, then look at the student’s other work and talk to the student before you escalate anything. Ask for drafts or notes. A detector score alone should never be the basis of an academic misconduct case.
If you are an editor or publisher. Run submissions through two detectors, check facts and sources, and keep a human in the loop. A submission that fails the fact check is a problem whatever wrote it; a submission that passes the fact check but reads as machine-written may just need an edit.
If you are an employer or recruiter. AI-written applications are now normal, and many are honest about it. Judge the substance — does the candidate actually know the work? — rather than policing the prose. A cover letter is a poor place for a forensic test.
If it is your own writing. This is the most useful case of all. Run a draft through a detector before publishing: if it reads as machine-written, the fix is usually more specifics, more opinions and shorter sentences — the same edits that make any writing better.
Frequently asked questions
Can you tell if something was written by AI just by reading it?
Often, yes — with practice. Uniform rhythm, generic filler phrases, vague attribution, missing specifics and a flawless but empty surface are the classic signs. But reading alone is not proof: careful human writers show some of these signs too, and heavily edited AI text can read perfectly human.
What is the best free tool to check if text was written by AI?
GPTZero is the most widely used free option, with a free tier of 10,000 words a month and sentence-level highlighting. ZeroGPT is a good free second opinion that needs no account. Use two tools, not one — if they disagree, treat the result as unknown.
How many words do you need to check?
At least 300 words for a meaningful detector result. Very short passages give detectors too little to work with, and scores on them are unreliable. For the manual read, even a few paragraphs are enough to spot the classic signs.
Can AI detectors be wrong?
Yes, in both directions. They miss AI text that has been paraphrased or heavily edited, and they wrongly flag human writing — Stanford research found 61.3% of non-native English speakers’ essays were flagged by leading tools. A score is a signal, not a verdict.
Is it illegal to use AI to write something?
No — using AI to write is not illegal in itself. What matters is the context: passing AI-written work off as your own can breach university rules, workplace policy or a publisher’s terms. Check the rules that apply to your situation.
Final word
Was this written by AI? Read it for the signs, run it through two free detectors, check the facts — and then treat the result as the start of a conversation rather than the end of one. For the tools themselves, our Detect AI guide, our free AI detector guide and our ZeroGPT guide are good next reads.
Sources: GPTZero — https://gptzero.me/ · OpenAI — https://openai.com/



