⚠️ Affiliate Disclosure: This post is not an affiliate associated review. It may become one in the future and this disclosure will reflect that. This does not affect my reviews and I only recommend tools I genuinely believe in.
Can You Tell If Something Was Written by AI?
Sometimes — but not reliably, and not with certainty. You can often spot lazy, unedited AI writing by its patterns. Carefully edited AI text, or human writing that happens to be plain and tidy, is much harder to call. No tool or technique can prove either way on its own.
That uncertainty is the most important thing to understand before you accuse a student, a freelancer or a colleague. This guide covers the signs worth noticing, how AI detectors work, why they get it wrong, and what to do instead.
Common Signs of AI Writing
None of these proves anything by itself. Several together, in a piece that also lacks anything personal or specific, are worth a closer look.
- Signs of AI writing in the content — confident generalities with no specific examples, names, dates or numbers that could be checked.
- Even, polished rhythm — paragraphs of similar length, every point neatly balanced, nothing rough or surprising.
- Stock phrases and structures — “in today’s fast-paced world”, “it’s important to note”, “not just X, but Y”, and lists of exactly three.
- Hedging everywhere — every claim softened, every side presented, no real opinion.
- Facts that do not check out — invented quotes, sources or statistics are a strong signal, because AI hallucinations are rare in careful human writing.
- A mismatch with the writer — a sudden change in vocabulary, spelling conventions or quality compared with their earlier work.
- Leftovers — “As an AI language model”, “Certainly! Here is”, or placeholder text in square brackets that nobody removed.
The catch: plenty of human writing has these features too. Corporate copy, formulaic school essays and anything written by someone working in a second language can tick several boxes without any AI involved.
How Do AI Detectors Work?
AI detectors such as GPTZero, Turnitin’s AI checker and the detectors built into tools like Grammarly and QuillBot look for statistical patterns. Language models tend to choose predictable words and produce sentences of fairly even length. Detectors measure how predictable a text is and how much it varies, and estimate the likelihood that a model wrote it.
The output is a probability, not a verdict. A score of “80% AI” does not mean 80% of the text was written by AI, and it does not mean the detector is 80% sure. It means the text looks statistically similar to AI writing the detector was trained on.
How Accurate Are AI Detectors?
Less accurate than their marketing suggests, and the evidence is hard to ignore.
| Evidence | What it found |
|---|---|
| OpenAI’s own AI classifier (2023) | In OpenAI’s tests it caught only 26% of AI-written text and wrongly flagged 9% of human writing. OpenAI withdrew it in July 2023, citing its low rate of accuracy. |
| Stanford study of seven detectors (2023) | The detectors misclassified more than half of essays written by non-native English speakers as AI-generated — an average false-positive rate of 61% — while judging native speakers’ essays almost perfectly. |
Both findings point the same way. Detectors penalise writing that is simple and predictable, which describes a lot of honest human work. And light editing or paraphrasing of AI text is often enough to slip past them.
Better Ways to Check Whether Something Was Written by AI
- Check the facts in the text — search any specific quote, source or statistic. Fabricated details are the most solid sign you will find.
- Look at the process — version history in Google Docs or Word shows whether a document was typed and revised over time or pasted in all at once.
- Ask about the work — someone who wrote a piece can usually explain their reasoning, sources and choices in their own words.
- Compare with earlier writing from the same person.
- Watch for provenance labels — some AI tools now embed hidden watermarks or attach content credentials. They are far more established for images than for text; my guide to AI watermarking explains what can and cannot be checked.
What to Do If an AI Detector Flags Your Own Writing
It happens, and it is not proof of anything. Stay calm and gather evidence of your process: drafts, version history, notes, research and browser history. Offer to talk through how you wrote the piece. If you are a student, ask what the institution’s policy says about detector results — many universities advise staff not to rely on them alone.
If you use a tool like Grammarly or QuillBot to tidy your writing, be aware that heavy rewriting features can make text look more AI-like to a detector. Spell-checking is not the same as having AI write for you, but it is worth knowing what your tools are doing.
Frequently Asked Questions
Can AI writing be detected 100% of the time?
No. No detector or manual check can identify AI writing with certainty. Edited or paraphrased AI text often passes as human, and plain human writing is regularly flagged as AI. Detection gives a likelihood, never proof.
Are AI detectors reliable?
Not reliable enough to act on alone. OpenAI withdrew its own AI text classifier in 2023 for low accuracy, and a Stanford study found detectors flagged more than half of essays by non-native English writers as AI-generated. Use them as a prompt to look closer, not as evidence.
What words does AI use most?
AI writing tends to overuse safe, polished phrasing — words like “delve”, “crucial”, “landscape” and openers like “in today’s world”. Word lists are weak evidence, though: models change, and many people write this way naturally.
How can I prove I did not use AI?
Keep evidence of your process: drafts, version history in Google Docs or Word, research notes and sources. Being able to explain your reasoning and choices in conversation is often the most convincing proof.
Can Google tell if content is written by AI?
Google says it does not penalise content simply for being AI-generated; it targets low-quality content made mainly to manipulate rankings, however it was produced. What matters for search is whether the content is genuinely helpful and accurate.
More AI Explained guides: What Is an LLM? · AI Hallucinations: Why AI Makes Things Up · What Is AI Watermarking? · How to Write AI Prompts That Actually Work
James is an SEO and content specialist with over 12 years’ experience, and tests AI tools so everyday people don’t have to. WhatAITools.co.uk covers honest, no-jargon reviews to help you find the right tool for your needs.

Leave a Reply