Paste a text and this AI content checker shows which patterns of unedited AI writing it contains and exactly where: stock phrases such as “delve into”, sentences that all run to the same length, inflated adjectives, “not just X, but Y” framing. Each signal is explained and highlighted in your text, so you can see what to edit. What it will not do is tell you who wrote the text, because no tool can know that from the words alone.
That limit is not modesty. OpenAI built a classifier for exactly this job and withdrew it on 20 July 2023 “due to its low rate of accuracy”: in its own evaluation it caught 26% of AI-written text and wrongly labelled human text as AI-written 9% of the time. A Stanford study published in Patterns in July 2023 found seven popular detectors flagged essays by non-native English speakers as AI-generated 61% of the time. A tool that prints “87% AI” over that kind of error rate is guessing with confidence. This checker shows its evidence instead.
What the AI content checker looks for
The checker measures eight signals. None of them proves anything on its own; together they describe how much a text reads like a first draft from a chatbot. Every signal is counted per 1,000 words or as a share of sentences, so long and short texts are comparable.
| Signal | What it measures | Why it points to AI drafting |
|---|---|---|
| Stock AI phrases | Phrases such as “delve into”, “in today’s fast-paced world”, “it’s important to note”, “plays a crucial role” | Chatbots reuse a small set of openers and filler phrases far more often than people do |
| Inflated adjectives | Crucial, seamless, robust, pivotal, comprehensive and similar, used as filler | Generated text praises without evidence |
| Connector openers | Share of sentences that start with Moreover, Furthermore, Additionally, Ultimately | AI drafts stitch paragraphs together with them |
| “Not just X, but Y” | Contrast frames such as “it’s not just a tool, it’s a strategy” | A favourite template of chatbot prose |
| Even sentence rhythm | How much sentence length varies (coefficient of variation) | People mix short and long sentences; generated text tends to even them out |
| Same-length paragraphs | How much paragraph length varies | Generated drafts come in blocks of near-identical size |
| Lists of three | Comma lists of exactly three items | A natural device that chatbots overuse |
| Em dashes | The long dash (—) per 1,000 words | Heavily overused in English chatbot output |
The phrase lists are written separately for English, Spanish and German, because the clichés differ: a Spanish draft says “en el mundo actual” and “cabe destacar”, a German one “in der heutigen schnelllebigen Welt”. Choose the language of the text in the menu next to the button.
How to read the result
The verdict has three levels: few, some or many AI-style signals. It comes from adding up the eight signals, with a strong signal counting double. It is a style reading, not a probability, and it never says “written by AI”.
When we ran the checker on two test texts on 12 October 2026, a deliberately generic paragraph about SEO strategy came back with five strong signals: eleven stock phrases, nine inflated adjectives and five sentences opening with a connector, in 132 words. A first-person account of a server migration, written with specific dates and numbers, came back with none. Most real texts land in between, and that middle band is where the highlighted text is most useful: it shows the three or four phrases worth rewriting.
Under the signals, a bar chart shows the length of every sentence. A row of bars of nearly the same height is the “even rhythm” signal made visible.
Who this is for, and who it isn’t
The checker is built for editors and writers who want to take the machine sound out of a draft, whether the draft came from a chatbot, a hurried freelancer or themselves on a Friday afternoon. It is not built to catch anyone. If you are a teacher, an employer or a client deciding whether someone used AI, the honest answer is that this tool, and every tool like it, cannot settle that. Non-native writers, legal and technical writing, and anything written to a strict template trigger the same signals.
Does Google penalise AI content?
Not for being AI content. Google’s guidance on AI-generated content, first published in February 2023, says that “appropriate use of AI or automation is not against our guidelines”, and that what matters is whether content is helpful and made for people. What Google does act on is producing many pages mainly to manipulate rankings, however they are made, which its spam policies call scaled content abuse. A text full of stock phrases is not a ranking problem because a classifier might spot AI in it; it is a problem because readers notice it says little.
What this checker cannot tell you
- Who wrote the text. No signal here, alone or combined, is evidence of authorship.
- Whether a text was edited after generation. A lightly edited AI draft and a formulaic human text can score the same.
- Watermarks. Some AI providers embed invisible watermarks; reading them needs the provider’s own detector, and their absence proves nothing.
- Facts. A text can pass with no signals and still be wrong.
- Very short texts. Under about 300 words the signals are noisy; under 80 the checker refuses to run.
The short version
- No tool can tell from the text alone whether a person or an AI wrote it.
- This checker shows eight explained writing signals and where they occur, never an “AI %”.
- Use it to edit: rewrite stock phrases, vary sentence length, cut filler adjectives.
- Google does not penalise content for being AI-made; it acts on unhelpful content produced at scale.
- Your text is analysed in your browser and never sent to a server.
FAQ
Is the AI content checker free?
Yes, with no account and no limit on checks. Each check accepts up to 5,000 words.
Is my text stored or sent anywhere?
No. The analysis runs in your browser. Nothing you paste reaches doctor-seo.net or anyone else.
Why doesn’t it give a percentage of AI content?
Because the percentage would be invented. The best-documented attempt, OpenAI’s own classifier, caught 26% of AI text with 9% false positives before it was withdrawn. A number with that much error behind it looks precise and misleads.
My own writing shows many signals. Does that mean anything?
Only that it uses patterns chatbots also use. Formal, technical and second-language writing often does. Read the highlighted phrases and keep the ones that say what you mean.
Does it work in Spanish and German?
Yes. Each language has its own lists of stock phrases, inflated words and connectors; the rhythm signals work the same in every language.
Where to go next
This tool is part of the Doctor SEO tools section. For copied rather than generated text, use the plagiarism checker. How Google treats AI content and scaled content is covered in the free SEO course, and the SEO glossary defines any term that is new to you.
Sources
- OpenAI, New AI classifier for indicating AI-written text, 31 January 2023, with the note added on 20 July 2023; checked 12 October 2026.
- Liang, W. et al., “GPT detectors are biased against non-native English writers”, Patterns, July 2023, as reported by Business & Human Rights Resource Centre, 14 August 2023.
- Google Search Central Blog, Google Search’s guidance about AI-generated content, 8 February 2023.
- Google Search Central, Spam policies for Google web search (scaled content abuse), checked 12 October 2026.
- doctor-seo.net, first-party test of the checker on two test texts, 12 October 2026.