AI Content Detector
See the patterns that make writing read as AI-generated, and how to fix them
How it works
No black box. Here's exactly what AI Content Detector checks.
- 1
Paste text or a URL
Up to 15,000 characters, or a public page whose main text we pull out. Nothing you submit is stored.
- 2
We look for patterns you can see
Stock phrases, stock transitions, runs of same-length sentences, repeated openings, list-heavy layout and heavy dash use.
- 3
You get the evidence, sentence by sentence
Every flagged sentence says what we noticed, and for the strongest ones we suggest a rewrite you can use or ignore.
Why this matters
Readers skim past writing that sounds like everything else, and search engines reward pages that are useful and specific. The patterns this tool finds are the ones that make text sound generic, whoever wrote it. Fixing them usually makes the writing clearer.
- Stock phrases ("in today's fast-paced world", "plays a crucial role") take up space without saying anything.
- Sentences that are all the same length read flat. Varying length is one of the easiest fixes.
- Specific detail, a real example or a number you can source does more than any rewrite.
How does this AI content detector work?
It counts features you can check yourself: stock phrases and words that chat models overuse, sentences that open with stock transitions such as "Moreover", runs of sentences of nearly identical length, sentences that keep starting the same way, list-heavy layout and heavy dash use. Each feature adds points, and the total sets the level. "Many AI patterns" also needs stock phrasing, so rhythm alone can never get a text there. A language model is used only to explain flagged sentences and suggest rewrites, never to judge who wrote the text.
Why can no tool prove a text was written by AI?
Because people and models write in overlapping ways. Plenty of people use stock phrases and even rhythm, and a lightly edited AI draft can lose every pattern a detector looks for (in our own test, all ten lightly edited samples came back clean). Detectors that print a precise "AI probability" are presenting a guess as a measurement, and a 2023 Stanford study (Liang et al., published in Patterns) found popular detectors flagged essays by non-native English writers far more often than essays by native speakers. So we show patterns and evidence and leave the judgement to you.
Does Google penalize AI-generated content?
Google's published guidance on AI-generated content says its systems reward helpful, original, people-first content however it is produced, and that using automation mainly to manipulate rankings breaks its spam policies. In other words the question Google asks is whether a page is useful, not who or what typed it. Removing generic patterns helps because it pushes a page toward specific, useful writing. See Google Search Central: developers.google.com/search/blog/2023/02/google-search-and-ai-content.
Is my text stored?
No. Text and URLs are analyzed and discarded. They are not saved, logged or used for training. When suggestions are on, the flagged sentences (not the whole text) are sent to OpenAI to write the rewrites, under their API terms.
How did we test it?
We built a labeled sample of 132 texts: public-domain books and pre-2020 Wikipedia articles (human), answers from two OpenAI models to the same prompts (AI generated), and AI answers we edited lightly by hand. We chose every threshold on one part of the sample, froze them, then scored the other part once. These are the raw counts from that held-out part.
| Held-out texts | No AI patterns | Some | Many | Inconclusive |
|---|---|---|---|---|
| Human writing (46) | 43 | 3 | 0 | 0 |
| AI generated (24) | 4 | 2 | 18 | 0 |
| AI, then lightly edited by hand (10) | 10 | 0 | 0 | 0 |
- Human texts marked "Many AI patterns": 0 of 46 (0%). Our launch rule was at most 10%.
- Writers for whom English was a second language (Joseph Conrad, Jacob Riis): 6 of 6 texts came back with no AI patterns.
- Lightly edited AI text: 10 of 10 came back clean. Light editing removes these patterns, which is exactly why a pattern check cannot prove authorship.
Sample-specific, not a general accuracy guarantee. These counts describe how this analyzer behaved on this one labeled sample; other writing, other models and other editing will behave differently. A pattern level says nothing about who wrote a text.
Want the full story across every page?
The AI Content Detector checks one text at a time. CrawlTide audits every page on your site, flags low word counts and duplicate titles, and turns your Search Console data into a list of what to write or improve next.
No credit card. Free tier covers a small site end-to-end.
More free tools
See all →Content Gap Checker
Spot missing topics, FAQs, and internal-link opportunities for your site
AI Answer Readiness Checker
See if a single page is structured to win direct answers in ChatGPT, Perplexity, and AI Overviews
Meta Title & Description Checker
Preview your title and meta description in Google search and get rewrite suggestions