Before submission
What the AI assignment checker does in your routine
This is one part of a short pre-submission routine, not a verdict machine. Paste a section, press Analyze, and read back a written account of what in your prose looks machine-written and why. There is no percentage and no ruling on who wrote the text, because the engine is instructed not to claim certainty about authorship.
It runs from the box on this page. Free, 5 runs per rolling 24 hours, up to 12,000 characters each, no account. Your text goes to our AI server to be read, and there is no file upload, so copy the section out of your document rather than attaching it.
The routine
Four checks, in the order that saves the most time
Two of these have definite answers and two are judgement. Do the definite ones first, because they are quick and they are the ones that get you in trouble.
- Invisible characters
- Run the hidden character scanner on this site. It covers 14 specific marks, works in your browser, and either finds them or does not.
- Sources
- Open every citation. Confirm the source exists and says what you attributed to it. Invented references are a problem no detector is needed to expose.
- Prose
- Use the box above on the sections you are least sure about, and fix the specific sentences it names.
- Policy
- Read your course or institution's AI rules and confirm what you did is permitted and, where required, declared.
How to
How to run the AI assignment checker in three steps
One section per run, so spend the five on the parts you actually doubt.
- Step 1
- Pick the weakest section rather than starting at page one. Introductions and conclusions are where generated phrasing survives longest.
- Step 2
- Paste it in, add an instruction if you want a narrower read, such as ignoring quoted material, and press Analyze.
- Step 3
- Edit the sentences it names, one at a time, then move to the next section rather than re-running the same one.
The definite check
Strip invisible characters before you do anything else
If any part of your document was pasted out of a chat window, it may be carrying characters you cannot see. Zero-width spaces, word joiners, a byte-order mark that wandered into the middle of a paragraph, non-breaking spaces that look identical to ordinary ones.
This is the one item in the routine with a yes or no answer. The scanner on this site handles exactly 14 of them, locally, with no model and no probability involved. It takes a few seconds and it removes an argument you would otherwise have to have. Do it before you spend any of your five analysis runs.
The other definite check
Verify every source, one at a time
Open each reference in your list and answer two questions. Does this source exist, at this title, by this author, in this year? And does it actually say the thing you cited it for? Both are checkable in a minute each, and both are where AI-assisted drafting fails most visibly.
The University of Pittsburgh's teaching centre makes the point directly: proving AI use is difficult, but hallucinated sources are a violation independent of any detection tool. That cuts both ways for you. A checker cannot save you from a fabricated citation, and no checker is needed to find one.
The judgement check
The phrasings to look for while you edit
When you read the findings, these are the patterns most likely to be named. The checker on this site matches 17 in total, and these are the ones that survive into coursework.
- Announced information
- It is important to note that. Cut the phrase and keep the fact. The sentence gets shorter and better.
- Three connectives in a row
- Moreover, Furthermore, Additionally. Vary them or drop them, since most sentences do not need one at all.
- A universal opener
- In today's rapidly evolving landscape. If the sentence would fit any other essay, it is not doing work in yours.
- Importance without content
- Plays a crucial role. Replace it with what the thing actually does.
- Ornamental metaphor
- Tapestry, realm of, stands as a testament to. Rare in student writing, common in model output.
Straight answer
What to do if you have already been flagged
Nothing here predicts or reverses another tool's result, and there is no affiliation with Turnitin, GPTZero or any other service. What is worth knowing is that the published reliability figures conflict badly. Turnitin has claimed under 1% false positives, a Washington Post test reported up to 50% on a smaller sample, and a University of San Diego guide reports that the same checker misses roughly 15% of AI text.
The errors also fall unevenly. Research cited in that guide, from Stanford's Human-Centered AI institute and from Liang and colleagues, reports higher flagging rates for students writing in a second language and for neurodivergent students. Some institutions have responded: Pittsburgh's teaching centre has disabled Turnitin's AI detection and endorses no detector. If you are in a conversation about a score, that context is relevant and it is published.
In practice
Who the AI assignment checker is for, and what it will not do
It is for you, reading your own draft, before you submit it. Typical reasons: a model helped with an outline and you want to know how much of its voice is left, you write in English as a second language and have heard that gets flagged, or you want a second read on prose that feels stiff.
It is not a marking tool and not an invigilation tool. It returns observations, not a ruling, and it will not tell you which model produced anything because it is instructed not to try. If you teach, the teaching centres cited above agree on the recommendation: a detector score should not be the sole basis of a misconduct decision, and assignment design and clear policy do more than any checker.