Direct answer
AI generated content is text, images, audio, video, or code produced by a generative model such as ChatGPT, Claude, or Gemini in response to a prompt, rather than authored line by line by a person. In academic and publishing contexts the term usually means prose whose wording and structure were produced by a large language model, including drafts a student later edits. Detection tools such as Turnitin's AI writing indicator estimate what share of a submission looks machine-generated, and they report that share as an indicator rather than as proof of misconduct [1].
How Do Turnitin And Similar Systems Decide Whether Content Is AI Generated?
Detection is statistical, not forensic. Turnitin's model reads the submission as a sequence of tokens and asks how predictable each next word was given everything before it. Human writing tends to carry more surprise — unusual word choices, uneven rhythm, idiosyncratic transitions — while model output tends to sit closer to the most probable continuation. The system aggregates those signals into a per-segment probability and then rolls them up into one percentage for the document [2].
That percentage describes prose only. The AI Writing Report excludes quoted material, reference lists, and segments shorter than a minimum length, so a paper with heavy citation can show a lower AI share than its actual drafted body. The report also highlights individual sentences, which is why two students with the same overall number can face very different conversations with an instructor [2].
The threshold matters more than most readers expect. Below a 20% confidence band, Turnitin displays an asterisk (*%) instead of an exact figure, signalling a low-confidence result rather than a clean bill of health. A *% reading means the detector found some signal but not enough to state a precise number — it is not the same as 0% [2].
Finally, the indicator is evidence, not a verdict. Turnitin states plainly that its AI score should be read alongside context, drafting history, and the institution's policy, because detectors can be wrong in both directions. That framing is why most universities treat the report as the start of a conversation rather than the end of one [1][2].
Does AI-Assisted Editing Or Paraphrasing Still Count As AI Generated Content?
Yes, in most institutional definitions it does — and this is the point where students most often misjudge their own position. If a model produced the sentences and you then swapped synonyms, reordered clauses, or ran the text through a paraphrasing tool, the underlying authorship is still machine-generated. Turnitin's guidance on AI paraphrasing tools treats this as a form of misrepresentation when the assistance is not disclosed, because the submitted wording originates from a model rather than from the student [3].
Paraphrasing also does not reliably change the signal. Detectors score statistical patterns across sentences — perplexity and burstiness — not individual word choices, so a synonym swap leaves most of the underlying distribution intact. Students who assume "rewritten means safe" are often surprised by a report that still flags the same passages, only with different vocabulary [3].
The practical dividing line institutions draw is disclosure and proportion, not the presence of any AI at all. Light assistance — grammar checks, brainstorming, restructuring your own sentences — is commonly permitted or tolerated; generating the argument and the prose and presenting it as your own is not. When a policy is ambiguous, the safe move is to ask the module lead and to keep your draft history, because versioned files are the strongest evidence of how the work was actually produced [3].
Can You Check Whether Your Own Draft Would Be Flagged As AI Generated Before You Submit?
Usually not through your institution's own Turnitin assignment. Once work is submitted to a Turnitin assignment, it enters the instructor's view and, depending on settings, the student paper repository — you cannot run a private test through that same door. Some institutions enable draft or self-check options inside the assignment, but availability is decided per module and is far from universal [4].
Where no self-check exists, students typically turn to independent preview services. The useful ones return the same two artefacts an instructor sees: an AI detection report with a percentage and sentence-level flags, and a similarity report with matched sources. Seeing both before the deadline converts a guess into a decision — you either submit as-is, revise the flagged passages, or rewrite them in your own voice [4].
Two practical cautions apply to any preview. First, a preview is only meaningful if it is non-repository: a check that files your draft into a student paper database can create a self-match problem later. Second, read the number the way an instructor would — a low single-digit AI score or a *% reading is a weak signal, while a high score concentrated in specific paragraphs points to text you should genuinely rewrite rather than tinker with [1][4].
If you want that preview before your deadline rather than after it, turnitin0 gives you the instructor's-eye view in one checkout: a Turnitin AI detection report and a similarity report delivered together, typically in under 15 minutes, without adding your file to the student paper database.
※ Turnitin0.com - Actual Turnitin AI Report Cover, Score, Flag And Similarity Summary
FAQ
Is AI generated content always written entirely by a machine?
No. The label covers a spectrum from fully model-written text to human drafts that a model substantially rewrote. What matters for most policies is who produced the submitted wording, not whether a human touched the file afterwards [3].
Why does Turnitin show *% instead of a number?
Turnitin withholds an exact figure below its 20% confidence threshold because the signal is too weak to state precisely. A *% result is a low-confidence indicator, not a guaranteed 0% [2].
Do quotes and references count toward the AI score?
They are excluded from the AI writing analysis along with segments under a minimum length, so a heavily cited paper can report a lower AI share than its drafted body [2].
Can I check my own draft in Turnitin before submitting?
Only if your institution has enabled a draft or self-check option on the assignment. Otherwise you cannot run a private test through the same submission point, and independent preview services are the usual route [4].
Does paraphrasing AI text make it human-written?
No. Paraphrasing tools change vocabulary but leave the underlying statistical patterns largely intact, and Turnitin's guidance treats undisclosed AI paraphrasing as misrepresentation [3].