The core distinction
AI-Generated
The AI wrote it
The primary text was produced by an AI model. The human role was to provide a prompt, select from outputs, and possibly make minor edits. The words, structure, and arguments come from the AI.
Examples: submitting a ChatGPT essay, publishing an AI-written blog post without editing, sending an AI-drafted email as your own
AI-Assisted
You wrote it, AI helped
The human wrote the primary content. AI tools were used in the process for tasks like grammar checking, brainstorming, research summarization, or suggesting a word. The words and structure are primarily the human's.
Examples: using Grammarly, asking AI to suggest a better word, using AI to research a topic but writing in your own words, getting AI feedback on your draft
The line between these two categories is not always sharp. Heavy editing of AI output, AI-written sections embedded in human-written text, and iterative collaboration between human and AI all create a spectrum. But the two poles are clearly distinct, and most real-world cases fall closer to one end or the other.
How detectors treat each category
AI detectors are trained to identify the statistical and semantic signatures of AI-generated text. AI-assisted writing that is primarily human-written behaves very differently from AI-generated text under detection:
Unedited AI output
Strong AI signal across most detectors. Statistical smoothness, uniform structure, and pattern matches are all present.
AI output with light edits (corrected a few words)
Scores drop but remain clearly elevated. The core AI statistical pattern is preserved across light editing.
AI output substantially rewritten (restructured, personalized)
Variable. May fall below detection thresholds. Scores depend heavily on how much of the original AI structure remains.
Human writing with AI grammar correction
Grammar correction tools do not alter statistical patterns, phrase structure, or argumentative form. Typically low scores.
Human writing researched with AI help
If the text is written in the human's own words, AI use in the research phase does not appear in detection signals.
Human writing with AI-suggested rewrites incorporated
Depends on how many AI-rewritten sentences were adopted. Isolated adopted suggestions raise scores modestly; wholesale rewrites raise them significantly.
Why the distinction matters for policy
Most AI policies are not written to prohibit AI assistance entirely. They target AI generation: submitting text that an AI produced as if it were your original work. Understanding this distinction helps navigate policies accurately:
Academic integrity policies
AI-generated
Prohibited in nearly all cases. Using AI to write your essay and submitting it as original work is academic dishonesty regardless of how good the editing was.
AI-assisted
Usually permitted with disclosure, or permitted without restriction for research assistance. Policies increasingly distinguish between AI writing and AI tools in the research workflow.
Content publishing and journalism
AI-generated
Increasingly requiring disclosure. Many publications now require a note when AI generated primary content.
AI-assisted
Generally not subject to disclosure requirements. Using AI for headline brainstorming, grammar editing, or fact-checking is standard practice.
Professional and legal contexts
AI-generated
Some professions (law, medicine) have rules around signing work you did not author. AI-generated documents signed as your own analysis may create liability.
AI-assisted
Using AI tools in your research or drafting process is generally acceptable. The professional is still accountable for the content.
Employment and hiring
AI-generated
AI-generated cover letters and work samples presented as your own are misrepresentation. Detected AI content in samples has led to rejection and termination.
AI-assisted
Using AI to improve or polish your own writing is generally considered acceptable, similar to using spell check or editing software.
The gray zone: heavy collaboration
The hardest cases for both detection and policy are in the middle: writing that involves extensive back-and-forth between human intent and AI execution. Some examples:
- •Iterative drafting: human writes a paragraph, AI rewrites it, human edits the rewrite, AI polishes the edits. At what point is this AI-generated versus AI-assisted?
- •Prompt-heavy generation: the human writes an extremely detailed prompt specifying structure, arguments, examples, and tone. The AI fills in the prose. Is the result AI-generated when the human authored the entire architecture?
- •Section-by-section: an essay where some sections are fully human-written and others are AI-generated. Detectors may flag it overall but the flagged sections are a subset.
- •AI used to expand notes: the human writes bullets and rough ideas; AI expands them into full paragraphs. The ideas are human-originated; the prose is AI-produced.
These cases do not have universal answers. Most policies are still catching up to this complexity. When in doubt, disclose. “This document was drafted with significant AI assistance” is both accurate and appropriate for most contexts where disclosure expectations exist.
What Airno detects
Airno's 8-detector ensemble is calibrated to detect AI-generated content, not AI-assisted work. Writing that is primarily human-produced with AI editorial help will typically score low. Text that was primarily produced by an AI model will score high, even if the human made edits afterward.
The per-detector breakdown helps you understand what is driving a score. If only the statistical model is slightly elevated, that is consistent with AI-assisted editing. If statistical, pattern, and semantic detectors are all elevated, that is consistent with AI-generated content that was edited but not rewritten.
For deeper context on detection accuracy across content types, see What Percentage of AI Content Is Detectable? For information on how style imitation affects these signals, see Can AI Write in Your Voice?
See exactly which signals are elevated
The eight-detector breakdown helps you understand whether a high score is driven by AI generation or by something else. Free, no account.
Try Airno free