Plagiarism and AI at Liberty: the detector's published limits

The short answer

Liberty runs Turnitin’s AI detector — and then publishes, in unusual detail, the circumstances in which its own faculty are told not to trust it. That list is the most useful thing on this page, especially if English is not your first language.

Liberty’s position is also less restrictive than students assume: it does not ban ChatGPT or Grammarly. What it prohibits is letting a tool generate — or substantially rewrite — the content you submit as your own.

The benefit-of-the-doubt list

This is from Liberty’s own faculty guidance, under “Limitations”, and it opens with an instruction rather than a caveat:

“Even with a low error rate, the AI detector has limitations that all faculty must consider. It is highly recommended to provide students with the benefit of the doubt when these limitations are present.”

The four listed limitations, verbatim:

• “The error rate increases at results of less than 20%. 21%-100% results maintain the less than 1% error rate.”
• “Technical writing in the STEM fields, research methodologies, data-heavy content, and any other project that may reduce the presence of a ‘human-like’ writing style may produce a false flag at a higher rate.”
• “Resumes, poetry, song lyrics, lists, outlines, and tables may also produce unreliable results.”
• “A non-English-speaking student’s writing style may be affected by the use of AI-powered translation tools and could be interpreted as AI-generated. Purely human-written content by non-English speaking students should not influence the false flag rate.”

If you are writing your degree in a second language, that last bullet is worth having to hand. Liberty has written down that its own tool may misread you, and told its faculty what to do about it.

The same document is blunt about what a flag is and is not: “The AI detector is designed to highlight text that is ‘AI-like’. That highlighted text may or may not be AI-generated in violation of the Liberty Way. For students who write in a style that mirrors AI, their work is highly likely to be flagged. AI detection results alone are not an indictment regarding the student’s conduct.” And: “All factors that may reasonably influence the AI detector’s results should be considered.”

There are also compatibility limits worth knowing, because they explain odd results: documents under 300 words or over 15,000 words “are not compatible with the AI detector”, and “AI detection is only available in English”.

On the wider question of what these numbers can support, see how accurate AI detectors really are and AI detectors and non-native speakers.

The 0.8 threshold, and whose figures those are

Liberty explains the mechanism:

“Turnitin’s AI detector evaluates each sentence on a scale of 0.0 to 1.0 for the probability of being AI-generated. All text must meet a 0.8 thresholdof probability to be flagged ‘highly probable’ as AI-generated. Anything below the 0.8 standard will not be flagged. This is the primary safeguard against false flags.”

It also restates accuracy figures — “a less than 1% error rate for false flags”, tested “using 800,000 human-written papers”. We want to be careful here, because being careless about accuracy claims is the exact failure this site exists to avoid.

Those are Turnitin’s own figures, restated by Liberty. They come from the vendor’s internal testing and are not independently reproducible by anyone outside the company. We are not saying they are wrong — we have not measured Turnitin and cannot. We are saying they are vendor figures, and that a document-level error rate says little about your particular document, especially if one of the four limitations above applies to it.

For the same reason we publish our own error rates with their population attached, and repeat that they are measured on English and German corpora — and that no detector, ours included, produces proof. That is the same thing Liberty tells its faculty in its own words.

Liberty restricts the tooling for a second reason too: “No other AI detector may be used when investigating possible academic misconduct. This is due to the high reliability of Turnitin’s AI detector. Additionally, submitting assignments outside of the secure Canvas/Turnitin platforms presents concerns related to student privacy.”

Grammarly: where the line falls

This is the most practically useful passage Liberty publishes, because it distinguishes between features rather than banning a product:

“All Grammarly products provide the opportunity to use generative text AI features in some form. The use of those generative text features will regularly be flagged by the AI detector. Spelling, grammar, punctuation, and mild corrections to content will not cause that content to be flagged by the AI detector because those features do not use generative text technology.”

And the instruction to students:

“The use of Grammarly is very common and is not prohibited by the university. … If Grammarly suggests modifications to your work beyond basic spelling and grammar edits, those modifications should not be accepted. We encourage students to let their true writing voice be prominent in their work and not allow it to be diluted by AI generated content. When possible, turning off generative text AI features in Grammarly or any other writing aid is highly recommended.”

That is a workable rule you can apply while you write: accept the comma, decline the rewrite. And the practical tip — turn the generative features off — removes the temptation entirely.

Paraphrasing tools

Liberty names one directly, and we should quote it in full because it concerns the category our own product sits near:

“Many paraphrasing tools also use generative text AI and can be flagged by the AI detector. The use of Quillbot, as an example, is highly likely to be flagged by the AI detector. The university’s standards for academic integrity do not permit the use of paraphrasing tools to the extent that originality becomes compromised. … This act is considered academic misconduct.”

Note the test: not the tool, but whether “originality becomes compromised”. That is the same principle running through everything Liberty publishes.

Our own check sits on the permitted side of Liberty’s Grammarly line by design — it marks filler, hedging and tired transitions in your own sentences and suggests plainer wording, and nothing is rewritten for you. But the honest advice is Liberty’s own: if a suggestion changes what your text says rather than how it reads, do not accept it.

What Liberty actually prohibits

The Liberty Way was updated on 1 April 2023 to address AI. The standard, verbatim from the faculty guidance:

“The standard is that students should not use AI to generate content for their assignments in a manner that compromises originality and authenticity. This largely prohibits the use of generative text AI features/programs specifically. The university does not ban the use of AI tools as a whole and does not restrict access to AI products/platforms like ChatGPT on university servers. What is prohibited are AI-enabled acts that may compromise our long-standing expectations of academic integrity in student work.”

And to students, with the clearest formulation of the boundary:

“What we prohibit is allowing AI to generate content that is presented as original work or allowing AI to significantly modify existing work to the point that the assignment is no longer in the student’s own words. […] Just as students should not hand off their work to another person to add content to it, the same is true of AI.”

Liberty adjudicates AI misuse through two of its four stated principles — originality(“each student’s work should be unique to them”) and authenticity (“genuine, reliable, created with due diligence, and supported by evidence”) — rather than by naming AI inside the definition of plagiarism.

Liberty publishes no required AI disclosure form. It publishes a prohibition and a boundary instead. If you have used a permitted tool and want to be transparent anyway, our AI disclosure statement guide covers the shape.

One documentary caveat: the Liberty Way itself is published through a JavaScript portal that serves no readable text to anyone outside, so every quotation from it above is quoted as the Office of Academic Integrity quotes it. The two guidance PDFs also carry no printed date; their file metadata places them in April and May 2024.

No transcript notation

This is a meaningful contrast with several other large US universities:

“All academic misconduct records are confidential and housed with the Office of Academic Integrity. Grades of ‘F’ for academic misconduct receive no unique notation on your academic transcript. It will simply appear as an ‘F’.”

And there is a route back: “Any student who receives a grade of ‘F’ for academic misconduct will be allowed to repeat the class.”

Compare UCF, where a Z designation attaches to the grade and becomes permanent after a second finding, or Penn State, where an XF marks the failure as misconduct for up to two years. Liberty keeps the record internal.

Grade penalties are chosen by the professor from a fixed set — “Major, Minor, and Other” — and “students may petition for a lesser penalty through the appeal process”. For “severe and/or reoccurring academic misconduct, the sanctions of suspended or expulsion may be considered”. The full sanction ladder sits in a section of the Liberty Way we could not retrieve, so we are not going to describe it.

Reporting, appeals and the seven-day window

Four procedural facts that matter more than they sound:

• Reporting is mandatory and intent-blind. “Any incident of academic misconduct, regardless of known or perceived intent, should be reported.”
• One report per assignment. Filing separately for each “assists in alerting the university of possible patterns of misconduct.” Several small cases can therefore add up to a pattern.
• There is no time limit. “The Liberty Way does not place any time limitations on when academic misconduct can be reported.”
• The appeal window is seven calendar days from official notification, and “appeals do not go to the professor for review” — they go to “the academic leadership of that department”. You get two: “If the second appeal is denied, the decision is considered final.”

Seven calendar days is short. If a report lands, diarise the deadline before you do anything else, then assemble your material — drafts, dated version history, the settings you used in any writing tool. See what to do when you are accused and how to prove you wrote it yourself.

If a detector score is what started this, our guide to an AI detector flagging your thesis sets out what that number does and does not establish on its own.

The standard of proof is preponderance of the evidence — “the allegation must be proven by the greater weight of the evidence, in other words, a determination of what is more likely than not to have occurred”. Liberty adds that meeting it “should always involve a thorough review of all available information”, of which AI detection results are one item among several.

One more warning from the same office, about a common trap: “there are websites that contain complete or partial assignments from current and former LU students”, and accessing them is prohibited, since “stealing, buying, receiving, selling or transmitting coursework of any kind” breaches the Liberty Way.

Before you submit

• Turn off generative features in your writing aid. Liberty recommends it explicitly, and it removes the risk.
• Accept corrections, decline rewrites. That is exactly where Liberty draws the line.
• Know the limitations list if you write in STEM, in lists and tables, or in a second language.
• Keep drafts and version history. A flag is a starting point for a discussion; your process is what ends it.
• If a report lands, diarise seven calendar days.

If you want a read on your own text before you hand it in, our AI checkis free up to 1,500 words, with no account and no name — and note Liberty’s own point that work should stay inside its secure platforms where an assessment is concerned.

Sources

• Liberty University, Office of Academic Integrity: Artificial Intelligence and Academic Integrity — Faculty Guidance — no printed date; file metadata dates it to May 2024.
• Liberty University: Using Artificial Intelligence Responsibly — student announcement; no printed date, metadata April 2024.
• Liberty University: Office of Academic Integrity — definitions, sanctions, reporting and appeals.
• Liberty University: The Liberty Way — served through a JavaScript portal; quoted here only as the Office of Academic Integrity quotes it.

This page is orientation, not legal advice. What binds you is The Liberty Way and your course requirements.