Is 15% AI in Your Thesis Acceptable? What the Number Really Means

What Does a 15% AI Score Actually Mean on a Thesis?

A detection tool doesn't measure whether you used AI to write your thesis in some absolute sense. It measures how closely the statistical patterns in your prose resemble patterns typical of AI-generated text, then reports that likelihood as a percentage. A 15% reading usually means the tool found a modest amount of phrasing across your document that resembles AI-typical patterns, not that 15% of your words were literally generated by a chatbot and pasted in unchanged. Dense, formal academic writing, heavy use of field-specific terminology, and long complex sentences can all push a detector's score up even in fully human-written chapters, especially in disciplines where hedged, structured phrasing is the norm. Before deciding whether 15% is a problem, it helps to know that the number is a probability estimate, not a word count of AI-authored text.

A percentage from a detector describes how your writing patterns statistically resemble AI output — it doesn't describe how the thesis was actually produced.

Is There a Universal Threshold for Acceptable AI in a Thesis?

No. Unlike a plagiarism similarity score, where many schools do publish a rough ceiling, AI-detection percentages don't have anything close to an industry-standard cutoff, and 15% isn't a number that appears in any widely adopted policy as a bright line. Some graduate schools have published general statements about AI use in theses and dissertations, but the ones that mention detection tools at all typically frame the score as one input for a committee conversation, not an automatic gate. That means a 15% result that one advisor waves off without a second thought could trigger a meeting with another, purely based on how that program has chosen to handle AI so far. If you've been told a specific number by your program, that's your real threshold — the internet's general answer to how much AI is acceptable in a thesis won't override your own institution's stated policy.

How Do Thesis Committees and Graduate Schools Set Their Own Limits?

Where a percentage like 15% lands on the acceptable-or-not spectrum usually comes down to which layer of your institution is doing the judging, and each layer tends to weigh the number differently.

1. Graduate school policy: some universities have a blanket statement on AI use in theses, though it's more often about disclosure than a numeric ceiling
2. Department norms: departments in fields with heavier technical or quantitative writing sometimes treat higher baseline scores as normal, since detectors flag structured, formal prose more readily
3. Individual committee chair or advisor: many advisors set expectations directly with their own students based on writing they've seen before, rather than deferring to a university-wide number
4. Discipline-specific conventions: some fields already use AI tools for literature searching or citation management, which can shape how tolerant a committee is toward moderate AI-pattern scores
5. Defense-stage scrutiny: a score that raises no concern during a draft review can still come up as a question during the oral defense, particularly if a chapter reads noticeably differently from your earlier work

Does It Matter Which Chapter the 15% Shows Up In?

Yes, and this is one of the most overlooked parts of interpreting a thesis-wide score. An overall 15% could mean the entire document is evenly flagged at a low level, or it could mean your literature review is clean while your discussion chapter is scoring much higher — two very different situations that get averaged into the same headline number. Literature review and background sections tend to draw less scrutiny because summarizing existing scholarship is expected to sound formal and structured. Your original analysis, discussion, and conclusion chapters carry more weight in a committee's mind, because that's where your own scholarly contribution is supposed to be most visible. Before treating a 15% overall score as fine, it's worth checking the chapter-by-chapter breakdown if your tool provides one, since a single heavily flagged section can matter more than an evenly distributed low average.

An average score can hide a chapter that needs attention — the discussion and conclusion sections usually deserve the closest look.

Should You Disclose AI Use Even If the Score Is Only 15%?

Disclosure isn't only for cases where a detector flags a high percentage. If you used an AI tool at any point while drafting your thesis — for grammar polishing, outlining, translating your own writing, or summarizing background reading — most graduate programs' academic integrity codes ask you to say so, regardless of what any detector later reports. A low score doesn't retroactively make undisclosed AI assistance acceptable if your program's policy required disclosure in the first place, and a committee that later learns AI was used without being told tends to react more critically than one that was informed upfront. If your program has no explicit disclosure requirement, adding a short note in your acknowledgments or methodology section about how you used AI tools during the writing process is a low-cost way to get ahead of a question you might otherwise face at the defense.

What Evidence Should You Have Ready Before Your Defense?

A 15% detection score is easier to explain calmly if you can also point to normal evidence of your own authorship, which most thesis writers already have without realizing it counts as evidence.

1. Draft history: earlier versions of chapters saved in your word processor or cloud drive show your writing evolving over time, which is hard to fake after the fact
2. Advisor feedback trail: emails or tracked-changes comments from your committee across multiple drafts demonstrate an iterative process rather than a single generated document
3. Research notes and data files: raw notes, interview transcripts, or analysis scripts tied to your results support that the underlying work is yours even if a passage of prose reads oddly to a detector
4. Citation and source records: a reference manager library or annotated bibliography shows you engaged directly with the literature you're citing
5. A record of how you used any AI tool: if you used AI for editing or brainstorming, keep a general note of what it was used for and when, so you can describe it accurately if asked

Can You Lower a 15% AI Score Without Rewriting Your Own Argument?

If a chapter's score seems disproportionate to how it was actually written, the first step is finding out exactly which sentences are driving the number rather than guessing. NotGPT's AI Text Detection tool highlights the specific passages contributing to a score instead of returning one blended figure for an entire chapter, which makes it easier to see whether a genuinely human-written section is being flagged because of dense academic phrasing. Where a flagged passage does reflect a stiff, overly uniform sentence structure — something that happens naturally in formal thesis writing — the Humanize tool can vary phrasing and rhythm without touching your data, findings, or citations. This is a reasonable step before a defense if you're confident the substance is entirely your own work and just want the phrasing to read more like your natural academic voice, but it isn't a way to disguise AI-generated analysis that isn't actually yours, and no detector adjustment changes what your committee will ask you to explain in person.

Is a 0% AI Score Always Safer Than 15% for a Thesis Defense?

Not automatically. Committees that have dealt with AI-detection tools for a while are increasingly aware that scores can swing based on writing style, formatting, and even which detector was used, so a very low score isn't treated as automatic proof of anything either. What tends to matter more in practice is whether you can speak fluently and specifically about your own methodology, defend your findings under questioning, and account for how you produced the document if asked. A 15% score paired with a confident, detailed defense of your own work is generally in a stronger position than a 0% score from a student who can't explain their own analysis. Treat the percentage as one signal worth checking, not the deciding factor in whether your thesis gets accepted.