Compilatio Plagiarism and AI Detector: What to Verify Before You Trust a Score

What Is Compilatio and How Does It Work?

Compilatio started as a plagiarism-detection service built for the French education market and has since become one of the more widely deployed similarity-checking tools in French-speaking secondary schools and universities, with adoption extending into other European institutions as well. Its core function compares a submitted document against a large index of web pages, published academic work, and previously submitted student papers, then returns a similarity report showing which passages overlap with existing sources and how closely. That similarity percentage is the number teachers and administrators have relied on for years when reviewing student submissions for potential plagiarism. In the past few product cycles, the platform added a second layer to its reports: an estimate of how likely a passage is to have been generated by an AI writing tool, delivered as its own score sitting alongside the similarity figure rather than replacing it. Like most commercial detection vendors, it has not published the full technical details behind that AI-detection scoring — the exact signals it weighs and how it was trained are not public information. Availability of the AI-detection module, the exact report layout, and how prominently it's surfaced can vary by institution and subscription tier, so what one student or teacher sees in a Compilatio report is not guaranteed to match what someone at a different school sees. If you're trying to understand what your own report is showing you, checking with your institution's administrator or the vendor's own documentation is more reliable than assuming every deployment works identically.

How Is Plagiarism Detection Different From AI Detection?

Plagiarism detection and AI-writing detection solve two genuinely different problems, even when a single report presents both scores side by side. A similarity check works by matching text against an index of existing sources — web pages, journal articles, previously submitted papers — and flagging passages that overlap closely enough to suggest copying without attribution. It's fundamentally a search-and-compare process: the tool isn't judging who wrote the sentence, only whether that exact sentence (or something very close to it) already exists somewhere else. AI-writing detection works on an entirely different principle. Instead of comparing text to a source index, it looks at statistical properties of the writing itself, most commonly perplexity (how predictable each word choice is given the surrounding context) and burstiness (how much sentence length and structure vary across a document). Language models tend to produce text that is smoother and more statistically uniform than typical human writing, and detectors are built to pick up on that pattern. A passage can score low on similarity and high on AI probability, or the reverse, because the two checks are looking for completely different signals. Putting both numbers on one dashboard is a convenience, not a sign that they measure the same thing, and treating a low similarity score as proof a passage wasn't AI-written (or treating a high AI score as proof of plagiarism) misreads what each figure is actually telling you.

How Accurate Is Compilatio's AI Detection Feature?

Compilatio has not published independently audited accuracy figures for its AI-detection feature, which puts it in the same position as most commercial detectors on the market — vendors generally describe their tools as effective aids for review rather than as infallible verdicts, and the vendor's own materials frame the AI score as one signal among several rather than a standalone determination. Peer-reviewed research on AI detectors more broadly, covering tools built on similar perplexity- and burstiness-based approaches, has repeatedly found meaningful false positive rates on specific categories of writing, along with reduced accuracy on text that has been paraphrased or lightly edited after being drafted with an AI tool. There's no public reason to assume this AI-detection module is immune to the same categories of error that affect comparable tools, and no independent third-party benchmark currently available lets you verify exactly where its accuracy stands relative to competitors. That's a reasonable basis for caution rather than dismissal: the score can be a useful flag that prompts a closer look, but it isn't strong enough evidence on its own to support a high-stakes decision like a grade penalty or an academic integrity finding.

"A detection score is a reason to look more closely at a piece of writing, not a verdict on its own — that's true whether the tool is Compilatio, Turnitin, or anything else on the market."

What Should You Verify Before Trusting a Compilatio Score?

Because Compilatio's reports and feature availability can differ by institution, subscription tier, and even by assignment settings a teacher configures, it's worth confirming a handful of specifics before treating any score as reliable. Pricing, exact integrations, and which detection modules are switched on are also the kind of detail that changes over time, so verifying directly with your institution or with Compilatio is more dependable than relying on secondhand descriptions, including this one.

1. Confirm whether the AI-detection module is actually enabled for your course or institution — similarity checking and AI detection are configured separately in many deployments
2. Ask what threshold your teacher or institution treats as noteworthy, since Compilatio itself does not set a universal pass/fail cutoff
3. Check which languages the AI-detection feature currently supports well, since accuracy and coverage can vary across languages
4. Find out whether the report you're looking at reflects the most recent version of your document or an earlier draft that was previously submitted
5. Ask your institution what happens procedurally after a flag — whether a human reviewer looks at the file before any consequence follows
6. When in doubt about a specific feature, pricing, or integration detail, check Compilatio's own documentation or your institution's guidance rather than assuming it matches what you've read elsewhere

False Positive Risks and Other Limitations to Know About

The categories of writing most likely to trigger false positives on AI detectors generally are well documented, and there's no strong reason to expect this feature sits outside that pattern. Non-native speakers writing in a second language often produce prose that is more grammatically regular and less idiomatically varied than native-speaker writing, which can read as statistically 'smooth' in a way that overlaps with how AI-generated text tends to score. Heavily edited writing — a draft that has passed through several rounds of revision with a tutor, writing center, or grammar-checking tool — can lose some of the irregularity that detectors use as a signal of human authorship, even though every word was written and revised by the student. Formal or technical writing, including lab reports, legal-style analysis, and highly structured business writing, tends to read as low-perplexity even when written entirely by a person, simply because the genre itself rewards conventional phrasing. None of this means the tool's scores are unreliable across the board; it means the same caveats that apply to any commercial AI detector — human review before any consequence, corroborating evidence beyond a single score, and extra care with ESL and heavily edited writing — apply here too.

How Compilatio Fits Into an Institution's Academic Integrity Workflow

In most deployments, a Compilatio report — whether it's the similarity score, the AI-detection score, or both — is designed to sit in front of a human reviewer rather than trigger an automatic outcome. A teacher or academic integrity officer typically looks at the flagged passages directly, checks whether the highlighted text is properly cited or genuinely resembles the student's other work, and decides whether the file warrants a conversation with the student before any formal process begins. Institutions vary considerably in how they configure and act on Compilatio results: some set conservative thresholds and reserve the tool for spot checks, others review every submission as a matter of course, and specific escalation procedures are set at the institutional level rather than by Compilatio itself. If you're a student or teacher trying to understand what a flag actually means for a specific assignment, your own institution's academic integrity policy — not general information about how the tool works — is the authoritative source for what happens next.

Using NotGPT as a Pre-Check or Second Opinion

Running your own writing through a separate detector before submission is a reasonable way to catch a potential false positive early, and it also gives you an independent second reading rather than relying on a single tool's output. NotGPT's AI Text Detection gives a sentence-level probability breakdown with highlighted passages, built specifically for AI-writing detection rather than as an add-on to a plagiarism-matching product, which makes it a useful independent comparison point alongside a Compilatio report. If your own writing gets flagged unexpectedly by either tool, checking it against a second, differently built detector is one of the more practical ways to tell whether the flag is likely a false positive tied to your writing style, or a passage worth genuinely reviewing before you submit. For students specifically, the goal of a pre-check is not to game any detector but to catch stylistic patterns — overly uniform sentence structure, an unusually generic passage — that might draw unwanted scrutiny, and to revise them so the final submission reads clearly as your own work.

1. Paste your draft into NotGPT's AI Text Detection before submitting it anywhere a Compilatio check might run
2. Compare which specific sentences get flagged, since overlap between two independently built tools is a stronger signal than either score alone
3. If a passage is flagged and it genuinely is your own writing, revise for natural variation in sentence length and add specific, concrete detail rather than rewriting from scratch
4. Keep drafts and revision history for anything you submit, since documentation is the most persuasive evidence if a flag is ever formally reviewed
5. Treat any single score — from Compilatio, NotGPT, or any other tool — as one input into a human decision, not a conclusion on its own

"Cross-referencing an independent detector isn't about disputing a score for the sake of it — it's about giving a human reviewer more information before anything gets decided."