AI Detector Glossary: Key Terms Behind Turnitin0 Accuracy

AI detection is a field where imprecise language causes real harm. A student who confuses "AI score" with "similarity score" may panic over a number that has nothing to do with AI writing. A student who does not know what the asterisk (*%) symbol means may assume the worst when Turnitin is actually signalling low confidence rather than a high match.

Turnitin's own documentation and independent research make the stakes clear. In a study of 180 essays generated by GPT-5.6-Sol (156,955 words), Turnitin's word-level accuracy reached 97.88%, ranging from 88.81% in Physics to 99.67% in Business Administration [TT0-2026-0008]. In a separate study of 170 Claude Fable-5 essays (131,451 words), word-level accuracy was 99.01% [TT0-2026-0007]. Accuracy at that level means the terminology in your report is not academic decoration — it is the difference between a clean submission and an academic misconduct hearing.

The Vocabulary Problem in Practice

Most students encounter AI detection vocabulary for the first time in a stressful email. Learning the terms in advance converts that email from a threat into a readable document.

AI Writing Detection

AI writing detection is the process of estimating whether a passage of text was produced by a large language model rather than a human author. Turnitin's model does not search for matching sources the way plagiarism detection does; it analyses statistical properties of the prose itself.

AI Score

The AI score is the percentage of the document that the detector classifies as AI-generated. It is a proportion, not a probability, and it is calculated at the sentence level before being aggregated.

Confidence Threshold

The confidence threshold is the minimum certainty the model requires before it will display a precise number. Turnitin's threshold is 20%. When AI detection falls below that level, Turnitin shows *% instead of an exact percentage. This is a deliberate design choice: it prevents the system from publishing a number it does not trust.

Key point: Seeing *% on a Turnitin report does not mean "unknown" or "error." It means the detector's confidence fell below 20%, which is generally good news for the writer.

False Positive

A false positive occurs when human-written text is flagged as AI-generated. This is the single most damaging error in academic AI detection, because it penalises honest work.

The evidence on false positives is genuinely reassuring. In 504 human-written research papers from the PLOS corpus (135,712 words), Turnitin's word-level accuracy was 100.0%, with no false positives across 18 majors [TT0-2026-0005]. In 340 human-written ESL essays (263,329 words), accuracy was again 100.0%, with zero false positives across all domains and buckets [TT0-2026-0004]. ESL writers are the group most often assumed to be at risk, and the data does not support that assumption for Turnitin specifically.

False Negative

A false negative is AI-generated text that the detector misses. This is the error that matters to institutions rather than to honest students. Turnitin's accuracy varies by generator and discipline, which is why the studies above report ranges rather than single figures.

Sentence-Level Flagging

Turnitin flags AI writing at the sentence level, not only at the document level. A report may show a modest overall AI score while highlighting a handful of specific sentences. This granularity is why reading the report matters more than reading the headline number.

Perplexity

Perplexity measures how surprised a language model is by the next word in a sequence. Human writing tends to have higher perplexity because humans make unexpected word choices, use idioms, and vary register. AI writing tends toward lower perplexity because models select statistically likely continuations.

Low perplexity is not proof of AI authorship. Highly formulaic human writing — legal boilerplate, lab methods sections, standardised introductions — also produces low perplexity.

Burstiness

Burstiness measures variation in perplexity across a document. Humans are "bursty": we write a dense technical sentence, then a short plain one, then a long clause-heavy one. AI text tends to be more uniform, with sentence lengths and complexity clustering around a mean.

Perplexity and burstiness are usually discussed together because neither is decisive alone. A document with uniformly low perplexity and low burstiness looks statistically machine-like. A document with high burstiness looks more human even if individual sentences are predictable.

Tokenisation

Tokenisation is how a model breaks text into units (tokens) before analysis. It matters for students because tokenisation explains why unusual formatting, heavy symbol use, or non-standard punctuation can distort detection signals.

Training Data and Model Drift

Detection models are trained on text from known generators. When a new model appears — Gemini 3.5 Flash, Claude Fable-5, GPT-5.6-Sol — the detector must generalise. Turnitin's measured accuracy on Gemini 3.5 Flash essays was 98.35%, ranging from 94.36% in Information Technology to 99.82% in Business Administration [TT0-2026-0003]. That range shows model drift is real but bounded.

Plagiarism and Similarity Terms

Similarity Score

The similarity score is the percentage of a document matching sources in Turnitin's comparison database. It is not a plagiarism verdict. A high similarity score can reflect correctly quoted material, a standard methods paragraph, or a bibliography.

Similarity Checker

A similarity checker compares submitted text against journals, web pages, and previously submitted student papers. Turnitin0's Turnitin similarity checker report is delivered alongside the AI report in the same checkout, so you can review both signals together rather than guessing which one your professor will prioritise.

Source Matching and Overlap

Source matching identifies which specific sources overlap with your text. Overlap is the matched text itself. Reviewing overlap lets you confirm that quotations are properly attributed and that paraphrases are genuinely rewritten rather than lightly reworded.

Repository vs Non-Repository Submission

A repository submission adds your paper to the database that future submissions are checked against. A non-repository submission does not. This distinction is central to how Turnitin0 operates: files are not added to the Turnitin student paper database, and reports are not shared with third-party databases. Users can also delete files from their account.

Student Paper Database

The student paper database is the archive of previously submitted work used for cross-comparison. It is why submitting a draft to a repository-based service can create a future self-match problem. Non-repository checking avoids that entirely.

Humanizer and Rewriting Terms

AI Humanizer

An AI humanizer rewrites AI-drafted text so it reads as human-authored. Turnitin0's AI humanizer is built for text drafted with ChatGPT, Claude, or Gemini, and it preserves meaning, citations, headings, and.docx formatting.

Meaning Preservation

Meaning preservation means the rewrite keeps your argument intact. A humanizer that changes your claims to lower a score has failed, regardless of the number it produces.

Citation and Formatting Retention

Citation retention means references, in-text citations, and heading structure survive the rewrite. Formatting retention means the.docx structure survives too. Both matter because reformatting a 5,000-word paper by hand is where students lose hours.

Re-checking Humanized Output

Turnitin0 re-checks 98.2% of humanizer orders with Turnitin, which is how the humanizer score promise is verified: lower the Turnitin AI score to *% or below 20%, or even 0%, or receive a full refund.

How Turnitin0 Reports Differ From Generic Detectors

Generic detectors produce a number. Turnitin0 produces the two PDFs that mirror what professors see in their LMS. That difference is the whole point of pre-submission checking.

Competitors approach this differently, and it is worth being fair about their strengths. Humanizeaitext.ai advertises unlimited free words, file uploads, keyword freezing, and a Chrome extension, with a claimed 4.9/5 on Trustpilot across 189 reviews — though all benefits are vendor claims. Getsolved.ai bundles detection, rewriting, grammar checking, and paraphrasing into one workspace with claimed support for six languages. Ref-n-Write targets academic writing with cross-referencing, proofreading, a phrasebank, and a free trial. SupWriter offers 300 free credits, Apple App Store and Chrome availability, and MCP/API integrations. Winston AI markets an AI image detector and a broad tool suite. EssayDone claims bypass of 12+ detectors. Turnitindetectorai.com and MyDetector.ai both offer free, no-signup checking with sentence-level highlighting.

What none of them provide is the actual Turnitin report format. A free detector's percentage is not a Turnitin percentage, and a vendor claim of "bypassing Turnitin" is not the same as a re-checked Turnitin AI report. Turnitin0 is an independent service and is not affiliated with Turnitin, LLC — it simply gives you the report itself.

Reading Your Report: A Practical Walkthrough

Step 1: Check the AI Score and the Asterisk

If you see a precise percentage, read it against the sentence highlights. If you see *%, the detector's confidence was below 20%.

Step 2: Read the Similarity Report Separately

Do not blend the two numbers. A 15% similarity score and a 15% AI score describe completely different things.

Step 3: Inspect Flagged Sentences

Sentence-level flags tell you where to revise. A single flagged sentence in an otherwise clean document is a targeted edit, not a rewrite.

Step 4: Decide Whether to Humanize

If the AI score is high and the text was drafted with ChatGPT, Claude, or Gemini, the humanizer is the appropriate tool. If the text is your own writing, revise the flagged sentences directly.

Step 5: Re-check Before Submitting

Re-checking is how you confirm the revision worked. It is also how Turnitin0 verifies 98.2% of humanizer orders.

What Real Users Report

Reviewers describe the same pattern repeatedly: speed, clarity, and two reports at once.

• Raini Dipré (CA), 2026-09-17, 5 stars: described the process as easy, fast, and efficient, with the report arriving much faster than expected.
• may zin (SG), 2026-09-15, 4 stars: received a complete report in about 20 minutes and downloaded the AI and similarity reports at the same time.
• daniela pellegrini (GB), 2026-09-07, 5 stars: had used Turnitin0 several times, found reports delivered quickly, and found Humanize helpful when revising.
• b c (US), 2026-09-04, 5 stars: used the site for assignments, plagiarism checking, and awareness of AI.
• Shawn Thakur (AU), 2026-08-24, 5 stars: easy to use and arrived on time.
• Encrypted (GB), 2026-08-14, 5 stars: called it the best site for Turnitin scans, authentic and simple to use.
• Shubham Pachauri (IN), 2026-08-14, 5 stars: found it easy, quick, and helpful, and liked Humanize for sounding more natural while keeping the original meaning.
• Taksh Patel (AU), 2026-08-13, 5 stars: described it as a great service that is 100% legit and works.

These are individual experiences, not a statistical sample, but the consistency across countries is notable.

Honest Limitations You Should Know

Trust requires stating limits plainly. Turnitin0 accepts English documents only, for both checking and humanizing. The checking service requires more than 300 words and fewer than 30,000 words, with files under 20 MB, and accepts.docx,.pdf, or.txt. The humanizer accepts files under 90 MB and only.docx or.txt uploads, and there is no free word quota or free trial for the humanizer. Turnitin displays *% rather than an exact figure when AI detection falls below its 20% confidence threshold. On Trustpilot, the profile shows a TrustScore of 4.3/5 labelled Excellent from 9 reviews in the last 12 months, with 89% five-star and 11% four-star — but the company has not recently invited customers to review, so those reviews may not be representative. In rare queue spikes, delivery is still guaranteed within 30 minutes.

Conclusion

The vocabulary in this glossary — perplexity, burstiness, confidence threshold, false positive, non-repository — exists to help you read a report accurately rather than guess at it. The most practical way to apply it is to see your own document through the same lens your professor will use. Turnitin0 delivers the Turnitin AI detection report and similarity/plagiarism report as two downloadable PDFs in one checkout, matches what professors see in their LMS, keeps your file out of the student paper database, and backs its humanizer with a score promise. Start with a Turnitin AI detector check on your draft, read the terms on your own report, and submit with evidence rather than hope.