Decoding Your Originality Report
The SafeAssign AI Checker report is not a simple pass/fail grade. It provides a Combined Integrity Index. This metric synthesizes two distinct forensic layers:
Similarity Matching
Checks for direct text matches against the Global Reference Database (web pages, journals, and past student submissions).
Linguistic Forensics
Analyzes statistical probability (perplexity & burstiness) to detect synthetic content generated by AI models.
Risk Level Interpretation
Low Risk (Green Zone)
Status: Original Work. The text exhibits natural human variation in sentence structure (“burstiness”). Common phrases or citations may trigger minor percentages, which is normal for academic writing.
Action: No revision typically required.Medium Risk (Yellow Zone)
Status: Hybrid Composition. Indicates a mix of original thought and potentially AI-assisted editing or generation. Often seen when students use tools like Grammarly GO or ChatGPT to rewrite paragraphs.
Action: Instructor review recommended. Check version history.High Risk (Red Zone)
Status: AI Generated. Significant statistical patterns consistent with Large Language Models. The text lacks the semantic entropy (randomness) typical of human cognition.
Action: Formal academic integrity audit required.Understanding Deep Metrics
Global Reference Match
This percentage represents text found verbatim in our database of over 10 million student submissions and academic journals.
- Direct quotes should be excluded if cited.
- Bibliographies often trigger false positives here.
AI Model Signature
Identifies specific “watermarks” of logic used by models like Claude 4 or GPT-5.
- High consistnecy in sentence length often triggers this.
- Overuse of transition words (“Furthermore,” “In conclusion”).
Perplexity & Burstiness
Perplexity
The measure of how “surprised” a model is by the text. AI maximizes probability, so it writes with Low Perplexity (predictable). Humans write with High Perplexity (unpredictable).
Burstiness
The variation in sentence structure and length. AI tends to be monotonous (flat line). Humans vary short and long sentences dynamically (spikes).
Common False Positives
Certain writing styles or tools can inadvertently trigger higher AI scores without malicious intent. Be aware of these factors:
-
Grammarly & Spell Checkers
Heavy use of “Rewrite for Clarity” features can strip natural human “burstiness,” making text appear machine-generated.
-
Technical & Legal Writing
Formulaic documents (lab reports, legal briefs) naturally use repetitive structures and low-entropy language.
-
Non-Native English Speakers
Writers with limited vocabulary may use simpler, more predictable sentence structures that algorithms can mistake for AI.
FAQ & Appeals
How can I appeal a high AI score?
If you believe your work was flagged incorrectly, present your instructor with Version History (from Google Docs or Word) showing the evolution of your document. Drafts, outlines, and research notes are strong evidence of original work.
Why is my bibliography flagged?
Bibliographies often list standard citations found in millions of other papers. This triggers “Similarity Matching” but usually not “AI Detection.” Instructors typically exclude references from the final score.
Does this detect paraphrasing tools like Quillbot?
Yes. Paraphrasers often leave specific syntax patterns that our linguistic model detects as “unnatural flow,” even if the words themselves are unique.
Can I resubmit my paper to lower the score?
This depends on your institution’s policy. If allowed, we recommend focusing on adding personal analysis and varying your sentence structure rather than simply swapping synonyms.
What to do next?
Improve Your Integrity Score
Academic Fairness Clause
This report is a screening tool, not a verdict. A high probability score does not prove misconduct. It indicates a need for further human review. Instructors must verify results using version history, oral defense, or comparative analysis of previous work.