Evaluate AI detection for an Arabic dataset. Start with a language compatibility test. Record the response before choosing a threshold for human review.
Save an Arabic article you have permission to analyse as article-ar.txt in UTF-8. Install the Python package with pip install its-ai and set ITS_AI_API_KEY to an Enterprise API key. Run one file before processing a batch. The example prints the returned score and segments when the request succeeds, and reports an English-only validation response if the endpoint rejects the language.
from pathlib import Path
from its_ai import ItsAIClient, OnlyEnglish
text = Path("article-ar.txt").read_text(encoding="utf-8")
with ItsAIClient() as client:
try:
result = client.analyze_text_v2(text)
except OnlyEnglish:
print("This endpoint requires English for this request.")
else:
print(result.score, result.segments)
Deep Scan estimates AI likelihood for individual sentences. Use it as one signal when reviewing Arabic datasets, alongside provenance records and human-labelled evaluation samples.
Test the API with a representative file
Use the API to add text checks to your pipeline. Consult the documentation for current pricing, supported requests and text-length limits . Plan your workload around the applicable rate limits and validate the integration on your own data before scaling it.
Plan data handling before integration
Review the privacy policy and your organisation's requirements before sending a dataset. Discuss retention, access and deployment needs with the team; avoid uploading data you are not authorised to process.
I've tried many detectors, but It's AI remains the most accurate and convenient - especially when working with massive datasets.
Felix Enriquez
Fullstack/ML Engineer at Takumi Studio LLC
Flexible pricing, rich functionality, and a clear interface. It's a must-have tool for any ML engineering team.
Dinara Darkulova
Big Data Engineer at Google Cloud Platform
I rely on It's AI in every ML project. It's the best tool for checking model responses and optimizing training workflows.
Rahul Behal
Machine Learning Operations Engineer at Arctic Wolf
I integrate It's AI API into my custom LLMs. Fine-tuning goes faster, and I'm always aware of the quality of the data I use.
Anna Berger
Machine Learning Engineer at Bloomfield Robotics
Arabic AI detection benchmarks
Simple
Accurate
Affordable
AI detection forArabic and English writing
Our Arabic evaluations report 98.7% accuracy on ASJP and 99.8% on DetectRL-X. These datasets test different text distributions; their results do not guarantee accuracy on every Arabic document. Arabic benchmark details →
Arabic benchmark accuracy (%)
ASJP
98.7
DetectRL-X
99.8
Full Arabic Support
98.7% accuracy and 0.5% FPR on the Arabic ASJP benchmark.
98.7%
accuracy on ASJP
Arabic academic abstracts in our ASJP evaluation. A benchmark result is not a guarantee for an individual document.
False Positive Rate
FPR: 0.89% on the AR-AES corpus of authentic Arabic student essays. Review individual results in context.
Evaluate the detector on your own data
Evaluate labelled samples, inspect errors and choose thresholds that suit your use case before automating a filtering decision.
Check for AI
Check source matches
Validate results
Arabic AI Detector FAQ
Questions about checking Arabic text?
Validate thresholds on representative labelled data first. Length, topic, dialect and editing can affect results. Record false positives and false negatives, review uncertain samples and reassess when your data distribution changes.
No. Check the original Arabic text directly. The English interface is separate from the language of the document, and translating first would change the text being assessed.
Results depend on the text, its length, topic and editing. Consult the accuracy page for Arabic benchmark details. Multilingual benchmark figures are not an accuracy guarantee for an individual Arabic document.
Treat mixed-language results with extra care. Review passages in context and, where useful, inspect substantial Arabic and English sections separately. Short fragments may not give enough context for a useful assessment.
Use the free allowance currently shown in the app to try a scan. Check the current plans for document limits, batch uploads, reports and API access.
Send one representative file to the v2 text endpoint with your Enterprise key. Record the response before starting a batch. The Python SDK exposes a language validation error when the service requires English.
Deep Scan provides sentence-level AI likelihood. It does not establish the origin of each word or prove whether a training example was generated by a particular model.
The accuracy page reports Arabic results for ASJP and DetectRL-X and human-writing false positives on AR-AES. MGTD and RAID results shown elsewhere on the site evaluate English detection and should not be treated as Arabic accuracy measurements.
Validate the rule on labelled data from your target distribution first. Measure false positives and false negatives, keep provenance records and review uncertain samples before applying an automated filter.
Review the privacy policy, access requirements and your organisation’s data-handling rules. Confirm retention and integration requirements with the team before uploading confidential or restricted material.
![[object Object]'s avatar](/_next/image?url=%2Favatars%2Fimg_22.webp&w=96&q=75){kind=small}
![[object Object]'s avatar](/_next/image?url=%2Favatars%2Fimg_23.webp&w=96&q=75){kind=small}
![[object Object]'s avatar](/_next/image?url=%2Favatars%2Fimg_24.webp&w=96&q=75){kind=small}
![[object Object]'s avatar](/_next/image?url=%2Favatars%2Fimg_25.webp&w=96&q=75){kind=small}
