Do I need a developer or technical experience?

Not at all. Paste your website URL, choose your settings, copy the embed code. If you can copy and paste, you can use RaddBot.

Does it work on Salla, Zid, and Shopify?

Yes, all of them. One embed code that works on Salla, Zid, Shopify, WordPress, and any site that accepts HTML.

How does the bot understand different Arabic dialects?

We use AI models trained on Gulf, Egyptian, and Levantine dialects. The bot detects the dialect from the first message and replies in it.

How long from sign-up until the bot is live?

Under 2 minutes. Create an account, enter your URL, customize settings, paste the embed code. Done.

What's the difference between RaddBot and ChatGPT?

ChatGPT is a general tool. RaddBot is trained on your specific website content. It knows your products, prices, and policies, and replies in your customer's dialect.

How does RaddBot test Arabic language accuracy?

We run 17 automated tests across 7 categories: Arabic morphology, Gulf dialect, Egyptian dialect, Arabizi (Latin-script Arabic), cross-language matching, synonyms, and short queries. Each test embeds two text samples and measures cosine similarity against a minimum threshold.

What technology does RaddBot use for embeddings?

RaddBot uses a 768-dimensional embedding model with task-type optimization. Different task types (query, document, QA) are applied depending on context, improving retrieval accuracy compared to a one-size-fits-all approach. Arabic text is also normalized to handle tashkeel, hamza variants, and zero-width characters.

Do other Arabic chatbots publish accuracy benchmarks?

No. As of March 2026, no Arabic chatbot platform publishes independently verified accuracy benchmarks with reproducible methodology. Some claim accuracy rates of 95–99% without disclosing their test methodology or allowing independent verification. RaddBot is the first to publish transparent, reproducible Arabic AI benchmarks.

What is cosine similarity and why does it matter?

Cosine similarity measures how semantically close two pieces of text are, on a scale from -1 to 1. A score of 1 means identical meaning. For Arabic retrieval, scores above 0.5 indicate strong semantic matching. RaddBot averages 0.722 across all 17 tests — well above the relevance threshold.

How often are the benchmarks updated?

Benchmarks are re-run whenever we update our embedding model, normalization pipeline, or retrieval logic. The date of the last test run is always displayed on this page.

Why does Arabic need specialized embedding benchmarks?

Arabic has unique challenges: 10,000+ word forms per root, multiple dialects (Gulf, Egyptian, Levantine), right-to-left script, and widespread use of Arabizi. Generic English-first models typically show a 15–40 point accuracy drop on Arabic tasks. Purpose-built Arabic benchmarks ensure these gaps are measured and addressed.

Verified Transparency

Arabic AI Accuracy Benchmarks

The only Arabic chatbot platform that publishes transparent, reproducible accuracy scores. 17 tests. 7 categories. 100% pass rate.

100%

Pass Rate

Total Tests

100%

Pass Rate

0.722

Avg Similarity

Categories Tested

Why We Publish Our Scores

No Arabic chatbot publishes benchmarks

Other platforms claim 95–99% accuracy without disclosing methodology. We show every test case, every score, every threshold — because we can.

Reproducible by anyone

Our test suite is open. Run the same tests on any platform and compare. We believe accuracy claims should be verifiable, not marketing.

Updated with every model change

Whenever we update our embedding model or retrieval pipeline, we re-run all 17 tests and publish the results here. No hiding regressions.

How RaddBot Compares

Metric	RaddBot	Industry Standard	Source
Published accuracy benchmarks	17 public tests, fully reproducible	No Arabic chatbot publishes verified benchmarks	Industry survey, Mar 2026
Arabic dialect coverage	Gulf, Egyptian, Arabizi — all tested	Most platforms support MSA only	Multilingual LLM performance study, 2025
Arabic vs English accuracy gap	0% gap — 100% pass rate on Arabic	15–40 point drop is typical for Arabic	Multilingual LLM performance study, 2025
Self-service resolution rate	Built for Arabic-first retrieval	Only 14% of issues fully resolved via self-service	Customer service industry report, 2024 (n=5,728)

Total Tests2

Why Arabic Needs Specialized Benchmarks

10,000+ Forms Per Root

Arabic has 10,000+ word forms per root. RaddBot handles morphological variants so customers find answers whether they type the singular, plural, or verb form.

Multi-Dialect Coverage

Gulf, Egyptian, and Levantine dialects are all understood. A customer in Riyadh writing in Gulf Arabic gets matched to the same content as someone writing in MSA.

Arabizi Support

Even Arabizi (Latin-script Arabic like 'keef asawwi order') is matched correctly to native Arabic content. No customer query falls through the cracks.

Closing the Accuracy Gap

Generic English-first models typically show a 15–40 point accuracy drop on Arabic tasks. RaddBot was built Arabic-first to close this gap completely.

Methodology

Embedding Model

We use a 768-dimensional embedding model optimized for Arabic, with task-type-specific embeddings (query, document, QA) and Arabic text normalization (tashkeel, hamza, tatweel stripping).

Thresholds

Thresholds are set per category based on linguistic difficulty. Arabizi has a lower bar (0.30) while synonyms require higher similarity (0.60).

Test Frequency

Tests are re-run with every model or pipeline update. The date shown reflects the latest verified run.

Last verified: March 27, 2026

Frequently Asked Questions

Build on proven Arabic AI

RaddBot is the only Arabic chatbot that proves its accuracy with transparent, reproducible benchmarks. Try it on your store for free.

Get started free