How we built this

• Arab and Gulf houses (about 2,100 perfumes): collected automatically from the houses' own online stores (Lattafa, Amouage, Ajmal, Arabian Oud, Swiss Arabian, and 15 others, 20 houses in total). The scraper kept Arabic and English names, listed notes, concentration, and images.
• International brands (about 960 perfumes): selected from Parfumo, a public fragrance database, by number of reviews, excluding houses already covered above.
• Missing notes: where a store listing had no notes, some were filled from public Fragrantica and Parfumo listings. A note was kept only when the automated match was confident.
• Scent families: 18 families, using the Edwards fragrance wheel as the reference framework.

Perfume descriptions, brand histories, note and accord descriptions, and articles are drafted by AI language models. Models used so far:

• google/gemini-3.1-flash-lite-preview (main drafting model, via OpenRouter)
• openai/gpt-5.4-nano (fallback drafting model)
• OpenAI GPT-5.5 (used for a refresh of 35 articles in July 2026)

We may change models over time and will list them here when we do.

Arabic and English are written in separate passes from the same structured data. Some article pairs share a title, but the Arabic body is not a translation of the English body.

Some brand histories were drafted with a Wikipedia summary as background. In a number of cases that summary was about a different subject with the same name, and unrelated material ended up in the text. We are removing those passages.

Drafts go through automated checks (length, Arabic script in Arabic fields, internal links). These catch formatting problems, not factual errors. Most text on Mijmara has not been individually reviewed by a person.

Western brand and perfume names are written in Arabic script following the convention used by Gulf retailers (faces.com, ounass.com, noon.com). For example: Dior → ديور, Chanel → شانيل, Tom Ford → توم فورد. URL slugs stay in Latin script (for example, /perfume/dior-sauvage-eau-de-parfum).

Most brands do not publish accords, so we estimate them. A ridge-regression model, trained on a large external set of perfumes whose notes and accord strengths are known, predicts how strongly each accord appears from a perfume's note list. Each perfume's primary scent family is then derived from its leading accords.

These are estimates from the note list, not measurements. About one perfume in ten still carries accords from our earlier method (an IDF-weighted note score), which we know over-assigns "woody". Those perfumes are being re-scored.

Descriptions only change when we deliberately regenerate or correct them. We add perfumes as data becomes available. Listing pages (accords, notes, families) refresh at most once a day.

• About 4% of perfumes have no note data. Their records state when note data is unavailable.
• About 7% of perfumes have no official bottle image. We show a paper placeholder.
• The text is AI-written and mostly not individually reviewed by a person. It can contain factual errors.
• Accords and families are statistical estimates, not measurements.
• Some Arabic note names are mistranslated. We are correcting them.
• We do not test perfumes, and the site has no user reviews yet.

Perfume pages may include affiliate purchase links. If you buy through them, we may earn a commission at no extra cost to you. Disclosure appears beside the purchase information. These links do not affect which perfumes we list or what the descriptions say.

Official-source facts are now live for nine Arab perfume houses. Armaf is excluded because its terms prohibit automated collection. Each fact shows its source and fetch date. “Verified” means a second automated check confirmed that field on a sample with at least 98% agreement; it does not mean human review.