Arabic Doesn't Fail at Translation. It Fails at Render Time, Inside a Build You Don't Own.

Arabic localisation of a licensed HTML5 catalogue fails at render time, not translation: WebGL and bitmap-font game builds cannot shape Arabic script at all. The words are the cheap part. The expensive part is that the engine drawing them places one glyph after another in the order it was handed, with no idea that Arabic letters change shape depending on their neighbours — and that the build you licensed is a sealed bundle you may have no contractual right to open.

🌍 The Audience Is Large. The Web It Reads Is Not.

Start with the demand side, because it is the part everyone gets right. Niko Partners' MENA-3 series, which tracks Egypt, Saudi Arabia and the UAE, put those three markets at roughly $2.2 billion in 2025, up about 8.5% year on year, with 85.76 million gamers and 76% of them under 35.

Treat that number as a band rather than a fact. Forecasts published for the same three countries over the last two years have ranged from about $2.0 billion for 2024 to $3.1 billion for 2025 in an earlier vintage, and Niko's own longer forecast lands near $2.8 billion in 2029 — which is lower than some 2025 estimates that were circulating in 2023. The methodology and the cut-off matter more than the headline. If a deck quotes one MENA figure with no year and no publisher, it is decoration.

The number that actually changes how you plan is the spread inside the region. Niko's reported 2024 ARPU was $84.60 in the UAE, $54.89 in Saudi Arabia and $3.39 in Egypt, and revenue split roughly 59% Saudi, 34% UAE, 7% Egypt. Egypt has the players. The Gulf has the money. "MENA" is a linguistic grouping and a shipping convenience, not a monetization strategy.

Now the supply side. As of W3Techs' August 2026 survey, Arabic is the content language of 0.6% of websites whose language is known. English is 49.5%. Whatever you think the Arabic-speaking internet population is, the web being served to it is thin, and a games portal that renders correctly in Arabic is competing against a shallow field. That is the opportunity. Everything below is why most operators never collect it.

🔤 Arabic Is Two Algorithms, Not a Word List

Two things have to happen between a string in a database and legible Arabic on a screen, and neither is translation.

Contextual shaping. Arabic is cursive. Most letters take a different form depending on what sits either side of them — isolated, initial, medial, final. The letter is one character in your database and up to four different glyphs on screen. Certain pairs also fuse: lam followed by alef is drawn as a single ligature, not two letters side by side. Skip this step and you get Arabic text that a reader will describe as "broken" — disconnected letters that technically spell the word and are painful to read.

Bidirectional reordering. Arabic runs right to left, but numbers inside it run left to right, and so does any embedded Latin text — a level number, a brand name, a score. Unicode specifies exactly how to resolve that in Annex #9, the Bidirectional Algorithm. It is not optional and it is not a heuristic you can approximate with reverse().

Here is the part that misleads people: when you write Arabic into a web page's DOM, or draw it with the canvas fillText API, the browser runs both algorithms for you. Your portal chrome, your category names, your help pages, your checkout — those localise close to free, because you are handing text to a text engine that already knows what to do with it. Which is precisely why teams conclude the job is easy, sign an Arabic launch date, and only then look inside the games.

🎮 Your Engine Decides Whether Any of That Happens

A licensed HTML5 title is not one rendering strategy. It is whichever one the original studio picked, and you inherit it.

Text drawn through the browser's text engine

Phaser's standard Text object works by rendering to an internal hidden canvas via fillText. That path goes through the browser, so shaping and bidi happen. Games built this way are the good case: replace the string, and the glyphs come out connected and in the right order.

Text drawn glyph by glyph from an atlas

Phaser's BitmapText takes a texture file plus an XML or JSON descriptor and draws each character from the sheet, spaced proportionally. Unity WebGL with TextMeshPro does the same thing with its own atlas. Both are faster, both look identical across devices, and both are completely blind to Arabic. TextMeshPro has no native bidirectional or Arabic-script support; the text arrives as a sequence of codepoints and each one is stamped down in order.

There are fixes, and reading their own documentation is instructive. The most widely used open-source option, RTL Text Mesh Pro, is candid that it is a workaround rather than real RTL support: it shapes Arabic from a fixed presentation-form table rather than doing OpenType shaping, and it does not implement the full UAX #9 algorithm. The documented consequence is exactly the case a game menu hits constantly — Arabic text with a number or an English word in the middle of it. Commercial alternatives exist, and Unity's newer UI Toolkit handles Arabic properly through its Advanced Text Generator. None of that helps you if the title you licensed shipped three years ago on the old UI stack.

And then the atlas itself

Even with a correct shaping pass, a bitmap font page baked for Latin has no Arabic glyphs in it. Arabic needs multiple forms per letter plus ligatures — several times the glyph coverage. You are not editing a string. You are regenerating a font atlas and rebuilding the game.

📦 You Licensed a Build, Not a String Table

Here is what a standard non-exclusive HTML5 licence typically delivers: an index file, a minified JavaScript bundle, sprite sheets, audio, maybe a config. There is no strings.json. There is no font source. The text is inside the bundle, and in a bitmap-text title it is also inside a PNG.

So the localisation question turns into a licensing question, and it is one most buyers never ask: does your licence grant the right to modify the game at all?

Distribution rights and modification rights are separate grants. Plenty of catalogue licences explicitly prohibit creating derivative works, decompiling, or altering the build — the licensor's whole business depends on the same asset being licensable to the next portal in a known state. Where localisation is permitted, it is often a separately priced grant with conditions attached: approved translators, licensor review before release, and ownership of the localised version sitting with the licensor rather than with you.

On branded and character titles the position is tighter still. Any change to on-screen text is a change to how the licensed property is presented, which means it goes through an approval pipeline with its own calendar. Budget for the review round, not just the work.

✍️ What to Put in the Deal Before You Promise Anyone Arabic

• An explicit right to localise. Named as localisation, not left to be inferred from "modify". Say which languages and whether the right survives renewal.
• A delivery obligation. Extracted strings at minimum; font sources and a rebuild if the title uses bitmap text. A right to modify a file you were never given is worth nothing.
• Who does the engine work. The licensor rebuilding once for all its licensees is usually cheaper than each portal solving it alone — and it is a reasonable thing to ask for during renewal.
• Ownership and reuse of the localised build. If you pay for it, define whether you can keep using it after the licence term, and whether the licensor may resell your version to a competitor.
• Approval turnaround for branded titles, in business days, in writing.

🧮 Triage the Catalogue by Text Surface Before You Spend

The good news is that the reason casual games travel is that most of them barely speak. Before commissioning anything, sort the titles you actually intend to promote into three buckets:

1. Zero-text. Physics, arcade, puzzle, sport. Icons, numbers, a timer. These ship into an Arabic-language portal today with no work at all.
2. Chrome-only. A menu, a tutorial line, a game-over screen, a leaderboard label. Small string count, but this is where the engine question decides whether it costs an afternoon or a rebuild.
3. Text-dependent. Word games, trivia, narrative, anything with instructions per level. Frequently not worth localising at all — and word games are usually locale-specific by design, so a translated one is a different product, not a translated one.

Then localise the shell completely: portal navigation, category names, search, game titles and descriptions, help, terms, billing flows and support. A fully Arabic portal wrapped around a mostly wordless arcade catalogue reads as a local product. A half-English portal wrapped around perfectly translated games does not.

🔢 The Small Things That Read as Machine-Made

Numerals. Arabic-Indic digits (٠١٢٣) and Western digits are both correct Arabic depending on where you are — Gulf audiences see Arabic-Indic routinely, North African audiences generally expect Western digits. Pick per locale, not per language, and be consistent between the portal and the game.

Mirroring, but not everything. Set dir="rtl" at the document level and use logical CSS properties so padding and margins flip themselves. Mirror navigation, back arrows, progress bars and carousels. Do not mirror clock faces, media transport controls, or a game board whose rules depend on its orientation — a mirrored chessboard is a bug, not a localisation.

Machine translation at volume. Auto-translating a thousand game descriptions and publishing them as indexable pages is the fastest way to turn a catalogue into thin content. Translate the pages you intend to rank; leave the long tail alone.

⚠️ One Number Going Around That I Would Not Put in a Board Deck

You will meet a claim that 41% of MENA users only play a game if it is available in Arabic. It appears across several localisation vendors' blogs, always without a named study, sample or year. It may well be directionally true. It is not evidence, and a vendor-sourced conversion statistic used to justify a six-figure programme is exactly the kind of thing that gets found out in the post-mortem.

The stronger signal is behavioural. In February 2025, at the World Governments Summit in Dubai, Roblox launched full Arabic support across desktop, mobile, PlayStation and Xbox, extending it to its automatic translation tooling and its Arabic chat moderation. A platform of that size does not fund that for a rounding error.

And note what Roblox's own developer guidance says alongside it: creators may still need to enable settings so text renders correctly right to left. Even on a platform that has done the hard work at the system level, per-title rendering remains the developer's problem. That is the whole argument of this post, stated by someone with far more engineering budget than your portal has.

🚫 Five Ways Operators Get This Wrong

1. Translating the portal and announcing an Arabic catalogue. Players open the third game, see disconnected letters, and conclude the site is fake.
2. Buying an Arabic string pack before testing one build. The translation is the smallest line in the budget and the only one that gets quoted first.
3. Assuming a shaping library fixes it. A shaper produces correctly ordered, correctly formed glyphs — into an atlas that does not contain them.
4. Treating MENA as one market. A 25x ARPU gap between the UAE and Egypt means different pricing, different payment rails and different monetization mixes. Carrier billing and telecom bundles carry weight in parts of the region where card-based purchase does not.
5. Committing to a launch date before reading the modification clause. The build is sealed, the right may not exist, and the licensor's approval calendar is not yours.

🎮 Where a Licensed Catalogue Fits

Forestry Games has licensed games since 2017 and the catalogue runs to 1,049 titles across HTML5 and Android. For Arabic-speaking markets the practical route is usually the one above: pick a low-text arcade core from the catalogue, localise the shell properly, and settle the modification and delivery terms for the handful of text-heavy titles you genuinely want — including source-code arrangements where a rebuild is unavoidable. For operator and telecom portals in the region, the same triage decides what actually ships on day one versus what waits for a build cycle.

🧭 Do This Before Your Next MENA Commitment

Run a two-hour test this week, before any budget conversation. Take five titles you would put on the front page. Open each one and find on-screen text. For each, work out whether it is drawn through the browser's text engine or stamped from an atlas — a quick way to tell is whether the text scales crisply at any zoom level, or gets soft the way a bitmap does.

Then email your licensor two questions: does the licence permit localisation, and will they supply extracted strings and font sources. The answers decide whether Arabic is a fortnight of portal work or a rebuild programme with someone else's approval calendar attached. Ask before you announce a date, not after.