Back to home

Editorial Policy

Last updated: 9th September 2026

This page explains who writes what you read on this site and inside the ArabyBuddy app, how it is made, where we use AI and where we do not, and what we do when we get something wrong. It is written plainly on purpose. If you are deciding whether to trust or link to us, this is the page that should answer the question.

Who is responsible

Ahmed Gadir, the founder of ArabyBuddy, is accountable for everything published on this site and in the app. ArabyBuddy is a trading name of Gadir Tech Ltd, a company registered in England and Wales, company number 15101441, registered office 3rd Floor, 86-90 Paul Street, London EC2A 4NE.

We are a small team rather than a newsroom. There is no editorial board, no panel of reviewers, and no masthead of contributors. You can read who works on ArabyBuddy and what each of us does on our About page. If something we published is wrong, the responsibility for it is one person deep.

How our content is made

There are two different things here, made in two different ways, and it is worth keeping them apart.

The guides and posts on this website

Every page on arabybuddy.com is written and edited by the team. Each one is a file in our repository, published through a code review, and no page reaches the site without a person reading it end to end. Where a first draft is AI-assisted, as a great deal of writing anywhere now is, a person still edits it, checks every factual claim in it, and answers for it afterwards. We would rather tell you that than claim a purity you have no way to verify.

Every guide and post carries the date it was first published and the date it was last changed. Those dates are not decorative: our build checks the "last updated" date against the page's own edit history, and a page we quietly changed without changing its date will not publish at all.

The courses and audio inside the app

The app's curriculum is produced by an AI-assisted pipeline with human review at defined points. Being specific about that is the only useful way to describe it:

  • The shape of each course is set by per-dialect authoring files that define the concepts, their order, the pronunciation rules for that dialect and its transliteration scheme. These are drafted with a language model, and they have to pass validators that check the sequence is complete and internally consistent before any lesson is built on them. The last file of the set, the worked examples that every lesson prompt is filled from, is read by a person as well.
  • Lesson text, vocabulary and example sentences are generated by a language model working from those files.
  • Illustrations and character art are generated images.
  • All spoken Arabic in the app is synthesised speech. It is text-to-speech in the dialect being taught, not a recording of a person. Before we produce audio for a dialect, someone listens to a sample clip for each character and rejects the voice if it is wrong.
  • Generated content is audited automatically against that dialect's own rules. The audit is run several times over and its flags are scored by how often they recur, so a one-off is not treated as a finding. The flags that do recur are applied as a diff we can read rather than a silent rewrite, and the borderline ones are held back for a person to decide.
  • Nothing reaches learners until a person promotes it. New material sits in a draft mode where we can read and hear it, and it is released deliberately, not automatically.

Here is the limit of that review, stated plainly. We do not employ a native speaker of each dialect to check the content before it ships. That is a deliberate decision, recorded internally on 8 September 2026, and it has an honest consequence: the automated audit is our quality floor, not an expert's signature. Nothing we make is "native-speaker verified", we will not describe it that way on any page, in any store listing or in any directory, and if we ever engage a reviewer we will name the arrangement here rather than upgrade the adjective quietly.

ArabyBuddy teaches seven varieties of Arabic: Modern Standard Arabic, and Egyptian, Levantine, Gulf, Sudanese, Iraqi and Moroccan Darija Arabic. We name a dialect as available only once it is actually in the app, and not a day earlier.

Where we use AI, and where we do not

Most of what follows is disclosed in our Privacy Policy and Terms of Service because it involves your data. It is repeated here because you should not have to read a privacy policy to find out how a lesson was written.

We use AI for:

  • writing lesson text, vocabulary lists and example sentences for the app's courses;
  • adapting and translating lesson material between dialects;
  • generating illustrations and character art;
  • generating every piece of spoken Arabic in the app, which is synthesised rather than recorded;
  • grading the free-text answers you type in some activities, and returning short feedback on them;
  • moderating usernames against our acceptable-use rules;
  • powering Speak, the realtime voice conversation feature. The character you talk to is a language model and not a person, its replies are generated as you speak, and they can be wrong;
  • auditing generated dialect content against that dialect's own rules.

We do not use AI for:

  • Publishing on its own. Nothing goes live because a machine produced it. A page, a lesson, a dialect's audio: each one is released by a person who can hold it back, and pull it down afterwards.
  • Sourcing facts. A date, an attribution or a pronunciation claim on this site is either sourced in the text or it is cut. We do not let a model supply a citation.
  • Reviews, testimonials and quotes. We generate none, and we publish none.
  • Support replies. When you write to us a person reads it and a person answers. There is no support bot.
  • Telling you about your money, your data or your rights. Our pricing, privacy, cookie and terms pages are written by the team and checked against the systems they describe.

Corrections, and how to report one

If something we published is wrong, please tell us. Use the contact form or email support@arabybuddy.com. Point us at the page and quote the line, and if you have a source, send it. You do not need to be an expert and you do not need to be polite about it.

Here is what happens next:

  • A person reads your report. We aim to reply within two business days, and we will tell you what we decided even when the answer is that we are keeping the page as it is.
  • If it is wrong, we fix it and change the page's "last updated" date, so the change is visible rather than silent.
  • If we cannot verify a claim, we remove it. A sentence we cannot stand behind does not get to stay up while we think about it.
  • If the error was substantive, we say what changed on the page itself rather than editing it away quietly.
  • If the error is in the app rather than on the site, it goes to the same place. Dialect reports are the most useful ones we get.

This is not hypothetical. On 9 September 2026 we rebuilt this site's guides and posts, fact-checked them by hand, and corrected four errors in the process. We are listing them because a corrections policy with no corrections in it is not worth much.

  • A wrong claim about Gulf pronunciation. Our guide to the letter ق said the Gulf "preserves the classical /q/". That is wrong for Najdi, the Gulf variety ArabyBuddy is built on, where the everyday reflex is a hard [g], and [q] survives mainly in vocabulary taken back in from Modern Standard Arabic. The guide now says so, and separates the Gulf [g] from the Levantine and Egyptian glottal stop by history instead of implying one is a later stage of the other.
  • A mis-defined grammatical term. Our guide to the root system said that "five-letter roots (الخماسي) exist but are rare". الخماسي names a five-letter verb, which is a three-consonant root plus added letters, not a five-consonant root. The guide now describes genuine five-consonant roots as marginal and explains the terminology separately.
  • A mistranslated gloss. A post glossed عَرَب as "Arab, Arabic", which turned the name عُرَيْب into "little Arabic one". It means Arab or Arabs, so the name is "little Arab one". The sentence that leaned on the wrong gloss went with it.
  • An unsourced date. A post placed Juha in 13th-century Anatolia without a source. The older written record is Arabic: Ibn al-Nadim's Fihrist, compiled in Baghdad around 987, lists a collection of his anecdotes, while the Anatolian date is tradition rather than a surviving manuscript. The post now says both.

What we will not do

  • We will not publish a testimonial we did not receive. This site once had a component holding three invented five-star quotes with invented names attached. It was never shown to a visitor, and on 9 September 2026 it was deleted outright rather than kept around to fill in later.
  • We will not invent a statistic. We have no download count, learner count or star rating worth quoting, so we quote none. When we have a number that means something, we will say where it came from.
  • We will not claim a dialect we do not teach. Not on this site, not in a store listing, not in a directory entry, and not in a keyword field.
  • We will not claim human expertise we have not engaged. No page of ours will say "native-speaker verified" or anything that implies it while no reviewer is engaged.
  • We will not present synthesised audio as a recording of a person, or an AI character as a human tutor.
  • We will not quietly rewrite a mistake. Corrections are dated and, when they matter, described.

Questions

Anything about this policy, a specific page, or a claim you think we cannot support: get in touch or email support@arabybuddy.com. If this page ever stops describing what we actually do, that is itself a correction worth reporting.