Image Tracking Analysis Report

The most common position is praise with no reason given, held by 28% of items, which is 19 points ahead of the next, pride in the product as an Arab and Syrian achievement, at 9% of items. Based on the positions people hold, 47% of items hold a position supporting Thaura AI and none opposing it, and 16% hold the reverse. The opposing positions are about access (7% of items), performance (4% of items), doubts about ethical claims (4% of items) and pricing (3% of items). The supporting positions rest on pride, quality and politics. 36% of items hold none of the listed positions, and 101 of those hold a clear view that is not on the list.

Among the positions held by 30 items or more, those in which more than 15% of items imply the position rather than state it in the comment itself are: "The product is a proud achievement of Arab/Syrian innovation and resistance" in 21% of its items; "The product is a high-quality, competitive alternative to Big Tech AI" in 37% of its items; "The founders' or the company's political activism is a reason to support it" in 61% of its items; "The product is a trustworthy, ethical, and privacy-respecting tool" in 48% of its items. 133 items (5%) hold a view not on the list (101 of them hold no listed position), among them, in no order of size: "Boycotting is an effective tool"; "AI development is a waste of resources that should be redirected to solve urgent global humanitarian and environmental crises"; "AI in general is unnecessary and should be boycotted".

By month, audience items whose positions were read and that carry a date; periods with fewer than 30 items are left out.

Comments are quoted as written; see the content note in the Summary.

One comment for each of the leading positions, translated in full. Every quoted comment is in the appendix.

• Positive runs 48 points ahead of negative.

% of the 2,595 items counted · tone of the item

Positive items, 56% of items, are largely short blessings and encouragement, in phrases such as "May Allah bless your work" and "Mashallah". Some praise Thaura AI without giving a reason, and others call it a proud achievement of Arab or Syrian innovation, with some mentioning a Palestinian founder. Some call it a competitive alternative to Big Tech AI, and some ask in a polite, blessing-filled way how to download it. 536 of the 2,595 items hold only emoji or a tagged name, and 468 of those are positive. Among the 2,059 items with words, 48% are positive, 43% neutral and 9% negative. Neutral items, 37% of items: 553 of the 952 hold no listed position, and the largest single concern, about a third of the 952, is whether the app can be found and used. Others complain about weak performance, pricing and free-tier limits, or doubt that an Arabic AI can be called ethical when it may rely on other companies' models, and some simply suggest other privacy-focused chatbots. Negative items, 8% of items, include mockery of the model's answers, such as laughing that it named the wrong ruler of Syria, and doubts that its ethical claims hold up. Others are angry calls for boycott that name Palestine and Big Tech, and one comment calls for a global general strike and writes "HUNT NAZIS" and "ROPE" (see the risk card). Several use anti-Zionist or anti-Jewish jibes ("Fuckin' Zios"). A few turn on Syrians or Arabs ("Arab liars", "a bunch of refugees"), about as many argue over Syrian politics (Assad, al-Jolani), and some are quarrels between commenters. Some reject AI outright ("God please let ai die"), and some express contempt in a short phrase without giving a reason.

"Big Tech Boycott", "Palestinian Solidarity", "Ethical AI", "Data Privacy" and "Pricing and Subscriptions" are added to the chart because the findings cite them; they fall outside the themes it would otherwise show.

Not charted, at 0.5% of items or more: Decentralization (2%), Digital Colonialism (1%), Environmental Impact (1%).

Commenters in this theme ask where to get the app. They write that they cannot find it in the App Store, ask whether it is on Google Play or only a website, and report that login fails. 42% of items in this theme hold the position that the product is difficult to access or lacks necessary platform support, and about half of them name the iPhone or the App Store. One commenter asks whether the app is blocked in Syria. Other commenters answer with directions, such as searching for "play thaura" in the Google Play store. A smaller group mixes praise with complaints about access. One says it resembles ChatGPT but still lacks features such as image support, while 6% of items in this theme call it technically flawed or inferior in performance and 5% call it a high-quality, competitive alternative to Big Tech AI. One encouraging commenter asks for a mobile app and agent features, and for a free plan limited by usage percentage rather than by number of requests.

Commenters in this theme split between praise and doubt about how well the model works. 29% of items in this theme call Thaura AI a high-quality, competitive alternative to Big Tech AI. Some say it is wonderful and easy to use, and one says they do not trust ChatGPT and hope Thaura AI succeeds. 26% of items in this theme call it technically flawed or inferior. About 25 items in this theme say or suspect it is built on someone else's model (the Chinese model GLM, ChatGPT or an open-source model), or doubt that two people could have built one. One of them argues that talk of Thaura AI having been built from scratch is marketing hype, because that is technically impossible for individuals or a small start-up, and that building on a ready-made model would itself be a respectable achievement. One commenter calls it the worst app. Some commenters hold both views at once, saying they tried it and found it excellent but that it needs some fixes, and one adds that they will subscribe. 9% of items in this theme object to pricing or free-tier limits, with some saying it is good but too limited, and complaining about a cap of 5 or 10 free messages. Others ask practical questions, such as whether it can be used in Arabic or whether it runs on another company's AI through an API.

63% of items in this theme present Thaura AI as a proud achievement of Arab and Syrian innovation. Some add Palestinian and wider Muslim pride, one saying the founders' family is Palestinian and blessing "all persevering Muslim youth". Smaller groups back the product on its merits or on the founders' politics. 6% of items in this theme call it a high-quality, competitive alternative to Big Tech AI. 5% of items in this theme say the founders' or the company's political activism is a reason to support it. One commenter asks the builders to make Arab-made social platforms to replace the ones "force fed" to users. Some comments push back or ask practical questions. One angry commenter calls the product's ethical claims lies, writes "Arab liars" and mocks "Syrian AI" as "artificial imbecility". One complains that the app is not available on the iPhone.

Some commenters in this theme say the company's political stance is a reason to back it: 29% of items in this theme hold that view, and they frame the product as a way to leave big tech (the large American technology firms behind mainstream AI tools). Others present it as a working alternative to the big-name chatbots, which 14% of items in this theme argue. One commenter writes that people hate being big tech's "serfs" but have no alternative, and another says simply that an alternative is needed. Smaller groups call it a proud Arab and Syrian achievement (10% of items in this theme) or a trustworthy, privacy-respecting tool (8% of items in this theme). Some commenters use boycott language directly, such as calls to quit ChatGPT over its reported military deal and hashtags urging a boycott of Israel. About a dozen comments under one news clip insist that boycotts do work, some angrily, taking the clip to say they do not. Two comments warn against a rival app, base44, saying its founder worked with the Israeli occupation; one calls it an Israeli company, the other says an Israeli company bought it. Some doubt the whole idea, arguing that the only ethical alternative is not to use AI.

In this theme, 40% of items say the "ethical" label cannot be true. Some commenters argue that no AI can be ethical. One says that ethics and business never go together, and one that AI should be banned until guidelines exist. Others say the branding is never defined, and they ask whether Thaura AI publishes its training data or the models it uses. A smaller group, 17% of items in this theme, describes the product as trustworthy, ethical and privacy-respecting. One of these commenters hedges, calling it "the most ethical version of AI" rather than fully ethical. Others present it as the alternative to ChatGPT, one post accusing that tool of powering military targeting and spreading propaganda. Another 9% of items in this theme treat the founders' or the company's political activism as a reason to support it. One sceptical commenter suspects another AI company of wording a claim about "mass" surveillance to keep it apart from its work with Palantir.

Not charted, at 0.5% of items or more: Comparison with Elsewhere (2%), Economic Reasoning (2%), Religious Citation (1%).

Of the items holding each position, the share using each frame. An item can use more than one frame. Each row shows its three most used frames, more where tied; frames held by fewer than 3 of a position's items are left out. 1 of these groups holds fewer than 30 items (25), so each of its shares stands for a few comments, not a steady rate; any table that breaks a figure down by source, period or dialect shows no shares within such groups.

Frame descriptions

The chart counts the 2,224 of the 2,933 items that carry a date; the other 709 (627 comments, 36 posts by others and 16 posts and 30 replies by the accounts kept apart) carry none.

Activity peaked in the week of 5 January 2026 with 781 items, after only 51 items the week before and then 115 the week after, so the spike rose and fell within a few weeks. Most of that week's comments answer one reel, posted on 6 January by a creator with Thaura AI. Volume then moved in smaller waves, climbing again to 282 items in the week of 16 February 2026 and dropping to 54 in the week of 9 March 2026, against a weekly median of 64 items. The latest week, week of 23 March 2026, holds 92 items up to the last date in the export (27 March), far below the peak and above the median. Of the audience's items, 597 carry no date. They are mostly bare emoji and more positive (77%) than the dated ones (49%), so compare weekly tone between weeks, not with the 56% overall.

The bars above cover 1,958 of the 2,933 items (67%). The other 975: 252 published by the pages themselves, which carry no dialect; 86 with no text to read; 637 the report could not place in one of its six varieties: 615 unclear, 13 in other languages or varieties, 5 mixing two languages or varieties and 4 that are only a tagged name or a platform translation, not counted as English.

Share of each dialect group's items; groups under 30 items are left out.

English-writing commenters form the largest single group at 21% of items, followed by MSA at 19% of items and Levantine-speaking commenters at 16% of items. Gulf and Egyptian writers are far smaller groups, at 6% and 4% of items. The Arabic varieties together are 45% of items. The tone differs most in how positive it is. MSA items sound the most positive at 66% of MSA items, while Levantine and English items sound the least positive, at 38% of Levantine items and 38% of English items. Most of those positive MSA items are blessings or greetings, which are about half of MSA items. With blessings and greetings left out of every group, about 27% (Gulf) to about 40% (Egyptian) of items are positive, and MSA is at about 34%. The positions differ by group: 33% of MSA items, nearly all of them blessings or greetings, praise Thaura AI without giving a reason, while Levantine-speaking commenters most often call it a proud achievement of Arab or Syrian innovation (17% of Levantine items). The next most common Levantine position is that the product is difficult to access or lacks platform support (15% of Levantine items). 78% of the Levantine items, and 55 of their 63 negative ones, sit under the 6 January reel, so this is largely the reading of one thread. English items most often praise it with no reason given (15% of English items) and next doubt its ethical claims (10% of English items), a position the table above does not show; they hold 65 of the 94 items with that doubt.

Of all 2,933 items, Instagram holds 80% and Facebook 20%. Tone is positive in 55% of Instagram items with a tone and 64% of Facebook items with a tone. Negative tone is low on both, at 8% of Instagram items with a tone and 4% of Facebook items with a tone, and neutral tone reaches 37% on Instagram and 33% on Facebook. About 44% of the audience's items answer one reel, the 6 January reel posted by a creator with Thaura AI.

Engagement sits on a few posts: the five most-engaged accounts, each on one or two posts, hold 58% of the likes the audience's items drew, though the export records no writer for any Instagram comment, so only 405 of the 2,681 audience items have one. Voice_1 (5,854) and Voice_2 (5,301) are each one post telling the founders' story, one promotional and positive, one a neutral news clip; the classifier gave neither a listed position. Voice_3 (1,549) has a negative tone, aimed at ChatGPT, and its one item holds three positions that back the company, while Voice_4 has a positive tone and backs the company because of its founders' or the company's political activism. The pattern is mixed: the top voices differ in tone, and some hold no listed position. The account that wrote the most, among those with a recorded writer, is Voice_11 with 5 items, not any of these accounts. The pages' own posts and replies drew 84% of all likes, most of them on the 6 January reel (111,000 likes), and none of the pages are among these accounts.

Share of Voice: each account's engagement as a percentage of the engagement recorded across every item the audience wrote in this dataset. Engagement = likes and other reactions. Replies and shares are not counted. It is a share of the reactions these items drew, not of how much of the conversation an account wrote: the items column is that. The pages' own posts and replies drew 84% of all likes and are not in this table. The table lists the ten accounts with the most engagement, then the three that wrote the most among the rest (Voice_11 to Voice_13).

Favourable and unfavourable come from the analyst's tags on each position; the model is never asked. An item holding positions on both sides counts in both. Groups under 30 items are left out.

Of the items on each side, the share using each frame. An item can use more than one frame. An item holding positions on both sides counts in both.

Favourable items outnumber unfavourable ones on both platforms: 48% of Instagram items against 19% unfavourable, and 49% of Facebook items against 9%. Among dialects, Modern Standard Arabic items are the most favourable, with 54% favourable and 9% unfavourable, largely because about half of them are blessings or greetings. Levantine items are the most divided, with 35% favourable and 33% unfavourable, and Gulf items also carry a high unfavourable share at 33%. The favourable and unfavourable sides argue differently: 47% of favourable items are blessings or greetings and 18% are personal testimony, while unfavourable items lean on personal testimony (32%), questions (29%) and sarcasm or ridicule (13%).

None of the indicators it could check fired; most could not be checked, so this is not a clean result.

Not checked: bursts of posting: the timestamps carry dates only, with no time of day.

Not checked: identical text across accounts: 2,276 of the 2,681 items carry no recorded writer, so it read only 405.

Not checked: near-identical text across accounts: 2,276 of the 2,681 items carry no recorded writer, so it read only 405.

Not checked: one account repeating itself: 2,276 of the 2,681 items carry no recorded writer, so it read only 405.

Not checked: account names on a shared stem: 2,276 of the 2,681 items carry no recorded writer, so it read only 405.

The scan could run only 1 of its 6 indicators, the check for uniform hashtags, and that check found no anomaly across the 2,681 audience items, though reading the text shows one hashtag repeated nine times in a row under the 6 January reel. The remaining indicators could not run: the timestamps carry dates without time of day, so bursts of posting could not be tested, and 2,276 of the 2,681 audience items lack a recorded writer, which left the account-based checks (identical text, near-identical text, one account repeating itself, and shared account-name stems) unable to cover the set. The scan therefore gives no rating and cannot say whether coordinated behaviour is present or absent.

Answer the access problem first: 7% of items from the people who commented under 80 posts about Thaura AI hold that the product is difficult to access or lacks necessary platform support, which makes it the largest listed objection and one the Thaura AI team can act on. Across 2,595 audience items, 47% hold a position supporting Thaura AI and none opposing it, while 16% hold the reverse, and 86% of these items sit under posts on Thaura AI's own Instagram profile. So among these commenters the goodwill is already there, and their own questions, such as how to download it, show where to start. Publish clear instructions for where and on which platforms the product can be used, and state plainly which platforms are not yet supported and when that might change, rather than leaving the question to the comment threads. Among the next objections, each at 4% of items, are weak performance and the view that the claims of being ethical or sustainable are unsubstantiated or impossible. Some commenters asked what data Thaura AI was trained on, so publish that information if it exists, or say it is not available. Stand behind only the claims you can document yourselves, and do not rely on the praise in the comments as proof of performance, privacy or ethics.

Comments are quoted as written; see the content note in the Summary.

Each comment is shown whole, as collected, with a translation of the whole comment. Grouped by the first position each holds.

Thaura AI draws the most attention, named in 145 items, and the leading sentiment toward it is positive, with 92 of those items carrying a positive tone. ChatGPT follows with 104 mentions and Syria with 69. The leading sentiment toward Syria is positive, while ChatGPT's is mixed, split between 47 negative and 41 neutral items. Gaza also carries a positive leading sentiment, with 24 mentions, though its count holds one hashtag posted nine times, which makes it read positive; counted once it has 16 mentions. Israel, named in 19 items, and Google, named in 23, have a negative leading sentiment, at 12 and 11 negative items. The leading sentiment for App Store (44 mentions) and iPhone (28) is neutral, while for Claude (23) it is mixed.

Mentions: items in which the classifier listed the entity; a word search can find more. Leading sentiment: the sentiment the classifier recorded toward the entity where it recorded one, otherwise the tone of the item that names it. The two can differ: a comment angry at those who insulted someone is negative in tone and positive toward him. Mixed means the leading tone led the runner-up by a fifth or less.