Research conducted by Lunasol. In the eight answers we recorded, asking the question in Arabic returned a different and larger set of cited websites than asking it in English. Across four sectors in Google's AI Mode on 21 September 2026, the English question cited 31 websites and the Arabic question cited 50, adding the four sectors' counts together rather than counting distinct websites across the study. Of those 31 English websites, 15, or 48 per cent, were cited again in the Arabic run of the same sector, against the 74 to 77 per cent that the identical English question kept when we asked it again with nothing changed, and because the Arabic runs cited 50 websites against the English 31, that 48 per cent is a changed list rather than a shorter one. The 74 to 77 per cent is not measured here. It comes from our separate study on variability.
Lunasol sells work aimed at how businesses are described in these answers, so the reading here is commercially convenient for us, and the notes set out that interest in full.
AI Visibility is owned by Lunasol.
The short version. The Arabic question cited more websites than the English one in every sector, 50 against 31, and more links out of the answer every time, 93 against 63, but only 15 of the English 31 were among those 50. That 15 of 31 is below the 74 to 77 per cent the identical English question kept by itself, and well clear of nothing. Two official or government websites appeared in the Arabic runs. Across every English run in this series, one has appeared, in the most detailed question of a separate study. With one Arabic run per sector we can set these figures beside the English ones, and we cannot say how much of the difference the language caused. It is a pilot: one surface, Google's AI Mode, one day, four sectors of twelve, one run per language.
What we did
Where the English side came from, stated plainly. The English run of each sector is not new. It is the counted run from our earlier study of which named businesses were also cited, recorded earlier the same day under that study's own protocol, frozen before its first counted run. The same four answers are run 1 of our separate study on variability and level 1 of our separate study on question detail. Four articles in this series therefore rest partly on the same four English answers, and this one says so.
The four sectors were chosen by a counting rule. The rule, fixed in advance, was every third sector in the order used by our count-only twelve-sector study: business setup consultants, IT support companies, gyms and moving companies.
The two questions. The English question is the plain category question, "Which are the best [sector] in Dubai?" The Arabic question puts the same request in Arabic, one per sector. It was written to be the same question, and whether it reads as the same question is not something this study tested. They are recorded in our protocol, which fixed all four before the first Arabic run. Our record does not say how each Arabic question was produced or who checked it, so what we measured is what these four strings returned. This article is written in plain ASCII characters, so the Arabic prompts are not reproduced here and are not transliterated either.
What was counted. For each run: the total outbound links in the answer that were not Google's own, and the distinct websites among them. Then, for each sector, how many of that sector's English websites were cited again in the Arabic run. Two websites count as the same website when they share the same domain name, so the www and non-www forms of one address count once.
What was not counted, and why it matters. For the English runs we also recorded how many businesses each answer named, reading the names the answer sets out as highlighted entries rather than names mentioned only in passing. For the Arabic runs we did not. We had fixed that measure in the protocol before the runs, so this is a measure we set out to take and then did not take. Our method for reading those highlighted entries had not been checked against Arabic script, so we have no figure we would stand behind, and a one-sided figure would invite exactly the comparison the data cannot support. The links and the website addresses were read the same way in both languages, and it is the reading of the business names inside the answer that the script affected. There is no English-to-Arabic comparison of businesses named anywhere in this article, and that silence is a limit of our method, not a result.
The problem with this study, written down before we started rather than found afterwards. In our separate study on variability, the same English question asked twice did not return the same sources in most of the sectors we ran. So a difference between two languages is only evidence of a language effect if it is larger than the difference the same question produces by itself. That is why the 74 to 77 per cent figure appears beside the language figure throughout this article. We wrote that requirement into the protocol before running anything.
What was not done. We did not open a single cited page. We ran the Arabic question once per sector, so we have no repeat measurement in Arabic at all.
Eight runs, eight answers. No failures and no re-runs.
How much of the English answer's source list came back in Arabic?
In the four sectors Lunasol ran in Google's AI Mode on 21 September 2026, the English question cited 31 websites, counted sector by sector. The Arabic runs cited 15 of those 31 again across the four sectors, each sector counted against its own English run, or 48 per cent, against the 74 to 77 per cent that the identical English question kept when Lunasol asked it again with nothing changed. Those baseline figures, 24 of the same 31 on a second asking and 23 on a third, are measured in our separate study on variability and quoted here, not measured again. The Arabic runs cited 50 websites in all against the English 31, so the 48 per cent marks a changed source list and not a shorter one.
Table 1. How many of the English question's 31 cited websites came back. Lunasol, four Dubai sectors in Google's AI Mode, 21 September 2026.
| What was asked | Websites kept from the English question's 31 | Share |
|---|---|---|
| The identical English question, second asking (baseline, from our variability study) | 24 of 31 | 77 per cent |
| The identical English question, third asking (baseline, from our variability study) | 23 of 31 | 74 per cent |
| The question written to be the same, asked in Arabic (this study) | 15 of 31 | 48 per cent |
Forty-eight per cent sits below the 74 to 77 per cent the identical English question kept by itself. Of the 31 English websites, 16 were not cited in Arabic, where seven of the same 31 dropped out when the identical English question was asked a second time and eight on a third, so the Arabic run is associated with roughly twice as much movement as the same question produces by itself. With one Arabic run per sector we cannot say how much of that the language caused.
It is also a long way above nothing. In these eight answers the two source lists overlapped by 15 of 31, so they were not separate lists.
Arabic returned more sources in every sector, and not the same ones
The Arabic question cited more distinct websites and more outbound links than the English question in all four sectors, and fewer than half the English websites were among them. Across the four sectors the Arabic runs cited 50 websites against the English 31, and 93 outbound links against 63, and 15 of the English 31 were cited again in Arabic, against 24 and 23 of the same 31 when the identical English question was simply asked again. Both website totals are sums of the four per-sector counts, not counts of distinct websites across the study. A larger count of cited websites is a count. We did not test whether a larger source set makes an answer better, worse or more useful, and nothing here says that it does.
Table 2. One English run and one Arabic run of the same question in each sector. Lunasol, four Dubai sectors in Google's AI Mode, 21 September 2026.
| Sector | English links | Arabic links | English websites | Arabic websites | English websites cited again in Arabic |
|---|---|---|---|---|---|
| Business setup consultants | 21 | 27 | 10 | 16 | 4 of 10 |
| IT support companies | 10 | 20 | 5 | 9 | 2 of 5 |
| Gyms | 12 | 19 | 6 | 11 | 4 of 6 |
| Moving companies | 20 | 27 | 10 | 14 | 5 of 10 |
| All four sectors | 63 | 93 | 31 | 50 | 15 of 31 |
Lunasol, four sectors in Google's AI Mode, 21 September 2026, one English run and one Arabic run in each.
Most of what the Arabic runs cited was not in our English answers. Of the 50 websites the Arabic runs cited, 21, or 42 per cent, appear somewhere in the 51 websites that our separate study on variability recorded across three English runs of the same four questions, the 51 being a sum of four per-sector totals like the 31 and the 50. The other 29 were not in any of those English answers. That comparison uses three English runs rather than a single English run, so it is the most generous version available, and it still leaves a clear majority of the Arabic sources outside it. These two figures are counted from different starting lists and they do not disagree. Counting from the English list, slightly under half of it came back in Arabic. Counting from the Arabic list, most of it was not in any of our English answers. The Arabic runs also cited more websites than the English runs in every sector, so they had more room to carry the English ones, not less. None of this says the Arabic answers were better or worse, only that they were largely made of different sources.
Two things we did not go looking for: official websites, and social platforms in one sector
One, two of the 50 websites our Arabic runs cited were official or government bodies, where our English runs across this series have produced one. One was cited in the moving sector and one in the gyms sector. The single English one appeared in the most detailed question of our separate study on question detail. Our count-only twelve-sector study, which folds web addresses differently from this one, recorded no government or official source at all. We are reporting the appearance, not explaining it. Two websites in four Arabic runs is two websites in four Arabic runs.
Two, in one sector the Arabic answer cited social and user content platforms and the English one did not. In the moving sector the Arabic run cited three user content or social platforms where the English run of the same question cited none. This is not a uniform difference: in the business setup sector both the English and the Arabic run cited the same user content platform. We did not assess the quality or reliability of any cited website in either language, and the category a website falls into is not a ranking of it.
What this study cannot tell you
Read this before you quote any number above.
Eight answers is eight answers. Four sectors of twelve, one run per language, one day. None of the shares above is a rate.
The Arabic side was run once per sector. We have no repeat measurement in Arabic, so we cannot subtract the same-question movement from the 48 per cent. We can only set them beside each other, which is what Table 1 does. A single Arabic run that happened to land low would look like a language effect, and we would have no way to tell an unlucky run from a real one.
It cannot tell you the cause. We did not measure why the Arabic answers cited different websites, and nothing in our record tells us that the language is what caused the difference. The Arabic question is a different string as well as a different language, and we did not test the change of string separately from the change of language.
We cannot vouch for the Arabic wording. The four Arabic questions were fixed in our protocol before any of them was run, and our record does not say how each one was produced or who checked it. They were written to be the same question. Whether each is the phrasing an Arabic-speaking buyer in Dubai would actually type is a judgement we did not test.
Businesses named were not recorded on the Arabic side. This article therefore says nothing about whether the Arabic answers named more businesses, fewer, or the same ones. That is a gap in our method and not a finding.
We did not read the Arabic answers as a reader would. We counted links and websites. We did not assess what the Arabic answers said, whether they were accurate, or whether they were useful.
We did not record the language settings. The Arabic questions were typed into the same signed-in account and the same browser as the English ones, and we did not record or vary what language those were set to. Somebody whose account, device and browser are all in Arabic may not see what we saw, and we cannot tell you what part those settings played in any figure above.
The baseline is four sectors on one day too. The 74 to 77 per cent figure comes from our separate study on variability, which ran the same four sectors on the same day. It is the right baseline to use and it is not a large one.
Everything here was recorded in one place. Google's AI Mode on 21 September 2026, signed in to one account, in one browser, on one connection, from one location. Our location did not vary, and we did not test whether a different one would change these figures.
Nothing here is a promise. Nothing in this article makes a business named, cited or recommended. Google states that "No one can guarantee a #1 ranking on Google" (Google Search Central). Our article on whether anyone can guarantee a ChatGPT recommendation goes through the position in full.
What a UAE business can do with this
This is the only section where a finding turns into advice. It is general information about what our eight answers recorded, and not legal, financial, medical, regulatory or other professional advice. Nothing here tells you anything about licensing, or about what any regulated sector requires. If you work in a regulated sector, take your own professional advice before you act on any of it. None of it tells you what being cited is worth to your business, because this study did not measure that. None of these checks makes a business named, cited or recommended anywhere. The checks below need a browser and a few minutes. One of them also needs somebody who reads and writes Arabic, and that item says what to do if that is not you and you have nobody to ask. For guidance on working in both languages, rather than for these measurements, our separate article on English and Arabic AI search is the one to read.
If your customers ask in Arabic, check in Arabic. In our four sectors the Arabic answer cited 15 of the English answer's 31 websites, against 24 and 23 of the same 31 when the identical English question was simply asked again. Checking only in English may be telling you less than you think about the answers that appear when the question is asked in Arabic.
Run both, then compare the two lists of cited websites. Write the websites of each answer in two columns and count how many appear in both. Then ask the English question a second time, unchanged, and count the same way, so you have your own version of the same-question figure before you read anything into a language difference.
Use the Arabic wording a real customer would type. Ours were written to be the same question, and we could not test whether they are the phrasing a real buyer would type. If you read and write Arabic, write it yourself. If you do not, the version that costs nothing is to ask an Arabic-speaking customer, colleague or supplier what they would actually type into a search box, in their own words, and use that. Start from that, not from the English. If you can get neither, run the check anyway and treat the result as weaker, for the same reason ours is weaker.
Do not build a plan on eight answers, including ours. Run the same shape of check on your own sector first. Our article on how to check what AI says about your company sets out how to do that.
The first three are about which language you ask in, which is what this study varied, and the fourth is about running your own check on your own sector.
How this study relates to Lunasol's published process
Lunasol's published process lists an audit first. Our service page (Lunasol) calls it "The AI census", and lists "Presence in sources AI models trust and quote" among the GEO deliverables. GEO is this article's shorthand for work aimed at how a business is described in AI answers.
This study asked questions of Google's AI Mode in two languages and recorded which websites each answer cited. That page describes the published audit as "what ChatGPT says about you today". The audit therefore looks at one business on ChatGPT. This study ran on Google's AI Mode instead.
Where the published scope is. Lunasol's published scope for this work is at lunasol.ae/geo-aeo, and our separate article walking through that published service page goes through the scope in detail. If that is the reading you want for your own business, ask for the AI visibility audit that the same page lists.
Common questions about asking Google AI Mode in Arabic
Why is the same-question figure in the headline table? Because without it the 48 per cent cannot be read. When the identical English question kept 74 to 77 per cent of its own websites, the 48 per cent the Arabic runs kept sits below it. Publishing 48 per cent on its own would invite a reader to think that just over half the answer moved because of the language, when the same English question repeated with nothing changed moved about a quarter of it by itself.
How many answers is this built on? Eight. One English run and one Arabic run in each of four sectors, on one surface, on one day. The four English answers are the same four answers used in three other articles in this series, so they are not new here.
Does 48 per cent, against the 74 to 77 per cent the identical English question kept, mean Arabic gets a different answer? We cannot say that from four sectors and one Arabic run per sector. What we can say is that the Arabic answers cited more websites than the English ones in every sector, 50 against 31, and that 15 of the English question's 31 websites came back, against 24 and 23 of the same 31 when the identical English question was simply asked again.
Why are the Arabic questions not printed here? Because this article is written in plain ASCII characters and an Arabic string cannot be reproduced in them. The questions were fixed in our protocol before the first Arabic run.
Why is there no comparison of how many businesses were named? Because we recorded that for the English runs and not for the Arabic ones. We could not read the highlighted business names in the Arabic answers to the same standard, so publishing one side of that comparison would be worse than publishing neither. We had fixed that measure before the runs, so it is one we planned to take and did not get.
Can I see the raw data? The per-run record exists, with every website in every run. It carries company and domain names, which we do not publish. What this article gives you is the counts taken from that record, which is the per-sector table above. There is no published data file and no separate published report anywhere. That means you cannot check these counts from outside this article, and the way to get a figure you own is to run the check yourself.
Method, sources and notes
Research conducted by Lunasol. Data collected 21 September 2026 in Google's AI Mode, signed in, one browser, one connection, one day. Four sectors, chosen by a rule fixed in advance, one English run and one Arabic run each. The four Arabic questions were fixed in the protocol before the first of them was run. No failures and no re-runs. Counts were read from the rendered page; no cited page was opened. Two websites count as the same website when they share the same domain name.
The shared English runs. The English runs are not this study's. They were recorded under the protocol of our earlier study of which named businesses were also cited, frozen before that study's own first counted run, and they also serve as run 1 of our study on variability and level 1 of our study on question detail. Four articles, four answers. This one says so, and they are not four independent confirmations of anything.
The baseline was not measured in this study, and it is still ours. The 74 to 77 per cent figure is measured in our own study on variability, on the same four sectors on the same day, and is quoted here rather than measured again. It is not an independent figure and it is not a published standard. If that study is wrong, every comparison in this article is wrong with it.
One measure recorded on one side only. The count of businesses named was fixed as a measure in the protocol, then recorded for the English runs and not for the Arabic ones, because our method for reading the highlighted business names in an answer was not checked against Arabic script. No named comparison appears in this article.
The website totals are sums, not distinct totals. The 31 English websites, the 50 Arabic websites and the 51 in the three-run English comparison are each a sum of four per-sector counts. We did not compute or report whether a website appeared in more than one sector.
The Arabic prompts. Held in our protocol record, described here rather than reproduced, and not transliterated.
Names. No host, domain, company, agency, publication or individual from the study is named anywhere in this article, including the two official bodies cited in the Arabic runs, which are described by category only. Sector labels are categories, not businesses.
Who publishes this. AI Visibility is owned by Lunasol, whose published work is aimed at how businesses are described in these answers. This protocol, the sector selection rule, the four Arabic questions and every measure were frozen before the runs they govern, and one of those frozen measures, the count of businesses named, was not recorded on the Arabic side and is reported nowhere in this article. Freezing fixed what would be asked. It did not establish that the Arabic wording was checked by anyone, and our record does not say that it was. The English runs were governed by the earlier study's protocol rather than this one, as the method note above sets out. The advice section is separated and labelled. Our commercial interest runs with a reading in which what an AI answer says about a business matters, because Lunasol's published service page (lunasol.ae/geo-aeo) lists an "AI visibility audit, what ChatGPT says about you today" and "Presence in sources AI models trust and quote" among its GEO deliverables. Readers should weigh the numbers above knowing that. The author writes for this publication, which Lunasol owns, and is not independent of either. This article makes no claim about any other provider of these services, named or unnamed, in any direction.
This is a pilot. It is general information and not legal, financial, medical or professional advice of any kind. Nothing here makes a business shown, named, cited or recommended in any AI product, and nobody can promise that it will. OpenAI says of ChatGPT Search that "Search results and citations can be incomplete, outdated, or incorrect" (OpenAI Help Center), and quoted documentation changes without notice, so check the current version of any page quoted here before relying on it.
Product and company names are used here for identification only. Neither this publication nor its owner is affiliated with, endorsed by or sponsored by any of them, nothing here is an endorsement by us of any of them, and none of them has reviewed, approved or contributed to this article. ChatGPT is a product of OpenAI, and the OpenAI Help Center is a publication of OpenAI. Google Search, Google Search Central and AI Mode are products or publications of Google. All are trademarks of their respective owners.
AI Visibility is owned by Lunasol.
