Research conducted by Lunasol. Across twelve Google AI Mode answers collected on 21 September 2026, one run in each of twelve sectors, 64 of the 78 businesses those answers named, or 82 per cent, also had their own website cited in the same answer. Read the match the other way round and it is looser: of the 100 cited website slots those same twelve answers produced, counting a website once in each answer that used it, 34 belonged to no business the answer named. The two figures count different things against different denominators, and both describe one day. To get them we put one question to Google's AI Mode, "Which are the best [sector] in Dubai?", ran it across twelve kinds of supplier in Dubai, and recorded two things separately: every business the answer named, and every website the answer cited. Then we matched them. The matching rule and its tie-break were both written and frozen before the first counted run, so the buckets could not be chosen to fit the numbers. Lunasol sells work aimed at how businesses are described in these answers, and this article says so at the top rather than at the foot.
AI Visibility is owned by Lunasol.
The short version. In twelve Google AI Mode answers collected on 21 September 2026, a business named in the answer was usually linked to as well: that held for 64 of the 78 businesses named, or 82 per cent. Thirteen more were named with no website of theirs cited, and one was too close to call and is held out of both counts. Run the match the other way and it is looser: the twelve answers cited 100 website slots between them, counting each website once in each answer that used it, and 34 of those 100 belonged to no business the answer named at all. The thirteen are not scattered. Three sectors hold all of them, and in two of those three every business named without a citation was recorded as international. It is a pilot: one surface, Google's AI Mode, on one day, twelve counted runs, one per sector. It says nothing about any other assistant.
What we did
Twelve sectors, one question shape, one counted run each, on one surface, on one day. Lunasol froze this protocol, and the matching rule with it, before the first counted run. A matching rule written after the data arrives is a rule written to suit the data. This one was written first. Freezing a rule does not make a small sample any bigger, and it is no substitute for one.
One question shape, twelve sectors. Every run used the same sentence with only the sector term changed: "Which are the best [sector] in Dubai?" The twelve sectors, in the order fixed before any run: interior fit-out companies, real estate agencies, business setup consultants, restaurants, accounting firms, IT support companies, law firms, dental clinics, gyms, nurseries, car service centres, moving companies. Because the shape never changed, no difference between the answers is attributable to a change in the shape of the question.
Two lists, then a match. For each run we recorded the businesses the answer named, and separately the distinct websites it cited. Then every named business went into exactly one bucket. Named and self-cited means the answer both listed the business and cited a website that belongs to it. Named and not self-cited means it listed the business and cited nothing of the business's own. Every cited website went into exactly one bucket too: it either belongs to a business the answer named, or it does not.
The tie-break, fixed before the data was seen. A website counts as a business's own only when its domain plainly corresponds to the name the answer gave. A directory page about the business, a social profile, a video channel and a news article about it are not the business's own website. Where the correspondence was arguable, the pair was recorded as arguable and held out of the counts rather than resolved silently. That happened to two pairs. One of those two pairs involves a named business and a cited website at once, so three placements are held out. A placement here means one named business or one cited website that the rule had to place, and there are 178 of them, the 78 named businesses plus the 100 cited websites.
What was not done. We did not open a single cited page. So this is a record of what was cited, not of what those pages say, and not of whether any of it was any good.
One method limitation, stated here rather than buried. Businesses were counted as named when the answer gave them an entry label, meaning a heading or a line in bold at the head of an entry. Where one entry label covered several businesses by name, each of them was counted, which happened once and is set out under the per-run table. A business mentioned only in running prose, with no entry label covering it, was not counted as named. That rule was applied identically to all twelve runs, and one of the two arguable cases below exists because of it.
Twelve runs, twelve answers. No failures were recorded, so no re-runs were needed.
Start from the names: does Google AI Mode link to them?
On 21 September 2026 Lunasol asked Google's AI Mode the same question in twelve sectors, one run each: "Which are the best [sector] in Dubai?" Of the 78 businesses those twelve answers named, 64, or 82 per cent, also had their own website cited in the same answer.
That leaves thirteen, or 17 per cent of the 78, that the answer listed but did not link to. One more was recorded as arguable and is not assigned.
So in this sample, most of the time, naming and citing travelled together. Thirteen of the 78 did not, and one more was too close to call.
Start from the links: do they lead to businesses Google AI Mode named?
Of the 100 websites cited across those twelve Google AI Mode answers on 21 September 2026, 64 belonged to a business the answer named and 34 belonged to no named business at all. This is the citation side of the match, not the naming side: the 82 per cent above counts named businesses and this figure counts cited websites, so the denominators are different. Two more websites were recorded as arguable and are not assigned.
Those 64 are the same 64 matches read from the other side. In every one of the twelve runs the two counts came out equal, which means each self-cited business accounted for one cited website and no more.
In those twelve Google AI Mode answers on 21 September 2026, a third of the cited websites, 34 of 100, did not belong to any business on the list. In one sector that included a plant retailer answering the other sense of the word nurseries. Which categories the rest fell into is the subject of our separate study on which categories of source get cited, and this article does not restate it.
The runs behind that study-wide third looked very different from each other. In these twelve Google AI Mode answers of 21 September 2026, five of the moving run's ten cited websites belonged to nobody it named. In the law run it was five of eleven, and in the real estate run four of thirteen. Two runs, IT support and dental clinics, had none at all. One run per sector is not enough to call that a pattern, but it is enough to say your own sector may look nothing like the study-wide third.
Both directions of the match. The first three rows count the 78 named businesses and the last three count the 100 cited websites, so the two halves do not add together. The source is Lunasol's twelve Google AI Mode answers, one per sector, 21 September 2026.
| Direction of the match | Count | Share |
|---|---|---|
| Named businesses that were also self-cited | 64 of 78 | 82 per cent |
| Named businesses with no website of theirs cited | 13 of 78 | 17 per cent |
| Named businesses recorded as arguable | 1 of 78 | 1 per cent |
| Cited websites belonging to a named business | 64 of 100 | 64 per cent |
| Cited websites belonging to no named business | 34 of 100 | 34 per cent |
| Cited websites recorded as arguable | 2 of 100 | 2 per cent |
The twelve runs in full
Here is every one of the twelve runs, as Lunasol recorded them in Google's AI Mode on 21 September 2026. Links are the total outbound non-Google links counted on the rendered page, and hosts are the distinct websites those links point to, counted once each per run. Across the twelve runs those per-run host counts add to 100, which is what this article means by the 100 cited websites: a website cited in two runs counts once in each. Named is how many businesses the answer gave an entry label to. Self-cited and not self-cited split those named businesses. The last two columns split the hosts again, by whether the host belongs to a business the answer named.
| Sector | Links | Hosts | Named | Self-cited | Not self-cited | Hosts, named business | Hosts, nobody named |
|---|---|---|---|---|---|---|---|
| Interior fit-out companies | 17 | 8 | 7 | 4 | 3 | 4 | 4 |
| Real estate agencies | 26 | 13 | 9 | 9 | 0 | 9 | 4 |
| Business setup consultants | 21 | 10 | 6 | 6 | 0 | 6 | 4 |
| Restaurants | 15 | 9 | 5 | 5 | 0 | 5 | 4 |
| Accounting firms | 17 | 7 | 11 | 5 | 6 | 5 | 2 |
| IT support companies | 10 | 5 | 5 | 5 | 0 | 5 | 0 |
| Law firms | 27 | 11 | 10 | 6 | 4 | 6 | 5 |
| Dental clinics | 11 | 5 | 5 | 5 | 0 | 5 | 0 |
| Gyms | 12 | 6 | 4 | 4 | 0 | 4 | 1 |
| Nurseries | 12 | 9 | 6 | 5 | 0 | 5 | 3 |
| Car service centres | 11 | 7 | 5 | 5 | 0 | 5 | 2 |
| Moving companies | 20 | 10 | 5 | 5 | 0 | 5 | 5 |
| All twelve runs | 199 | 100 | 78 | 64 | 13 | 64 | 34 |
Two sector rows and the totals row do not add up, and the arguable cases are why. In the nurseries run one named business is left out of the self-cited split, and one cited website is left out of the host split. In the gyms run one cited website is left out of the host split too. All three are held out and described below. So the named split is one short of 78, at 64 plus 13, and the host split is two short of 100, at 64 plus 34.
The accounting row is the only one where the number of businesses named is larger than the number of entries on the answer's list. Five firms each had an entry to themselves. A sixth entry covered six international networks together, each of the six named inside that entry, so under the rule fixed in advance all eleven count as named. A reader who would rather count that entry as one name than as six should know where it lands: they account for six of the thirteen named businesses with no website of theirs cited.
Which sectors the named but not cited businesses came from
Thirteen named businesses had no website of theirs cited.
All thirteen sit in three of the twelve sectors Lunasol ran in Google's AI Mode on 21 September 2026. Six are in accounting, four are in law, and three are in interior fit-out. The other nine sectors produced none at all.
In two professional sectors, on this one day, every business named without a citation was recorded as international. In the accounting run, the six businesses named without a citation are all classed as international networks in our record of the run, a category we assigned as we read the answer. The five firms whose own websites were cited are classed as smaller local firms. In the law run, four of the ten named businesses went uncited, and all four are classed as large international firms. Our record does not describe the other six law firms either way, so nothing here says that being international is what separated them.
We did not measure why. No assistant we know of publishes the reason a source is chosen, and a reason we made up would be worth nothing to you. What we can say is what was there: in these two runs, on this day, every name the answer did not link to was one the record placed in the international category.
Interior fit-out is the third. Three of its seven named businesses were named without a link of their own. Our record puts the accounting and law cases in the international category and puts no such description on these three. On one run per sector we cannot say whether that is the same phenomenon or a different one.
The two sectors where the match was perfect
Two of the twelve sectors, IT support and dental clinics, matched in both directions in Lunasol's runs. In each, every named business was self-cited and every cited website belonged to a named business. Each cited five websites against five named businesses.
Those two runs were also among the shortest of the twelve, at ten and eleven outbound links against a median of sixteen, and one other run had eleven as well. Whether a shorter answer is a tighter answer, we cannot say from two runs.
The two arguable pairs
Reporting these was in the protocol, not an afterthought.
In the gyms run, a cited website plainly belongs to a fitness business the answer discussed in prose but did not give an entry label. Under the frozen rule the business is not a named business, which would make its website one that belongs to nobody the answer named. Because the business is plainly there in the answer, we recorded the pair as arguable and held the website out of both counts. A reader who thinks the rule should simply have been followed can move it, and the count of websites belonging to nobody named goes from 34 of 100 to 35 of 100.
In the nurseries run, one cited website's domain is close to, but not the same as, the entry label the answer gave. We could not say the domain plainly corresponds to the name, and we could not say it does not. So the pair was recorded as arguable: the named business is held out of the self-cited count, the website out of the host counts. A reader who takes the domain as a match can move both. Self-cited named businesses then go from 64 of 78 to 65 of 78, and websites belonging to a named business from 64 of 100 to 65 of 100.
Two cases out of the 178 placements are arguable, and between them they leave three placements unassigned, one named business and two cited websites. We are not going to pretend the other 175 were all obvious, but these are the two where we would not defend the call against a reasonable person who disagreed.
What this study cannot tell you
Read this before you quote any number above.
Twelve runs is twelve runs. One question per sector, one sector per run, one day. None of the shares above should be treated as stable.
One surface, one day. Everything here describes Google's AI Mode on 21 September 2026. It says nothing about any other assistant and nothing about any other day.
It cannot tell you what a naming is worth. We counted whether a business was named and whether it was cited. We did not measure whether anyone clicked, called or bought. A naming is not a sale and this study does not pretend to connect them.
We did not settle what a recommendation is. These answers list businesses under a question that asks for the best ones. Whether that constitutes a recommendation, in the sense a customer would use, is not a question a link count can settle, and we will not settle it by assertion.
The matching is our judgement, under a rule we fixed in advance. Another person applying the same rule would place some pairs differently. We have named the two we found hardest, and we have not hidden how the rule treats a business named only in prose.
There is no why in this data. We know of no assistant that publishes the reason a source was chosen. Correlation between being named and being cited is not evidence of a mechanism in either direction.
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.
One run each, and runs vary. Repeated runs of the same question are the subject of our separate study on variability, and anyone reading a single run of anything, including this one, should read that first.
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 twelve runs 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 a naming or a citation is worth to your business, because this study did not measure that.
Check both things separately, because they are two different facts about you. Run the question your customers would type, then read the answer twice. The first time, write down every business it names in its list. The second time, scroll to the foot of the answer, where the websites it used are listed, and write those down. Then answer two separate questions: is your business on the first list, and is your own website on the second. Across the 78 businesses named in our twelve runs, those two were the same fact 82 per cent of the time, so one does not stand in for the other.
If your business is part of an international network, do not assume the citation follows the name. In the two professional runs where this came apart, the businesses named without a link were the ones recorded as international networks or large international firms. That is two runs on one day, so treat it as a thing to check rather than a thing to believe. Checking it costs nothing: run the question and look.
Do not treat your own citation as the whole picture. In our twelve runs, 34 of the 100 cited websites belonged to no business the answer named. Checking only whether your own site is cited leaves most of the answer's source list unread. Open the other sources it cited and see whether your business appears on any of them. That costs nothing but the time.
If an answer names your business, know what it says before a customer does. In one of our twelve runs the answer carried a labelled line of praise for each business it listed. Reading your own entry takes a minute.
Do not build a plan on twelve runs, 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.
All five are checks you can run yourself, and none of them is a finished picture.
How this study relates to Lunasol's published process
Lunasol's published process begins with an audit, described on our service page (Lunasol) as "The AI census". The same page lists "Presence in sources AI models trust and quote" and "AI visibility audit, what ChatGPT says about you today" among the GEO deliverables, GEO here meaning generative engine optimisation, which is work aimed at how a business is described in AI answers.
The published audit is described as "what ChatGPT says about you today", which is a reading of one business on the assistant that line names. This article asks a different question on a different surface: which named businesses were also cited across twelve sectors in Google's AI Mode. This study speaks only for the one AI product it ran on. The published process ends with a step called Track, described as "Numbers, every quarter". Our article on what an AI search audit should cover goes through that in general.
What to do next. Lunasol's published scope for this work is at lunasol.ae/geo-aeo, and our article on what Lunasol's service covers goes through it in detail. If that is the reading you want for your own business, ask for the AI visibility audit.
Common questions about being named and cited in Google AI Mode
How many runs was this? Twelve. One per sector, on one surface, on one day. Every study-wide share here is a share of either the 78 businesses those twelve answers named or the 100 websites they cited, and each figure says which.
Is being named the same as being recommended? We do not know, and this study was not built to find out. We counted namings and citations. What a naming is worth to a business is a different question and it needs different evidence.
Does 82 per cent mean my business will be cited if it gets named? No. It means that across our twelve Google AI Mode answers of 21 September 2026, 82 per cent of the businesses those answers named also had their own site cited. It is a description of twelve answers on one day, not a rate, and it says nothing about yours.
Why did you publish the arguable cases? Because a matching study that only reports the easy matches is a matching study with its thumb on the scale. Two cases out of 178 placements were arguable, and both are described above.
Is this the same data as your separate study on which categories of source Google AI Mode cites? No. That study recorded how many businesses each answer named and not which ones, so the matching in this article could not be done from it. This is a separate set of runs, under its own protocol frozen before its first counted run. Same sectors, same day, same surface, but not a second confirmation of the first study.
Can I see the raw data? The per-run record exists, with every host, every named business and every placement. It carries company and domain names, which we do not publish. What this article gives you is that record with the names removed and the counts kept, which is the per-run table above. There is no separate data file and no separate published report anywhere.
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, under a protocol written and frozen before the first counted run. Twelve sectors, one prompt shape, one counted run each. No failures and no re-runs. Counts were read from the rendered page; no cited page was opened. The matching rule and its tie-break were fixed before any data was seen. The arithmetic was checked in both directions and per run, and both sums are written out under the per-run table above.
One discarded run, disclosed rather than omitted. The extraction script was tested once on the fit-out question before the protocol was frozen, to confirm the browser could return the two lists at all. That test run is discarded and appears nowhere in this article. Fit-out was run again under the frozen protocol and the counted run is the one in the table.
One method limitation. Businesses were counted as named only when an entry label named them, whether that label covered one business or several. A business named only in running prose was not counted. The rule was applied identically to all twelve runs.
The related study. Our separate study on which categories of source Google AI Mode cites ran on the same twelve sectors, on the same day and the same surface, under its own frozen protocol. It is a different set of runs and a different dataset, and it recorded how many businesses each answer named and how many websites it cited rather than which ones they were. Two studies on the same twelve sectors are not two independent confirmations of each other, and both articles say so. The two studies also count hosts slightly differently, which the other article explains.
Names. No host, domain, company, agency, publication or individual from the study is named anywhere in this article. Sector labels are categories, not businesses. Where the article describes a group of named businesses as international networks or large international firms, that is the category the study record assigned, and the article does not say which businesses were in it.
Who publishes this. AI Visibility is owned by Lunasol, a Dubai agency whose published work is aimed at how businesses are described in these answers. This protocol, the matching rule and the tie-break were all frozen before the data existed, and the section where a finding turns into advice is separated and labelled. We have a commercial interest in one of the findings above. 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.
