Why GCC Brands Must Rethink Media Planning for AI Search

Media planning in the GCC still mostly answers one question: how do we get in front of the right person, in the right place, enough times. That question is becoming incomplete, because a growing share of commercial decisions now passes through a system that does not see media at all. It reads pages, picks a handful, and writes an answer.

The strategic shift is not that search is changing. It is that the thing being competed for has changed. Ranking first is a position. Being cited is a judgement about whether a source is worth repeating. Those are not the same contest, and the second one is decided largely on property you do not own.

The click is going away, and that part is measured

Pew Research Center collected the browsing data of 900 US adults across 68,879 Google searches in March 2025. Where an AI summary appeared, users clicked a traditional result in 8% of visits. Where no summary appeared, they clicked in 15%, which Pew describes as nearly twice as often.

Clicks on a source inside the summary happened in 1% of visits. Users exited the browser for five seconds or more on 26% of pages carrying an AI summary, against 16% without one.

Two details matter more for planning than the headline. First, longer and more conversational queries trigger summaries far more often: 8% of one or two word searches produced one, rising to 53% for searches of ten words or more, and 60% for queries beginning with question words. The queries most likely to be absorbed are the considered, research-stage ones, which is exactly the part of the funnel media planning is usually trying to influence.

Second, the method has limits worth stating. Pew did not observe what users saw on screen; it re-ran their March queries through a scraping service between 7 and 17 April 2025 and reconstructed which ones produced a summary, and it notes that summaries change over time. Only the first three sources in each summary were tracked, so the 1% in-summary click figure is a floor rather than a ceiling. Google rejected the study, telling press that it used “a flawed methodology and skewed queryset that is not representative of Search traffic” and that it had not observed significant drops in aggregate web traffic. Google released no data to support that.

What AI actually reads about a brand

If answers are replacing clicks, the operative question becomes which sources the answer is built from. There is now research on this, with real caveats attached.

A June 2026 preprint analysed 167,551 URL-grounded citations covering 128 brands across 13 languages and 12 markets. Its clearest finding comes from the Nordic-Baltic subset of 131,514 citations, the only portion with reliable owned-domain detection:

AI grounded brand answers in third-party sources 85.7% of the time, and in the brand’s own site 14.3% of the time.

The direction held across every cut. B2B brands were slightly more exposed than B2C, at 13.1% owned versus 15.7%, which is the opposite of what most people assume. Several brands in the sample drew zero owned citations: the model never grounded in their website at all.

The second finding has budget consequences. Within that same subset, around 80% of citations came from roughly 18% of domains, and half came from just 547 hosts. A log-log regression on the top 1,000 domain ranks fits a Zipf distribution closely. There is a small head of sources that decides most answers, and a long tail the paper itself calls real but thin.

Mostly third-party, and heavily concentrated. That combination is why media planning rather than content marketing is the right frame. Presence on a short list of sources is a buying and earning problem, not a publishing one.

The finding that should stop GCC planners

Here is where most commentary on this topic goes wrong, and where the honest answer is more useful than the confident one.

That study covers eleven Nordic and Baltic markets plus Hungary. It contains no Gulf market and no Arabic-language data. The Pew data is US-only. The best available evidence on how AI sources brand information has not been collected anywhere in this region.

And the same study shows exactly why that gap matters. Wikipedia was the most-cited domain in 11 of 12 languages, which looks like a universal rule until you examine the exceptions. In Lithuanian, the national business daily Verslo žinios outranked it. In Poland, the most-cited domain was YouTube, ahead of the price-comparison sites, and four recruitment and careers portals together out-cited Polish Wikipedia by roughly two to one.

Think about what that means. In one market, what AI says about a company is shaped substantially by employer-review and jobs sites. Not by its advertising, not by its website, and not primarily by the encyclopedia. A planner who imported a source list from a neighbouring market would have missed it entirely.

So the correct conclusion for a GCC brand is not “buy Wikipedia and trade press”. It is that the source head is market-specific, that nobody has mapped the one that applies here, and that the brands which map it first will be planning against evidence while their competitors plan against imported assumptions.

Why this lands harder in the Gulf

The regional context makes this urgent rather than theoretical.

Deloitte’s November 2025 Middle East predictions report, citing its own State of AI in the Middle East research, puts generative AI use among UAE and Saudi consumers at 58%, which it describes as significantly outpacing UK and European markets. On the organisational side, it reports that over 80% of organisations in the region feel intense pressure to adopt AI and 69% plan increased investment.

Read those two numbers together. Consumer behaviour here is ahead of the markets where the measurement studies were actually run. The exposure is larger and the evidence base is thinner.

There is also a useful piece of regional continuity. KPMG’s 2025 Customer Experience Excellence research, covering 104 UAE brands, found that integrity has overtaken personalisation as the strongest driver of customer experience in this market. Personalisation still scores highly; what changed is the rank order. Integrity in that context means accurate information, fair pricing and dependable follow-through.

That is close to the same quality AI retrieval rewards. Models ground answers in what the open third-party web says about you, so a brand whose claims hold up when checked elsewhere is well positioned in both systems. One whose story only holds together on its own website is exposed in both.

What actually changes in the plan

This is not an argument for moving budget from paid to PR. It is an argument for adding a layer the current plan does not have.

Audit where AI already reads you, per market and per engine. In that study, one engine cited across nearly 16,000 domains, another across roughly 6,600, and a third across about 3,300. A single English-language check against one assistant describes one slice of reality. Run your category questions in Arabic and English, across more than one assistant, and record which domains get cited. That list is your real competitive set for this channel.

Treat the source head as a media buy. Once you know the twenty or thirty domains that decide answers in your category and market, coverage on them is worth disproportionately more than scattered placement. This is closer to how trade and tier-one media were planned before programmatic, and the logic is the same: a small number of credible outlets carry the weight.

Fix the reference layer. Wikipedia led in 11 of 12 languages in that research. A thin, outdated or missing entry is a weak spot across many languages at once. The same applies to the structured, checkable facts about your organisation sitting on third-party registries and directories.

Expand what counts as brand media. The Poland finding is the warning. Recruitment portals, review platforms, marketplaces, community forums and video are brand media now, because they are what the model reads. In a region where employer brand, marketplace presence and video consumption are all unusually strong, assuming the same is not happening here is a position without evidence behind it.

Change what you measure. Rankings and click-through will describe a shrinking share of the journey. Citation share, meaning the proportion of category answers in which you appear and how you are characterised, is the metric that matches the behaviour. The measurement principle is the same one that applies to customer experience: instrument the behaviour you actually care about rather than the proxy that is easiest to collect.

What I would not claim

Four caveats, because this topic attracts more certainty than the evidence supports.

The citation study is a single-author preprint, not peer-reviewed. Its author is affiliated with Rankfor.AI, which develops AI brand-intelligence tools, and the datasets analysed are that company’s own research data. He declares this in the paper. The finding is documented and internally consistent on its main result, and the conflict is disclosed rather than hidden, but it is not independent work.

It measures what models cite, not what people believe or buy. The author states plainly that the citation mix is not a verified reputation signal.

Retrieval behaviour changes frequently. Specific domain shares reflect particular model versions in a particular window. The structural pattern, third-party dominant and concentrated, is more durable than any individual number inside it.

And nobody can currently tell you the GCC source head with evidence, because the measurement has not been published. Anyone who quotes you one is extrapolating from somewhere else.

The actual argument

Search optimisation was a competition for position against a ranking algorithm you could study. What is replacing it is closer to a reputation audit conducted continuously by a system that reads the open web and reports what it finds.

You cannot write your way into that from your own domain, because in the best measurement available roughly six in seven of the sources it reads are not yours. What you can do is become the kind of organisation that third parties describe accurately and consistently, and make sure the specific third parties that matter in this region have something accurate to say.

That is a media planning problem, an earned media problem and a corporate communications problem at the same time, which is precisely why it keeps falling between the teams that own each piece.

Frequently asked questions

What is the difference between SEO and optimising for AI answers?
SEO competes for a ranking position on a results page using signals you can largely influence on your own site. AI answers are assembled from retrieved sources, and research across a 131,514-citation dataset found 85.7% of those sources were third-party rather than brand-owned. The practical difference is that ranking rewards what you publish, while citation rewards what credible third parties publish about you.

Are AI summaries really reducing website traffic?
Pew Research Center found Google users clicked a traditional search result in 8% of visits where an AI summary appeared, compared with 15% where none did, and clicked a link inside the summary in just 1% of visits. This is observational US data from March 2025 showing association rather than proven cause. Google rejected the methodology and query set as unrepresentative of Search traffic, without releasing data of its own.

Which sources do AI assistants cite most about brands?
In a study of 128 brands across 12 markets, Wikipedia was the most-cited domain in 11 of 12 languages, followed by YouTube, Statista and Reddit. Critically, the mix is market-specific: in Poland the leading domain was YouTube and four recruitment portals out-cited Polish Wikipedia roughly two to one. No equivalent study has been published for GCC markets.

How should a GCC brand start?
Run your category’s buying questions through more than one AI assistant, in both Arabic and English, and record which domains are cited. That produces a market-specific source list rather than an imported one. Then treat the small head of recurring domains as a media target, since around 80% of citations in the research came from roughly 18% of domains.

Does paid media still matter?
Yes, and nothing in this research suggests otherwise. Paid media builds the awareness and demand that cause people to ask about you in the first place. What the research shows is narrower: when an AI assistant answers a question about a brand, it grounds that answer in third-party pages far more often than in the brand’s own, so owned content alone cannot determine what it says.

Sources

Chapekis, Athena and Anna Lieb. Google users are less likely to click on links when an AI summary appears in the results. Pew Research Center, 22 July 2025. Google’s response was given to trade press including The Register, 22 July 2025.

Żatuchin, Dmitrij. How Large Language Models Source Brand Reputation Across Languages and Markets. arXiv:2606.25787, 24 June 2026. Preprint, not peer-reviewed. Author affiliated with Estonian Entrepreneurship University of Applied Sciences and Rankfor.AI, whose data the study analyses; competing interest declared in the paper.

Deloitte Middle East. Agentic, Physical, and Sovereign AI: 2026 Predictions Shaping the Middle East, 27 November 2025, citing Deloitte’s State of AI in the Middle East report.

KPMG Lower Gulf. Customer Experience Excellence Report, 5 May 2025.

Further reading

Related: Answer Engine Optimization covers the technical layer beneath this argument. AEO vs GEO vs SEO separates the three disciplines. Marketing budget benchmarks covers where this sits against the rest of the plan. B2B marketing in the UAE covers the longer consideration cycles most exposed to this shift.

Martin Alva

Marketing, Brand & Digital Transformation Leader, United Arab Emirates

Senior Manager, Employee Experience (Brand Positioning, Digital Platforms & AI) at Space42 (A G42 & Mubadala Company). 15+ years in marketing, brand and digital transformation across MENA and Europe. 5 MENA Effie Awards.

More about Martin Alva → Connect on LinkedIn →