AI assistants recommended permanently closed restaurants 93 times in a new local-discovery study. That is the number UAE venue owners should care about more than another promise about “AI SEO.”
The research is not from Dubai or Abu Dhabi. It is an August 7 preprint that audited 4,776 cafes, restaurants and bars in Canggu and Ubud, Bali. The researcher collected 2,208 search-grounded responses from ChatGPT, Claude, Gemini and Perplexity. Most venues never appeared at all, while stale information allowed closed businesses to keep resurfacing.
A new preprint suggests local AI recommendations can omit most venues and can keep surfacing stale businesses that are already closed.
UAE restaurants, cafes, salons, clinics, attractions, local retailers, hotels and agencies responsible for local discovery.
Accurate websites, prices, hours, third-party mentions and business profiles may improve the chance that AI systems can correctly understand a venue.
The study covered two Bali markets, not the UAE, and it is a preprint rather than a final peer-reviewed result. Its associations do not prove causation.
Fix your Google Business Profile, own website, opening hours, closure status, menus and third-party listings before paying for a specialist AI-visibility product.
That does not prove the same percentages apply in the UAE. It does reveal a useful mechanism to test here: local AI discovery appears to have an eligibility problem before it has a ranking problem. A restaurant first has to be legible to the system. Then reputation can help decide where it appears.
What the Study Actually Tested
The preprint, titled Invisible to the Machine, built a census of 4,776 food and drink venues across Canggu and Ubud. It then compared that market against 2,208 search-grounded responses from four production AI assistants across 96 persona-based queries over seven days.
The headline result was invisibility. The paper reports that 85.6% of all venues were never recommended by any system. Even among established venues with at least 50 ratings, 72.6% were never recommended.
Those percentages belong to the Bali markets the researcher studied. Robius is not going to transplant them onto Dubai. The more transferable finding is which signals were associated with getting into the answer at all.
The First Gate Looked Like Documentation
The study reports that entry into AI answers was associated with higher review volume, having an own website, listed price information and third-party web mentions. Star rating was not a meaningful predictor at that first entry stage in the reported model.
Once a venue had entered the recommendation set, the pattern changed. Among recommended venues, rating was associated with a better chance of appearing first.
That suggests a two-stage way to think about local AI discovery. First, can the system confidently identify and describe your business? Second, once you are in the candidate set, do reputation and relevance help you outrank another option?
This is observational research. An own website did not magically cause the AI to recommend a venue. The variables are associated, and better-documented businesses may also differ in many other ways. Treat it as a useful audit result, not a guaranteed ranking formula.
The 93 Closed-Venue Recommendations Are the Bigger Warning
The paper found outright fabricated venue mentions were rare, at 0.08% of mentions. The practical failure was staleness. Permanently closed venues were recommended 93 times.
That distinction matters for a UAE business owner. The common fear is an AI inventing a restaurant that never existed. The more ordinary problem may be an AI reading real but outdated information about a business that changed its hours, moved, closed a branch or stopped selling the thing the assistant recommends.
Google explicitly tells verified Business Profile owners to keep address, hours, contact details and photos accurate and up to date. It also provides controls to mark a location temporarily or permanently closed and to set special hours for short changes.
The UAE Checklist Is Boring on Purpose
- Claim and verify the Google Business Profile for every active location.
- Make the address, phone number, website and normal opening hours consistent across your major listings.
- Use special hours for holidays or short closures and mark longer closures correctly.
- Keep a real website page for the venue with current menu or service information and prices where practical.
- Check major third-party listings for old branches, old phone numbers and old opening hours.
- Test real customer prompts in more than one AI assistant. Do not assume ChatGPT, Gemini, Claude and Perplexity surface the same places.
- Do not buy fake reviews or stuff descriptions with keywords. Clean data and genuine reputation are different assets, and you need both.
This is the local-business version of the same lesson in our AI shopping and product-data guide. A machine cannot reliably recommend information it cannot reliably understand.
Do Not Buy “AI SEO” Before You Fix Ordinary Search
There is already a new software category promising to optimize brands for AI recommendations. Some tools may become useful. But a UAE restaurant with the wrong hours on Google, an outdated menu PDF, no current website and three conflicting addresses does not have an AI optimization problem yet. It has a data hygiene problem.
Fixing that foundation has a useful side effect: it also helps Google Search, Maps, delivery platforms, travel sites, customer support and ordinary people trying to find the door.
Our Shopify vs Salla UAE guide made the same point from ecommerce: distribution depends on the underlying operating setup, not the fashionable layer added on top.
And as Google keeps pushing AI deeper into search, the boundary between traditional local search data and AI answers is likely to get even thinner.
The Robius Read
The paper is useful because it separates visibility from ranking. A business can have excellent reviews and still fail to enter the AI shortlist if the system does not have clean, current information to work with.
For UAE venues, the practical order is simple: become accurately machine-readable first. Then worry about whether you rank first. If your closed branch is still being recommended six months later, no AI visibility dashboard has solved the real problem.
Sources
- arXiv: Invisible to the Machine – August 7, 2026 preprint auditing AI food-and-drink recommendations in Canggu and Ubud, Bali.
- Google Business Profile Help: Edit your Business Profile – Official UAE-facing guidance for maintaining business information.
- Google Business Profile Help: Mark your business as closed – Official closure-status guidance.
- Google Business Profile Help: Edit business hours – Official opening-hours and special-hours guidance.
Robius.news — Dubai, UAE — 2026 | Built to be first. Built to be trusted.
