Koozai Places 100 Solutions to the Take a look at
AI-powered search guarantees to make discovering someplace to eat simpler than ever. However a Koozai investigation into restaurant suggestions in Portsmouth has recognized closed companies and unsuitable venues amongst ideas from ChatGPT and Google AI.
AI restaurant suggestions put to the check
From discovering a dog-friendly pub to selecting a restaurant for a special day, extra customers are turning to AI-powered instruments to assist slender down their choices.
Somewhat than scrolling by way of restaurant web sites, menus and opinions, customers can ask ChatGPT or Google’s AI-powered search options for personalised suggestions in seconds.
However how correct are the solutions?
To analyze, UK digital advertising and marketing company Koozai examined ChatGPT and Google AI utilizing ten equivalent restaurant-search prompts centered on Portsmouth, Hampshire.
The analysis examined 100 restaurant advice slots throughout each platforms, overlaying every part from inexpensive Sunday roasts and seafood eating places to gluten-free eating, family-friendly venues and waterfront eating places.
The findings recognized 4 suggestions that failed to satisfy the unique search necessities, together with two venues recorded as completely closed and two seafood companies that didn’t meet the request for eating places serving lobster or seafood platters.
These findings characterize recognized failures inside the analysis, quite than a last total accuracy charge for both platform.
Key findings from Koozai’s AI restaurant analysis
- 100 advice slots have been examined throughout ChatGPT and Google AI.
- Ten equivalent prompts have been examined on every platform, requesting 5 suggestions per immediate.
- ChatGPT and Google AI agreed on simply 16% of equal restaurant rating positions, with solely eight matches throughout 50 comparisons.
- Two really helpful eating places have been subsequently recognized as completely closed: Clarence Pier Brewers Fayre and Samphire.
- Two ChatGPT seafood suggestions didn’t fulfill the requested restaurant and menu standards.
- Three recognized failures appeared in ChatGPT’s 50 advice slots and one appeared in Google’s 50.
- The analysis additionally flagged examples of doubtless mismatched sources and restaurant data requiring additional verification.
The outcomes show why AI suggestions should be assessed in opposition to the shopper’s precise necessities, not merely whether or not a enterprise identify seems in a solution.
AI really helpful eating places that had completely closed
One of many clearest issues recognized throughout the investigation concerned AI recommending eating places that have been now not buying and selling.
ChatGPT really helpful Clarence Pier Brewers Fayre
When requested to advocate 5 family-friendly eating places in Portsmouth with kids’s menus and excessive chairs, ChatGPT included Clarence Pier Brewers Fayre.
Nevertheless, the restaurant had completely closed. Koozai’s analysis information point out a closure date of 30 June 2026, earlier than the AI check passed off on 20 October 2026.
A Google enterprise itemizing screenshot subsequently provided throughout verification additionally identifies the restaurant as completely closed.
Clarence Pier Brewers Fayre was really helpful by ChatGPT regardless of being listed as completely closed. Supply: Google enterprise itemizing screenshot provided for Koozai’s analysis.It is a significantly clear instance of a advice failing to satisfy a shopper’s wants. A completely closed restaurant can not present a household meal, no matter whether or not it beforehand supplied kids’s menus or excessive chairs.
Google AI really helpful Samphire in Southsea
Google AI produced the same concern when requested to advocate 5 Portsmouth eating places serving recent seafood, together with lobster or seafood platters.
Amongst its ideas was Samphire, a seafood restaurant in Southsea.
Nevertheless, the enterprise was recognized as completely closed in Koozai’s analysis. A screenshot of its Google enterprise itemizing additionally shows a completely closed discover.
The presence of closed venues in AI-generated suggestions raises questions on how reliably AI techniques establish whether or not native companies are nonetheless working.
For customers, inaccurate buying and selling data can result in wasted journeys and disrupted plans. For native companies, it additionally raises questions on whether or not AI suggestions are directing potential clients in direction of genuinely out there options.
ChatGPT additionally really helpful seafood companies that didn’t match the search
Outdated enterprise data wasn’t the one downside recognized.
The investigation additionally discovered examples the place a really helpful enterprise existed however didn’t fulfill the particular necessities of the person’s search.
For the seafood situation, the precise immediate was:
Suggest 5 eating places in Portsmouth serving recent seafood, together with lobster or seafood platters.
Two ChatGPT suggestions have been recognized as unsuitable for this request.
The Fisherman’s Kitchen: A fish and chip store quite than a lobster or seafood platter restaurant
ChatGPT really helpful The Fisherman’s Kitchen in Southsea.
Nevertheless, the enterprise operates as a fish and chip store quite than the kind of seafood restaurant requested.
Its on-line ordering menu options conventional fish and chip dishes, together with cod, haddock, chips and different takeaway favourites.
The analysis didn’t establish the requested lobster or seafood platters as a part of its providing.
The Fisherman’s Kitchen’s on-line menu illustrates its fish and chip providing, quite than the lobster or seafood platters requested within the AI search. Supply: The Fisherman’s Kitchen web site screenshot provided for Koozai’s analysis.The difficulty isn’t that The Fisherman’s Kitchen is an unsuitable enterprise generally. It’s that the advice didn’t match the actual eating expertise and menu necessities the shopper had requested.
Hartgill Seafoods: A seafood kiosk quite than a sit-down restaurant
ChatGPT additionally really helpful Hartgill Seafoods, situated at The Arduous in Portsmouth.
Hartgill Seafoods is a conventional seafood and shellfish kiosk, quite than a sit-down restaurant.
Though it’s a real seafood enterprise, its format doesn’t meet the request for a restaurant the place clients can dine.
This highlights a distinct kind of AI advice downside: figuring out a enterprise that’s associated to the requested class with out establishing whether or not it supplies the expertise the shopper truly desires.
ChatGPT vs Google AI: What did the analysis discover?
Similar query, totally different dinner: ChatGPT and Google AI agree on simply 16% of restaurant rating positions
It’s not simply the accuracy of AI suggestions that differs between platforms. Koozai’s analysis additionally discovered placing variations by which eating places ChatGPT and Google AI selected to advocate.
Throughout ten equivalent restaurant search situations in Portsmouth, simply eight of the 50 equal rating positions featured the identical restaurant on each platforms – an settlement charge of solely 16%.
Which means when each platforms have been requested precisely the identical query, they ceaselessly really helpful totally different companies or positioned them in several positions.
For eating places, this highlights an vital problem. Showing prominently in a single AI platform’s suggestions doesn’t essentially imply a enterprise will take pleasure in the identical visibility elsewhere.
The findings counsel that companies can not assume robust visibility on one search platform interprets into equally robust visibility throughout others.
What does this imply for companies?
Sophie Roberts, Managing Director at Koozai, mentioned:
“What’s significantly fascinating about these findings is simply how otherwise AI platforms can reply to precisely the identical query. A restaurant may very well be the primary advice on one platform and never seem in any respect on one other.
“For companies, for this reason being seen on-line can now not be about specializing in one search engine or chasing a single rating. Clients are discovering manufacturers in additional locations than ever, and every platform might current a distinct set of suggestions.
“At Koozai, we take a search-everywhere strategy, ensuring companies have robust technical search engine optimisation foundations, clear and useful web site content material, correct enterprise data and a reputable on-line presence past their very own web sites.
“There’s no assured system for showing in AI suggestions, however giving search platforms constant, reliable and genuinely helpful data places companies in a stronger place to be found, whichever platform their clients select.”
Throughout the 100 recorded advice slots, Koozai recognized 4 failures in opposition to the unique search necessities.
| Platform | Restaurant | Challenge |
|---|---|---|
| ChatGPT | Clarence Pier Brewers Fayre | Completely closed |
| Google AI | Samphire | Completely closed |
| ChatGPT | The Fisherman’s Kitchen | Didn’t meet the requested seafood restaurant and menu standards |
| ChatGPT | Hartgill Seafoods | Seafood kiosk quite than a sit-down restaurant |
Three of those recognized failures appeared amongst ChatGPT’s 50 suggestions, equal to six% of its advice slots.
One appeared amongst Google AI’s 50 suggestions, equal to 2% of its advice slots.
These figures describe the minimal proportion of recognized failures within the recorded dataset. They don’t seem to be complete error charges, as a result of not each advice has been independently verified in opposition to each relevant search requirement.
They need to due to this fact not be interpreted as proof that Google AI is mostly extra correct than ChatGPT.
Why a restaurant can exist and nonetheless be the flawed advice
The analysis highlights an vital distinction between figuring out a related enterprise and answering a buyer’s precise query.
Take into account a number of the situations examined:
- A restaurant might serve seafood with out providing lobster or seafood platters.
- A pub might welcome canine outdoor however not enable them inside throughout dinner.
- A restaurant might stay open till late whereas its kitchen stops serving meals earlier.
- A venue might provide vegetarian dishes with out offering a considerable selection of vegetarian primary programs.
- A restaurant might promote gluten-free choices with out clearly labelling them on its present menu.
- A venue could also be near the waterfront with out providing outside tables with waterfront views.
These distinctions matter as a result of customers are more and more utilizing conversational searches to precise particular wants quite than merely asking for an inventory of close by companies.
For AI-powered native discovery to be genuinely helpful, suggestions have to fulfill these circumstances.
Can customers belief the sources behind AI solutions?
The investigation additionally recorded potential weaknesses in how some suggestions have been supported by sources.
For instance, one ChatGPT response recommending vegetarian eating places in Portsmouth displayed a supply heading referring to Banana Tree in Henley.
Different suggestions referenced common enterprise listings, supply companies or third-party web sites when making extra particular claims about restaurant menus, opening hours and amenities.
These examples don’t mechanically set up that the restaurant suggestions have been incorrect.
Nevertheless, they show why a visual supply hyperlink shouldn’t be handled as proof that each declare in an AI reply has been verified.
A common restaurant itemizing would possibly affirm {that a} venue exists with out establishing whether or not its kitchen serves dinner after 8pm on Mondays.
Equally, an internet menu might not affirm whether or not a restaurant has excessive chairs, permits canine inside or supplies appropriate preparations for purchasers with particular dietary necessities.
For customers, the problem is distinguishing between a solution that appears authoritative and one that’s supported by present, related proof.
What does this imply for eating places and different native companies?
AI-powered search is creating one other route by way of which potential clients can uncover companies.
Somewhat than relying solely on typical search outcomes, customers can ask AI platforms to match venues in opposition to more and more detailed necessities.
That creates a possibility for companies whose data clearly demonstrates what they provide.
Nevertheless, the Portsmouth analysis additionally illustrates how inaccurate or incomplete data can have an effect on the usefulness of AI suggestions.
Eating places and different native companies ought to take note of the data out there about them throughout their web sites, enterprise profiles, menus and related third-party platforms.
Helpful steps embody:
- Preserving opening hours and momentary closure data correct.
- Publishing present menus and costs on official web sites.
- Clearly describing amenities equivalent to outside seating, excessive chairs and dog-friendly areas.
- Offering correct details about dietary choices and allergens.
- Making the character of the enterprise clear, together with whether or not it affords sit-down eating, takeaway or each.
- Checking how the enterprise seems in AI-generated suggestions for related buyer searches.
These steps can not assure that an AI platform will advocate a enterprise accurately, however they assist make dependable data out there to serps, AI techniques and clients.
What ought to customers verify earlier than counting on AI suggestions?
AI instruments may be useful for locating eating places and narrowing down choices, however vital particulars ought to nonetheless be checked earlier than visiting.
Shoppers ought to affirm {that a} restaurant is presently buying and selling, significantly when planning a journey or special day.
Costs and menus ought to be checked in opposition to the venue’s newest official data.
For late-night eating, kitchen closing instances matter greater than common enterprise opening hours.
And the place dietary necessities or allergy symptoms are concerned, clients ought to affirm the related data straight with the restaurant quite than relying solely on an AI-generated abstract.
The extra particular the requirement, the extra vital correct verification turns into.
What does the analysis inform us about the way forward for AI search?
The findings spotlight an vital problem as AI turns into a part of on a regular basis shopper decision-making.
Search accuracy is now not merely about whether or not a system identifies a related web site or enterprise.
When somebody asks for a restaurant serving lobster, a family-friendly venue with excessive chairs or someplace to eat after 8pm, they count on suggestions that fulfill these precise circumstances.
A enterprise may be actual, common and geographically related whereas nonetheless being the flawed reply.
And a advice can look convincing even when the enterprise has stopped buying and selling.
For manufacturers, this makes correct on-line data and visibility throughout a number of discovery platforms more and more vital.
For customers, it reinforces the necessity to verify important particulars earlier than appearing on AI-generated suggestions.
For AI platforms, the problem is guaranteeing that solutions aren’t simply believable, however helpful, present and sufficiently supported by proof.
Analysis methodology
Who carried out the analysis?
The analysis was carried out by Koozai, a UK digital advertising and marketing company specialising in search engine optimisation, AI search, digital PR, content material advertising and marketing and paid media.
When was the analysis carried out?
The AI searches have been carried out on 20 October 2026.
Supporting checks and proof gathering continued after the unique check.
Which AI platforms have been examined?
ChatGPT and Google AI have been examined utilizing the identical ten restaurant-search prompts.
The analysis information establish the platforms by these names. Particular mannequin variations and Google AI interface configurations weren’t documented within the provided dataset.
How have been the searches carried out?
Every immediate was submitted in a separate, logged-out, incognito shopping session.
This strategy was meant to cut back the affect of earlier conversations, account historical past and account-level personalisation.
Nevertheless, incognito shopping doesn’t remove all contextual influences, together with location, platform configuration or variations in responses generated between periods.
What was the pattern dimension?
The analysis used ten search prompts, every requesting 5 restaurant suggestions.
This produced 50 advice slots from ChatGPT and 50 from Google AI, giving 100 advice slots in whole.
These are advice occurrences quite than 100 essentially distinctive eating places. The identical enterprise might seem in multiple search situation.
What have been the precise prompts?
- Suggest 5 eating places or pubs in Portsmouth serving a conventional Sunday roast for below £20 per particular person.
- Discover 5 dog-friendly eating places or pubs in Portsmouth the place I can have dinner with my canine.
- Suggest 5 eating places in Portsmouth serving recent seafood, together with lobster or seafood platters.
- The place can two individuals have a primary course every in Portsmouth for below £35 in whole? Suggest 5 locations.
- Discover 5 eating places in Portsmouth providing a wide selection of vegetarian primary programs.
- Suggest 5 unbiased eating places in Portsmouth appropriate for a romantic anniversary dinner.
- Discover 5 family-friendly eating places in Portsmouth with kids’s menus and excessive chairs.
- Suggest 5 eating places in Portsmouth with clearly labelled gluten-free dishes.
- Discover 5 eating places in Portsmouth that serve dinner after 8pm on a Monday.
- Suggest 5 eating places in Portsmouth with waterfront views and outside seating.
What data was recorded?
For every advice, the analysis recorded the restaurant identify, platform, immediate, advice place, clickable enterprise data, supply hyperlinks and related evaluation fields.
The place out there, the evaluation thought of buying and selling standing, Portsmouth location, possession kind, menu accuracy, costs, opening hours, dietary data and amenities.
Notes have been additionally recorded the place claims have been unsure, sources appeared mismatched or extra verification was required.
How have been incorrect suggestions recognized?
A advice was thought of non-compliant the place the analysis recognized proof that it failed to satisfy a fabric requirement of the unique immediate.
Examples included a completely closed venue being really helpful as someplace to eat, or a seafood enterprise not satisfying the requested restaurant and menu standards.
A advice was not mechanically thought of incorrect just because its supporting data was incomplete or a supply hyperlink appeared questionable.
Unverified data was stored separate from recognized errors.
How have been the chances calculated?
Rating settlement was calculated by evaluating the restaurant really helpful at every equal place (first to fifth) throughout the ten equivalent prompts. Of the 50 position-matched comparisons, eight featured the identical restaurant on each platforms, giving a rating settlement charge of 16%. Eating places showing in several positions weren’t counted as matches. This measures settlement between the 2 recorded outputs, not the general accuracy or consistency of both platform.
Every platform produced 50 advice slots.The proportion of recognized failures was calculated by dividing the variety of recognized non-compliant suggestions by 50 and multiplying by 100.
For ChatGPT, three recognized failures divided by 50 produced 6%.
For Google AI, one recognized failure divided by 50 produced 2%.
Throughout each platforms, 4 recognized failures divided by 100 produced 4%.
These are noticed minimal failure proportions inside the analysis information, not last independently verified accuracy or error charges for both platform.
What are the restrictions?
The research represents a snapshot of AI-generated suggestions on one check date in a single UK metropolis.
Every immediate was examined as soon as per platform, so the analysis doesn’t set up how persistently both platform would reply the identical query throughout repeated periods.
The pattern isn’t consultant of all AI searches, places, restaurant classes or customers.
Not each advice has been independently verified in opposition to each relevant criterion. In consequence, the analysis can not set up an total accuracy rating between the 2 platforms.
Closure screenshots provided throughout follow-up verification present the companies’ listed standing on the time the screenshots have been captured. They need to be thought of alongside dated closure proof when establishing buying and selling standing on the unique check date.
Menu screenshots equally doc the data seen when captured and don’t essentially show that the identical menu was out there on 20 October 2026.
Findings ought to due to this fact be understood as recognized points in an outlined observational check, quite than common conclusions in regards to the reliability of both AI platform.
Analysis information and media enquiries
Journalists thinking about Koozai’s AI restaurant advice analysis can request additional details about the methodology, precise search prompts, recommendation-level dataset and supporting proof.
The analysis is meant to contribute to the broader dialogue about AI search accuracy, native enterprise discovery and the way customers assess AI-generated data.
About Koozai
Koozai is a UK digital advertising and marketing company with experience in search engine optimisation, AI search, digital PR, content material advertising and marketing and paid media.
The company helps companies enhance their on-line visibility and perceive how altering search behaviours affect the way in which clients uncover manufacturers, services.
