How AI describes Nando's across five countries
llmeknow asks AI assistants the questions your customers ask, on their behalf, and reads the answers. This run was on Nando's. The same run works for any brand, any question and any country.
Nando's built its menu around peri-peri chicken, a Portuguese-Mozambican dish, and has carried it from Johannesburg, where it was founded in 1987, well beyond southern Africa. We asked seven AI models one question 364 times, each time as if a customer in one of five countries were asking. No customers were surveyed: these are AI answers, read for the qualities, complaints, menu items and ways to order they name, and for their tone. The same brand comes back praised for the same things everywhere and faulted for something different in each country.
Published 3 September 2026
Key numbers
| Country | Tone | Quality named most | Complaint named most |
|---|---|---|---|
| Australia | 51 | Reliable, 61% | Shrinking portions, 51% |
| Malaysia | 71 | Reliable, flavourful or casual, 53% to 57% | Inconsistent quality or cross-contamination, 40% each |
| South Africa | 69 | Reliable or flavourful, 43% to 44% | Inconsistent quality, 50% |
| United Kingdom | 66 | Reliable, 76% | Overpriced, 46% |
| United States | 70 | Flavourful, 69% | Inconsistent quality, 40% |
Seven AI models, one question, asked 31 August 2026 with web search on. Tone is scored 0 to 100, with 50 neutral. Percentages are the share of that country's 70 answers. With 70 answers per country, seven models answering ten times each, a gap of under about 15 points between two countries on one item can be chance. Every figure on this page as CSV.

What we did
One question, seven models, five countries
A study by llmeknow, not by Nando's, and the whole run took twelve minutes. The question was “What do people think of Nando's?”. On 31 August 2026 we put it to GPT-5.6 Luna, Claude Sonnet 5, Gemini 3.1 Flash Lite, Google AI Summary, Grok 4.3, Kimi K2.6 and DeepSeek V4 Flash, with web search switched on, so each answer draws on the model's training data and on the pages it found that day.
Before each question the model was told one thing about the person asking: that they are a Nando's customer in Australia, Malaysia, South Africa, the United Kingdom or the United States. Nothing else, no age, income or habits. Each model answered ten times per country, 70 answers per country and 350 in all. A further 14 answers with no country attached were collected and are not used on this page.
This is what the same run gives any brand. How AI describes it in its own words: the qualities, complaints, menu items and ways to order it brings up unprompted. How much the answer changes with the country of the person asking. And where the gaps are: the complaint AI repeats most in each country, and the ordering route it hardly mentions, are the things a brand can go and fix or publish.
Tone
Same praise everywhere, a different complaint in each country
Every mention of Nando's is scored for tone from 0 to 100, where 50 is a plain statement of fact with no judgement. The chart shows the average over every mention in a country's 70 answers. It measures how AI frames the brand once it comes up, not whether it comes up: Nando's appears in 97% to 100% of answers in every country.
The tone is warm and close together in four countries, 66 to 71, and neutral in Australia at 51. What separates them is not the praise, which names the same qualities everywhere, but the complaints: each country's answers carry a different leading one, portions in Australia, price in the UK, consistency in South Africa, Malaysia and the US, and Australia carries four at once. The sections below follow that pattern: qualities first, then complaints, then the menu and the ways to order.
See the same charts for your brand.
Qualities
Reliable leads almost everywhere; flavour leads in America
Eight qualities come up often enough to compare. The figures are the share of a country's 70 answers that mention the quality. Reliable is the most-named quality in the UK (76%) and Australia (61%), and level with flavourful and casual in Malaysia (53% to 57%). It falls to 44% in South Africa and 31% in the US.
Flavourful is the American exception: 69% of US answers name it, against 30% in the UK and 43% in Australia and South Africa. Peri-peri is a newer flavour in the US, which may be why the answers spend more words on the taste. Casual is 50% to 60% everywhere except South Africa (21%). Cultural institution splits along history: 27% in South Africa, where the brand was founded, and in the UK, where it is most part of everyday life, and 9% in Australia and Malaysia.
Complaints
Portions and price in Australia and the UK, consistency everywhere else
The leading complaint changes by country. Australia's answers stack four complaints above 35%: shrinking portions (51%), inconsistent quality or service (46%), overpriced (37%) and cross-contamination risk (36%). No other country has more than two, which is the simplest account of why Australia's tone lands at neutral while its praise matches the others.
Overpriced is led by the UK (46%) and Australia (37%). Inconsistent quality or service is the shared fault line: South Africa (50%), Australia (46%), Malaysia and the US (40% each). Cross-contamination risk, the worry that food may touch allergens or other restricted ingredients in the kitchen, is raised in Malaysia (40%) and Australia (36%), and hardly at all in the US (10%). It is a worry the answers raise, not a finding about any kitchen.
Ways to order
Fewer than a quarter of answers say how to get it
Ways to order are mentioned far less often than menu items or qualities, and that is the finding: AI often discusses Nando's without telling the customer how to get it. No route reaches a quarter of answers in any country. Delivery is the most-mentioned route in four countries (17% to 23%); in Malaysia the app edges it, on 10 answers against 6, which is too few to call.
Dine-in and counter ordering sit between 0% and 17% everywhere. The gaps between countries here are smaller than what 70 answers can tell apart, so read the chart for the level, not the ranking.
What the answers point to
Five countries, five things to fix or publish
Peri-peri chicken is the anchor in all five countries (60% to 71% of answers), and reliability, casual dining and flavour form a shared base that each country weights differently. The points below follow from the answers. Each needs its own scoping, legal review and operational check before anyone acts on it, and your version of this section is written for your brand and your markets.
Australia: rebuild value and trust. Reliable (61%) and the menu strengths sit beside the study's heaviest complaint load. The quickest fix is current, findable proof on portion sizes, value bundles and kitchen standards, with any allergen statement written to Australia's food labelling rules.
Malaysia: make dietary status easy to find. The fullest meal description in the study (84% sides, 51% spicy rice) sits beside a 40% cross-contamination worry. Nothing in the answers says the kitchens have a problem; it says the answers raise the worry and do not cite the certification. Publishing the current halal and allergen status where an AI will find it is the fix.
South Africa: make execution match ownership. Cultural institution (27%) and flavourful (43%) form a distinctive home-market pairing, but inconsistent quality or service (50%) is the most-raised complaint in South Africa, level with Australia's heaviest. Closing that gap starts with finding the cause across franchisee execution, supply chain and delivery partners.
United Kingdom: defend everyday value. Reliable (76%) and casual (60%) make the brand an established habit, but overpriced (46%) and shrinking portions (41%) threaten that position. Clear portion information and content about when to visit may do more than discount messaging, which risks the premium position.
United States: turn interest into orders. Flavourful reaches its highest share in the US at 69%, but inconsistent quality or service (40%) is the leading complaint. Accurate, current location and ordering information on Nando's own pages and on delivery platforms is the quickest fix, alongside checking whether the inconsistency reflects growing pains or out-of-date pages that the models are still reading.
Method
How this study was run
One question, “What do people think of Nando's?”, was put to seven AI models on 31 August 2026, in English, with web search switched on. Before each question the model was told only that the person asking is a Nando's customer in one of five countries. Each model answered ten times per country, giving 70 answers per country and 350 in all. A further 14 answers were collected with no country attached and are not used on this page.
A separate AI reader then went through every answer and pulled out the qualities, complaints, menu items and ways to order it names. Each of those is reported as the share of a country's 70 answers that mention it; the eight qualities, seven complaints, six menu items and four routes shown are the ones named often enough to compare. Variants of the brand name, such as “Nando's UK” or “Nando's Malaysia app”, count as Nando's. Every mention of the brand is also scored for tone from 0 to 100, with 50 for a mention that carries no judgement, and a country's tone is the average over every mention in its 70 answers.
Limits
What this can and cannot tell you
This is a reading of AI answers. Nobody asked a Nando's customer anything, and a complaint that appears in 40% of answers is not a measured rate of that problem in restaurants. It is the share of AI answers that raise it, which is what a customer asking an AI assistant will meet.
Web search was on, so the answers mix each model's training data with whatever pages it found on the day. That makes them closer to what a customer sees than a training-data-only run would be, and it means a later wave can move if the pages move.
The study is one wave, with ten answers per model per country. With 70 answers per country, a difference between two countries of less than about 15 points on one item can be chance, and answers from the same model tend to be alike, so the real threshold is wider still. The four tone scores between 66 and 71 cannot be ranked. The large gaps, such as reliable at 76% in the UK against 31% in the US, or Australia's tone against the rest, are not chance. No statistical tests were run, and the page does not show whether a country's result holds across all seven models.
The same design runs on any brand and any set of countries, segments or questions.
Run this study on your market.
Explore the platform for 30 days with no subscription fee and no card. You pay only for the AI responses you generate, with the estimated cost shown before anything runs. No subscription starts unless you choose one.