TechmadeazyAsk the same question from London and from Dubai and you will get different brands back. Ask it in English and then in Arabic and you may get a different set again. Ask it from Bengaluru in the mix of English and Hindi that people actually type, and you might get a generic answer with no brands in it at all.
Most brands tracking their AI visibility are tracking one number, from one location, in one language. If you sell across borders, that number is describing a market you may not even be in.
Why answers move between countries
Four things change when the location changes.
The sources change. Assistants lean on local review sites, local publications, local community threads and local retailers. A brand that dominates the sources cited in the US may be entirely absent from the ones cited in Australia.
The competitive set changes. The three brands that take every answer in the UK are often not the three that take every answer in India. Your real competitor in an AI answer is whoever is being recommended instead of you in that market, which is a different list per country.
The practical constraints change. Can you buy it here, how long does it take, who pays the duty. These are the questions that decide whether a brand even qualifies for the answer, and they are country-specific by definition.
The price framing changes. "Under 5,000 rupees" and "under 80 dollars" are not the same filter applied to the same catalogue.
Language is not translation
This is the part that catches people out, and it cost us a month once, so I will be specific about it.
We built Arabic pages for a client by translating the English ones. Machine translation, lightly reviewed. They went live, and Arabic answers did not move at all for four weeks. We rewrote them natively with a copywriter in the market, and Arabic prompt coverage started climbing within three weeks of that.
The lesson is not that machine translation is bad. It is that a translated page answers the question the English speaker asked, phrased the way an English speaker would phrase it. Buyers in another language ask different questions, with different assumptions, in different words. The Arabic buyer asking about a fragrance in Gulf summer heat is asking something the English page was never written to answer.
Then there is the mixed case. In India, a large share of real buyer questions come in Hinglish, typed the way people speak. "Filter coffee ke liye best powder" is not a translation problem, it is a phrasing problem, and most brand content is written in a register that nothing about that sentence matches.
Track per market, not in aggregate
An aggregate share of answers across four countries can stay completely flat while one market doubles and another halves. It is the least informative number you can produce.
Run the same prompt set from each market you actually sell to, score each as its own set, and report them side by side. It is more work and it is the difference between knowing what is happening and having a figure for the deck.
Where two markets diverge sharply, that is usually your highest value finding of the quarter. It tells you where the sources are weak, and sources are the slowest thing to build, so knowing early matters.
The three facts that decide cross-border answers
If you sell internationally and do nothing else, put these in text on your site.
Which countries you ship to. Not visible only at checkout, and not implied by a currency selector. Stated.
How long delivery takes, by country. A range is fine. Silence is not.
Who pays the duty. This one is doing more damage than brands realise. When it is not stated anywhere, assistants fill the gap with general advice, and the general advice about ordering from small overseas sellers is a warning about surprise charges. You are being editorialised about because you left a blank.
If you sell in four markets and can fund one
Pick by where the gap between demand and your presence is widest, not by where revenue is highest today.
Your biggest market usually already has some sources, some reviews and some coverage. The market where you are invisible but the buyers are asking is where the same effort moves the number most. Look at your per-market scores, find the one where competitors are named in almost every answer and you are named in none, and start there.
What does not work
Running one prompt set from one location, usually wherever your agency happens to sit, and treating it as a description of your business. We have picked up accounts where six months of reporting described a market the client barely sold into.
Also: launching translated pages and concluding the language does not matter when they underperform. They underperformed because they were translated, and you will draw exactly the wrong conclusion from it.
Where to start
- List the markets that make up 80% of your revenue.
- Run the same prompt set from each, logged out, and score them separately.
- Compare the competitor names that come back.
- If the lists differ, you have confirmed you are running several businesses, not one.
- Put shipping countries, delivery times and duty responsibility in text this week.
