01Notes · 22 August 2026
Your US Prompt Set Measures Nothing in the UK
Quick answer
A June 2026 study ran identical prompts across three engines in seven markets and measured how much the cited sources overlapped. ChatGPT, the most globally consistent engine, carried about 21% of its citations across markets. Google AI Mode carried 14.6%.
The most divergent pair was the United States and the United Kingdom: the pair most teams serve with a single English page.
Key takeaways
- Roughly four fifths of your citation base does not transfer between markets.
- US and UK diverge most, which is the opposite of what most multi-market plans assume.
- Language is not the variable. The source pool is. British publications dominate UK answers and rarely appear in US ones.
- A translated prompt set measures demand that does not exist. Build it in-market.
- Awareness-stage content is the most locally specific, so broad category content is the least portable thing you own.
- The upside: what you build locally is defensible locally, precisely because it does not travel.
What the study found
| Measure | Finding | Consequence |
|---|---|---|
| ChatGPT carry-over between markets | ~21% | Four fifths of the citation base is market-specific |
| Google AI Mode consistency | 0.146 | The most geo-localised engine measured |
| Most divergent country pair | US and UK | A shared language is not a shared source ecosystem |
| Providers recommended in more than one market | None | There is no global default position to inherit |
What this breaks
- One English page for the US and UK. It competes properly in neither source ecosystem.
- Translated prompt sets. A US set converted to British spellings measures queries UK buyers do not type.
- Global share-of-voice reporting. An average across markets that share 21% of their citations describes none of them.
- Sequencing by market size. Building the US programme first and translating outward assumes a transfer that does not happen.
Why language is the wrong variable
The instinct is that markets sharing a language should share results. The data says otherwise, and the reason is straightforward once you look at what assistants are actually reading.
They ground answers in sources, and sources are national. British national publications, British trade press, British forums and British regulators dominate UK answers. Their US equivalents rarely appear, and the reverse holds. Two markets can share every word of vocabulary and share almost none of the corpus.
A better approach
- Build a prompt set per market, in-market. This is the largest cost driver in international work and the one worth paying.
- Agree a competitor set per market. Shortlists differ even in global categories.
- Report each market separately. Always.
- Sequence by revenue, not size. Start where you already sell.
- Consider the less contested markets. Non-English Europe and Latin America are described as largely unclaimed territory, because most brands concentrate on US English.
Method for building the set yourself is published in full, and the market-level detail is on the markets pages.
What this means for you
- Run your prompt set in one other market and compare the cited domains. If they barely overlap, you have your answer.
- Stop averaging markets in reporting. The average of two markets sharing 21% of citations describes neither.
- Treat a local citation base as an asset. It does not travel, which means competitors cannot bring theirs either.
02Common questions
Questions this raises
How much AI visibility carries between markets?
About 21% on ChatGPT, the most globally consistent engine measured. Google AI Mode scored 14.6%. So roughly four fifths of your citation base does not transfer.
Why are the US and UK the most divergent pair?
Because language is not the variable that matters. The source pool is: British national publications, trade press and forums dominate UK answers, and their US equivalents rarely appear.
Can I translate my prompt set?
You can, and it will measure the wrong thing. A translated prompt is not how buyers in that market ask. The set has to be built from in-market buyer language.
Does this mean I need separate content per market?
Where the market is a real revenue line, usually yes. Where the only difference is spelling, separation adds maintenance cost for little gain and we will say so.
Which markets are least contested?
Non-English European markets, India and Latin America are described in 2026 analysis as largely unclaimed AI citation territory, because most brands concentrate resources on US English.
Sources
Every figure on this page comes from one of the following. Where a claim is ours rather than a source’s, it carries an evidence label instead.
- Yotpo — cross-market citation study, June 2026. Identical prompts run across ChatGPT, Gemini and Google AI Mode in the US, UK, Germany and Australia, measuring citation-set overlap. Source for the ~21% carry-over, the 14.6% AI Mode consistency score and the US–UK divergence finding. Independent study
- EchoWi — three-market comparison, August 2026, US, Spain and France. No provider was recommended consistently in more than one market. The author discloses a competing interest in the category measured, which we repeat here rather than omit. Independent study
Ask an assistant
Hand this page to an assistant and ask it to summarise or challenge the argument. If it gets something wrong about us, that is itself a useful finding.
→Next step