# International GEO: Prompt Sets Across Markets and Languages

> Step-by-step tutorial: build, run and score an AI search prompt set across markets and languages, with a sheet layout and a monthly rerun routine.

Source: https://mentionx.ai/learn/international/international-geo-prompt-sets-across-markets  
Author: Dipuck Jones, founder of MentionX  
Updated: 30 October 2026  
Level: Practitioner

**To measure AI search visibility across markets, build one prompt set with a fixed intent structure, localise every prompt with native speakers, run it under recorded local settings, and score the answers the same way in every market. Then rerun monthly.** This tutorial gives the steps and a sheet layout.

*Method: we used Google Search Central's guides to multi-regional sites, hreflang and AI features, and Google's March 2026 Search Live announcement, all read on 31 October 2026 (CET). Quotes are verbatim. Opinions are marked "Our view"; examples are marked "Illustrative".*

> **Key takeaways**
>
> - AI answers differ by market and language, so test each market separately.
> - Keep the intent structure fixed across markets and localise the prompts, don't just translate.
> - Record run settings such as language, location and engine so results are comparable.
> - Score consistently and rerun monthly to separate real change from noise.

*Checked on 31 October 2026. "Our view" marks MentionX opinion. Ledgerline is a fictional company used for illustration.*

## Why do answers differ by market?

Google says "Google Search tries to find pages that match the language of the searcher" ([Google](https://developers.google.com/search/docs/specialty/international/managing-multi-regional-sites)). Hreflang annotations "help Google Search point users to the most appropriate version of your page by language or region" ([Google](https://developers.google.com/search/docs/specialty/international/localized-versions)). AI features draw on the same index: to appear as a supporting link, a page "must be indexed and eligible to be shown in Google Search with a snippet" ([Google](https://developers.google.com/search/docs/appearance/ai-features)).

So the same question in Germany and the UK can produce different sources, different brands and different prices. Voice adds more variation: Google says Search Live's model "is also inherently multilingual, which means that people around the world can now speak with Search in their preferred language" ([Google](https://blog.google/products-and-platforms/products/search/search-live-global-expansion/)).

## What do you need?

- A spreadsheet (one tab per market, plus a summary tab)
- A native speaker or local colleague per market
- Access to the AI surfaces you track (for example Google AI Mode, ChatGPT, Perplexity)
- 2–3 hours for the first build, about an hour per market per rerun

## Step 1: Pick two or three markets

Choose your home market and your largest or fastest-growing export market. *Our view:* fewer markets make results easier to compare and fixes easier to attribute.

## Step 2: Fix the intent structure

Define the intents once, in English, and keep them identical across markets. *Fictional example:*

| Intent | Share of set | Example (English master) |
|---|---|---|
| Category discovery | 30% | Best quiet dishwasher for a small kitchen |
| Comparison | 25% | Brand A vs Brand B dishwasher |
| Problem / use case | 20% | Dishwasher that fits a 45 cm gap |
| Brand facts | 15% | Does Brand A offer a 5-year warranty? |
| Purchase | 10% | Where to buy Brand A dishwasher with delivery |

Aim for 20 prompts per market to start.

## Step 3: Localise, don't translate

Send the English master to your native speaker with these rules:

1. Rewrite as a local shopper would actually ask.
2. Use local units, currency and price points.
3. Name local retailers and competitors where natural.
4. Keep the intent the same, even if the wording changes a lot.

*Fictional example:* "Best quiet dishwasher under £500 for a small kitchen" becomes, in Germany, "Welcher leise Geschirrspüler unter 500 Euro passt in eine kleine Küche?" Same intent, local phrasing and currency.

## Step 4: Record run settings

For every run, record:

- Date and time (with time zone)
- Surface and model or mode
- Account state (logged in or out, history on or off)
- Language and location settings
- Who ran it

*Our view:* without this, you can't tell a real change from a different setup.

## Step 5: Run the set

Run each prompt once per surface per market. Save the full answer text and every cited URL. A screenshot helps for disputes but text is what you score.

## Step 6: Score consistently

Use the same columns in every market tab:

| Column | Values |
|---|---|
| Brand mentioned | Yes / No |
| Position | First / Middle / Last / Absent |
| Our domain cited | Yes / No |
| Correct local domain | Yes / No / N/A |
| Facts correct | Yes / Partly / No |
| Competitors named | Free text |
| Notes | Free text |

"Correct local domain" catches a common problem: an answer in German citing your UK page with pound prices.

### How do you summarise?

On the summary tab, per market and surface: mention rate, citation rate, correct-domain rate and fact accuracy. *Fictional example:*

| Market | Mention rate | Cited | Right domain | Facts correct |
|---|---|---|---|---|
| UK | 55% | 40% | 95% | 90% |
| DE | 20% | 10% | 50% | 70% |

These numbers are made up to show the layout, not real results.

## Step 7: Diagnose and fix

Map each weak column to a likely cause:

- **Wrong domain cited.** Check hreflang. Google notes "Localized versions of a page are only considered duplicates if the main content of the page remains untranslated" ([Google](https://developers.google.com/search/docs/specialty/international/localized-versions)). Translate the main content, not just the template, and make annotations return-linked. See [does hreflang matter for AI answers](https://mentionx.ai/learn/international/hreflang-ai-answers-matter).
- **Brand absent.** Check whether you have local pages for that intent at all, and whether local publishers mention you.
- **Facts wrong.** Check local prices, warranty and delivery terms in plain text, and keep them consistent with feeds and markup.

## Step 8: Rerun monthly

Keep prompts frozen for at least a quarter. Add a version column; when you do change a prompt, log the old and new wording and the date. *Our view:* monthly reruns catch movement without drowning you in noise from day-to-day answer variation.

## How do you share results?

Keep the report to one page per quarter:

1. **Headline.** One sentence per market: better, worse or flat, and why.
2. **Summary table.** The four rates per market and surface, with last quarter alongside.
3. **Top three fixes shipped**, each with the prompts it was meant to move.
4. **Next quarter's fixes**, with an owner and a date.

*Our view:* local teams engage more when they see their own market's answers. Paste two or three real answer excerpts per market into the appendix, with the run date and settings, so readers can check the scoring themselves. Never edit or tidy an excerpt; quote it exactly or summarise it and say so.

## Common pitfalls

- Translating prompts word for word
- Changing prompts between runs
- Not recording location and account settings
- Comparing markets with different intent mixes
- Scoring by different people with no written rules

## Template checklist

| Item | Done? |
|---|---|
| Markets chosen (2–3) | |
| Intent structure fixed | |
| Prompts localised by native speakers | |
| Run settings recorded | |
| Scoring rules written down | |
| Summary tab built | |
| Fixes logged with dates | |
| Rerun scheduled | |

For a guided version with an exercise, see the course lesson International prompt sets.

## From MentionX data: why intent structure matters across markets

**Method.** Read-only aggregate queries over MentionX scans from 22 August to 29 October 2026 (UTC): 2,217 engine runs (2,154 completed) across 18 anonymised projects and six engines.

A fixed intent structure lets you compare markets fairly. In the MentionX sample, mention rates vary greatly by intent:

| Intent | Brand mentioned | Share of voice |
| --- | --- | --- |
| Brand-specific | 100% | 64.3 |
| Comparison | 91.9% | 44.6 |
| Recommendation | 29.2% | 26.8 |
| Best-of | 26.6% | 22.7 |
| Problem | 5.7% | 7.7 |

And by engine:

| Engine | Brand mentioned | Answers with no citations |
| --- | --- | --- |
| Gemini | 47.3% | 50.7% |
| OpenAI | 42.9% | 30.2% |
| Perplexity | 31.5% | 4.1% |
| AI Mode | 31.3% | 12.6% |
| Copilot | 26.0% | 47.0% |
| DeepSeek | 24.0% | 99.0% |

*Our view:* if one market's prompt set has more brand-specific prompts than another's, its mention rate will look higher for reasons that have nothing to do with visibility. Keep the mix the same.

**Limits.** A convenience sample of mostly B2B software projects, run through engine APIs, which can behave differently from the consumer apps. Perplexity, Copilot and DeepSeek figures come from a single day (26 October 2026), so treat them as a snapshot, not a trend.

## Worked example: Ledgerline compares the UK and Germany

*Ledgerline is a fictional accounts-payable tool. Figures are illustrative.*

Ledgerline sells in the UK and Germany. Its team writes 30 prompts in English for the UK, then asks a German-speaking colleague to write 30 equivalent prompts for Germany, using local terms for invoice approval and local competitors where they matter.

Both sets keep the same split: ten problem, ten recommendation, five comparison and five brand-specific prompts. Runs are recorded with the language and engine. Ledgerline appears far less often in German recommendation prompts, so the team prioritises German-language reviews and a localised buyer's guide. Every figure here is illustrative.

> **Pro move:** Have a native speaker review each localised prompt. Literal translations often miss how people actually ask.

## How many markets should you start with?

Two or three. More than that multiplies work before you know whether the method works. Add markets once the first ones run smoothly.

## How do you choose local competitors?

Ask local sales teams who they lose deals to. Global competitors may matter less than regional ones. Keep competitor lists per market, labelled consistently.

## Should prompts be identical in meaning?

Equivalent in intent, not identical in wording. A buyer in one market may ask about tax rules that don't exist in another. Keep the intent category the same and adapt the detail.

## How do you handle engines that aren't available everywhere?

Track the engines buyers actually use in each market. Note gaps rather than forcing the same list everywhere.

## How do you report across markets?

Show each market's mention rate by intent side by side, with sample sizes. Highlight the biggest gap and the plan. Avoid one blended global number, which hides the differences.

## What fixes tend to help in a weaker market?

Local-language content, reviews on local sites, local press coverage and accurate local listings. Check which sources engines cite in that market first.

## How do you write a good localised prompt?

Start from how people in that market actually ask. Use local terms, currencies, regulations and examples. A UK buyer might ask about VAT and a German one about e-invoicing rules. Keep the same intent category but let the detail reflect local reality. Ask a native-speaking colleague or local customer to review the wording.

## How do you score answers in another language?

Use the same scoring rules in every market: was the brand mentioned, where, was it described accurately, and which sources were cited. A native speaker should check accuracy, because a description that reads well in translation might still be wrong or odd locally.

## How do you budget time for multiple markets?

Each market adds prompt writing, runs, scoring and fixes. Estimate hours per market for the first month, then adjust. Automating runs saves time, but human scoring for accuracy still takes effort.

## How do you keep markets comparable over time?

Freeze the prompt sets for at least a quarter, and log any change you make with its date. If you must change prompts, run old and new versions side by side for a month so you can tell a real shift from a change in method. Keep engine, language and location settings recorded with every run.

## How do you start this month?

Pick your second-biggest market, write 20 localised prompts with a native speaker and run them alongside your home market. Score both with the same rules and compare mention rates by intent. That small test shows whether a fuller international programme is worth the effort, and where it should start.

## Common questions

**Can we just translate our English prompts?** Start from them, but have native speakers localise wording and examples.

**Do engines answer differently by location?** Often, yes. Record language and location so you can compare.

**How many prompts per market?** Enough to cover each intent several times; 30 to 50 is a practical start.

**Should hreflang affect this?** Hreflang helps search engines choose the right page; check that localised pages exist and are indexable.

## Sources

All accessed 31 Oct 2026.

- Google Search Central, [Managing multi-regional and multilingual sites](https://developers.google.com/search/docs/specialty/international/managing-multi-regional-sites); [Tell Google about localized versions of your page](https://developers.google.com/search/docs/specialty/international/localized-versions); [AI features and your website](https://developers.google.com/search/docs/appearance/ai-features)
- Google, [Search Live global expansion](https://blog.google/products-and-platforms/products/search/search-live-global-expansion/) (26 Mar 2026)

## Frequently asked questions

### How many markets should I start with?

Two or three. Start with your home market plus your biggest export market, so you can compare like for like before scaling.

### Do I need a VPN?

You need results as a local user would see them. Use the platform's own location and language settings where possible, and record exactly how each run was set up.

### Can I machine-translate prompts?

Use translation as a draft only. Have a native speaker rewrite each prompt in natural local phrasing, with local units, currency and retailers.

---

Written and reviewed by Dipuck Jones for MentionX Learn. The HTML version of this page: https://mentionx.ai/learn/international/international-geo-prompt-sets-across-markets