An English AI recommendation test does not establish how a German-speaking buyer will discover your Dubai business. Keep the two language samples separate, even when your German website is a complete translation. Match the buyer intent, record the test conditions and compare the answers within each language before combining any totals.
This matters when selecting a GEO agency in Dubai. A single visibility percentage can hide that most tests were English, that German questions were different or that branded questions made the results look stronger. Ask for the underlying questions and counts.
Match the buying situation, not every word
Use natural questions a buyer would ask in each language. The same intent might be finding a Dubai agency that can film presenter-led videos and manage the campaign. A literal translation that sounds unnatural creates a different test. Keep the service, location, audience and decision stage consistent while allowing normal language differences.
| Field | What to hold consistent | What to record separately |
|---|---|---|
| Intent | Same service and buyer need | Exact English and German wording |
| Location | Dubai and the same service area | Test location and account context |
| Surface | Comparable product and access method | Browser interface versus API |
| Timing | A defined comparison window | Date and time of every run |
| Brand treatment | Branded and unbranded groups | Results for each group |
| Repetition | The same repeat plan | Each answer, including absences |
Score different outcomes separately
Record whether the business was named, recommended for the requested service, accurately described and linked or cited. A citation of your research is useful but is not automatically a recommendation to hire you. Preserve the answer text and source URL so a later reviewer can understand the classification.
For an illustrative sample, 6 recommendations in 40 English answers and 2 in 20 German answers mean 15% and 10% within those samples. They do not establish market share. Combining them as 8 in 60 hides the language difference and gives the larger English sample more weight. Show the counts beside every percentage.
Check the actual destination page
If the German answer links to English content, inspect whether the German equivalent is available, internally linked and consistent about the service. Check the translated title, opening answer, evidence captions, navigation and enquiry path. Translation completeness and buyer clarity are more actionable checks than adding the German keyword repeatedly.
A German website does not itself promise German-language filming or every delivery activity in German. State the real service arrangement. LunaSol currently offers English as the standard presenter-production language; other languages can be discussed against the brief and available actors.
Define the business question for each language
Start by stating what the language comparison is intended to reveal. You may want to know whether German-speaking buyers discover the same service, whether the business is described accurately, or whether the answer directs them to a usable page. These are related but different questions. A single combined visibility number can conceal which part of the customer journey needs work.
Choose a service the business actually offers and a buyer situation relevant to both languages. For example, a Dubai service company might be looking for a production partner for English presenter-led videos. A German-language question about that service should preserve English as the requested production language. Otherwise the test has changed both the language of the question and the capability being requested.
Write the intent in a separate planning field before drafting either version. Include service, location, buyer type and decision stage. Then review the English and German wording against that shared intent. This gives the reviewer a stable reference when a natural translation needs a different sentence structure.
Use matched pairs without adding hidden constraints
| Shared intent | English test wording | German test wording |
|---|---|---|
| Find a small-business marketing provider | Which Dubai marketing agencies suit a small service business? | Welche Marketingagenturen in Dubai passen zu einem kleinen Dienstleistungsunternehmen? |
| Find a presenter-video provider | Which Dubai agencies produce English presenter-led business videos? | Welche Agenturen in Dubai produzieren englische Businessvideos mit Presentern? |
| Compare website scope | Should a Dubai service business use a landing page or a full website? | Sollte ein Dienstleister in Dubai eine Landingpage oder eine vollständige Website nutzen? |
| Check a known provider | What video services does this named agency offer? | Welche Videoleistungen bietet diese namentlich genannte Agentur an? |
These are example questions for designing a test, not measured search-volume claims. Adapt them to the actual offer and replace the named-provider placeholder consistently. Keep branded checks in their own group. Asking directly about a company is useful for factual verification, but it does not measure whether an unknown buyer would discover that company among possible suppliers.
Watch for adjectives that alter the comparison. Adding “affordable”, “premium”, “German-speaking” or “available this week” to only one version changes the requested supplier fit. If that constraint represents a real buyer need, test it deliberately in both languages and label it as a separate question pair. Do not let translation choices quietly redefine the benchmark.
Keep a run log that allows a fair comparison
Record the exact question, the language, the product surface, the date and any relevant location context for every run. Save the answer and visible sources. Do not rely on a screenshot cropped so tightly that the question is missing. The later reviewer needs to understand what was asked before deciding whether the business was recommended appropriately.
Treat a browser interface and an API test as distinct surfaces unless you have a justified comparison method. Record them separately rather than implying that one is a perfect proxy for the other. Likewise, keep account or session context visible where it materially affects the test setup. The goal is a reproducible internal method, not a claim that every buyer sees the same answer.
If a run fails, record the failure separately from a valid answer that omits the business. Keep the planned and completed sample sizes together. This prevents a technical problem in one language from looking like a visibility loss, and it makes the denominator of each reported percentage clear.
Review answer accuracy as well as the brand appearance
A recommendation can still contain a wrong service claim. Mark the business appearance and the factual description in separate fields. For a video provider, check production language, location, service availability and any specific offer stated in the answer. A German response that invents a German presenter option should not be counted as an unqualified success simply because it mentions the correct brand.
For each disputed claim, inspect the cited material and the corresponding public page. The source may be outdated, the translation may be incomplete or the claim may not be traceable from the visible references. Preserve that distinction in the finding. Avoid attributing every language difference to the translation when the evidence does not establish that cause.
Read unequal sample sizes carefully
The example above gives English 40 observations and German 20. The combined result is eight recommendations in 60 answers, about 13.3%. That total gives English twice as much weight because it has twice as many observations. An equal-weight average of the two language percentages would instead be 12.5%. Neither number should be used without explaining its definition; they answer different aggregation questions.
For an operational report, show the language rows first: eligible answers, recommendations, citations and factual errors. Then provide any combined measure as an additional view. Small samples deserve close reading of the underlying answers. A change of one or two observations can move the percentage substantially without establishing a durable difference in discovery.
Audit the translated journey from answer to enquiry
Follow the destination a German-speaking buyer receives. Read the opening answer, service scope and evidence captions, then use the navigation and contact route. Confirm that the customer does not encounter an old offer, an untranslated instruction or a different destination from the English version. Test the actual journey rather than checking only whether a German page exists.
Keep service delivery separate from website language. A German explanation can help a buyer understand an English production offer. It should not imply that every actor, filming session or delivery activity is available in German. State what is standard and what can be discussed. Consistency here matters both for accurate discovery and for the conversation the receiving team will have.
Turn a language gap into a small correction plan
| Finding | First useful action | Follow-up observation |
|---|---|---|
| Wrong service detail in German | Compare the German claim with the approved offer and sources | Repeat the matched factual question after correction |
| Missing German destination | Check whether the equivalent page is published and reachable | Follow the same destination path again |
| Accurate answer without a recommendation | Review buyer fit and public evidence separately | Retest the provider question under the same conditions |
| More English than German tests | Explain sample sizes and preserve language rows | Compare a stable matched set |
| Citation without supplier inclusion | Classify source use separately | Inspect the commercial page and the answer’s selection context |
Assign the source correction and the later visibility observation as separate tasks. A completed German-page fix should be recognised as completed when the public journey is correct. The subsequent AI answer remains a separate measurement. This makes the report useful without promising that a translation edit will produce an immediate recommendation. Keep the release date beside the correction so later reviewers can interpret the sequence.
Questions about bilingual GEO tests
Should the German prompts be literal translations?
They should preserve the buyer’s intent and constraints while sounding natural. Review the meaning against the shared intent field. Literal word matching can produce awkward questions, while an overly free rewrite can introduce a different service or decision. Keep the exact wording in the record so the comparison remains inspectable.
Can we combine the languages into one dashboard score?
You can provide a combined view if its calculation and sample sizes are explicit, but retain the separate language results. The combined number should not hide a factual problem or a discovery gap in one language. Use the detailed rows to decide what to improve next.
Turn differences into specific fixes
A wrong service description points to a factual-page review. A missing translation points to a publishing issue. A single absent recommendation may require repeated observation before action. Keep a fixed core question set and label new experimental questions separately. Our AI visibility measurement guide explains the wider reporting framework; bring both language samples to a GEO review.
