AI search, SEO & online discovery
Practical coverage from Dubai

Saved
Measurement

How to Compare English and Arabic Visibility Scores Fairly

Summarize with AI
Language editing and reference work; illustrative photograph.
Photo by Pixabay / PexelsIllustrative photography
In this article 6 sections

A bilingual visibility report can show that a Dubai business appeared more often in English. Before you decide which content to improve, check what the report compared. The English questions might ask for a shop in a particular district while the Arabic questions ask for delivery across Dubai. A reviewer might recognize the business under its English brand name and miss the same name in Arabic. A chain can appear in both answers while only one sends the customer to the requested branch.

A fair comparison starts with a shared customer task and a scoring rule that recognizes equivalent evidence in either language. Keep the results for each language, show the overlap in eligible question pairs, and publish branch and source outcomes separately when the decision depends on them. The resulting report can distinguish a language difference worth investigating from a difference in what was asked or how it was scored.

Define the customer task before drafting the two questions

Write a brief for each pair that specifies what the customer needs to do, the location and the constraints. For example, a customer wants to collect an online order from a specific Dubai branch on a weekday. That is different from asking which Dubai branches offer pickup. The first needs a branch-specific answer. The second can produce a useful list. Pairing those two questions would mix language with task.

Preserve the constraints that matter to eligibility. A neighborhood, budget, time window, service mode or requirement for an official link can change what qualifies. Compare the intended response as well as the wording. If English asks where to book and Arabic asks whether booking is possible, a reply that names a branch without a booking link may satisfy one question and fall short on the other. Resolve differences before collecting the primary comparison.

Use natural expressions for the intended audience in both languages. A literal rendering can preserve every noun while turning an ordinary customer question into formal copy. If you are testing a particular colloquial register or mixed-script query style, make that a separate condition. Record what the reviewer changed and why. A bilingual colleague can independently describe what each question asks, then compare that description with the shared brief. This adapts the idea of translation, review, adjudication, pretesting and documentation from survey design to a small business audit. European Values Study translation method

When the business has actual questions from each audience, preserve them in a separate view. That view can help you decide what each audience needs from your site. Its English and Arabic questions may have different mixes of tasks. A difference in scores there describes each sampled question set. An intent-matched panel serves the narrower comparison in which both languages cover the same tasks.

Record prompt language and answer language separately

Give each pair an ID and each submission a run ID. Record the prompt as submitted, its language and register, the interface language, the surface, capture window, market and session conditions. Record the language or languages of the actual answer in another field. An Arabic prompt may produce Arabic prose with English brand names, or a largely English answer. The primary comparison can remain grouped by the requested prompt language, while a separate answer-language check shows whether the customer received a usable response in the language they asked in.

Keep the market and surface stable when testing prompt language. If the English runs used one interface locale and Arabic runs used another, that may represent the real customer journeys you want to compare. Name the comparison accordingly. The run conditions then change together, so the result cannot isolate wording alone. Where practical, alternate which language goes first across pairs and keep runs close in time. Use separate sessions if earlier questions could shape later replies. Document the rule so next month's team can reproduce it.

A timed capture should preserve the reply and any sources the metric requires. If the answer is available but the source panel is missing, score the mention if there is enough evidence and mark citation eligibility separately. Use the same rules in both languages. If the reviewer must understand an Arabic passage to decide whether it recommends or merely names a business, route ambiguous passages to a qualified reviewer and keep the original wording.

Recognize the same brand without collapsing distinct branches

Build the scoring key around stable entity IDs. Record confirmed public names in both scripts, and keep branch IDs separate. Compare a candidate name with the rest of the evidence in the answer, including address, district, official link or surrounding description. The parent retailer can count for a brand-mention metric while a nearby branch counts for a different destination. Decide which outcome matches the customer's task before looking at the scores.

The official Carrefour UAE post below illustrates Arabic and Latin branding from the same publisher. A verified public example like that can inform the alias register. The register still needs separate evidence when a specific branch is in scope. A logo or short brand name alone cannot identify which store a customer should visit.

Save the original answer text and the match rule used. Unicode normalization can make canonically equivalent character sequences comparable. It does not translate a name, resolve a transliteration, or prove that similarly spelled businesses are the same. Additional search rules such as optional diacritic handling should generate candidates for review when they can affect identity. Store confirmed matches and unresolved candidates separately so a conservative rule does not silently turn uncertain Arabic names into absences. Unicode Standard Annex 15

Report the action a reader could take. For a location query, the most helpful split may be brand recognized, requested branch correctly identified, and usable official branch link. A correct branch result in this example requires a correct brand match. On another audit, you might independently score whether any source was cited or whether an answer explained the right service. Each metric needs a definition, an eligible population, and its own evidence.

Read the paired result and the language totals together

This constructed example follows eight English–Arabic pairs for the same fictitious business. Each pair asks for a specific branch under a shared task brief. For the two pairs with an incomplete capture, the other language has a scorable brand mention. The six remaining pairs support direct comparisons.

Measure English Arabic
Planned questions 8 8
Scorable captures for brand mention 7 7
Confirmed brand mentions among scorable captures 5/7 4/7
Confirmed brand mentions on the six shared scorable pairs 4/6 3/6
Correct branch on those same six pairs 3/6 2/6

Coverage is seven of eight for both languages, yet the missing question differs. The English conditional rate includes one positive on a pair without a scorable Arabic capture. The Arabic rate also includes one positive on a different pair without a scorable English capture. Keep those positives in their language totals and publish the common set separately.

On the six shared pairs, two mention the brand in both languages, two only in English, one only in Arabic, and one in neither. The matched difference comes from the three pairs with different outcomes. This lets the reader see the direction and scale of the observation without inferring that English performs better across Dubai. The branch result asks a stricter question. One confirmed mention per language on the common set did not identify the requested branch. Inspect the underlying answer before attributing the gap to the availability of branch information online.

When comparing months, report changes in the paired panel along with changes in the result. An English-only positive could become a positive in both languages, or it could leave the common set because of a missing capture. Those events can move the paired difference in the same direction while telling very different stories. Save a pair-level status for both months so you can account for additions, losses and changes on the overlapping set. If a task or prompt changed substantially, give it a new version and keep it out of the unchanged-panel trend.

Interpret differences in sources at the right level

A published Lunasol example measured how many English baseline sources reappeared after questions were asked in Arabic. Across four Dubai sectors on one day in Google's AI Mode, fifteen of thirty-one baseline source placements were retained under the guide's sector-by-sector counting method. That measures retention relative to the English baseline. It gives no brand-mention rate and does not supply the total number of Arabic sources from that figure alone. The same reporting guide records that repeated prompts also changed the cited source list. Both points matter when interpreting a single bilingual comparison. AI Visibility monthly reporting guide

If you compare source lists, name the unit and the direction. English baseline retention asks which English citations also appeared in Arabic. Arabic baseline retention asks the reverse. Those rates can differ when the lists have different lengths. If the goal is to find a gap in first-party support, a more useful field may be whether an official page was cited for the relevant branch. That page might be in either language. Record page language separately from source ownership, and check the page itself before classifying it from the URL.

The original AI Visibility English and Arabic guide covers what UAE brands should test, including matched questions and wording variants. The scoring record here picks up once the observations are captured. Its job is to show how each eligible pair contributes to the language totals and how the result changes when you tighten the outcome from brand mention to branch accuracy. English and Arabic AI search guide

Write the report around the decision it supports

For a team improving branch pages, start with the tasks where one language repeatedly identifies the brand but misses the right location. Look at the observed passages and verify the public branch evidence. A gap can reflect ambiguous entity names, a source page that serves several branches, differences in wording, or answer variability. If you find that the existing Arabic branch page uses an ambiguous location label, you have a specific item for the content owner to review. Record the finding separately from any explanation about what caused the AI answer.

For a team planning a broader multilingual content program, group the pairs by customer task before summarizing the results. Ten easy brand questions can overwhelm a small number of consequential booking questions in an unweighted overall count. Show the group counts and document weights if a business decision calls for a weighted view. Use the same weights across languages on the same comparison population. If a group lacks enough shared observations, give the available counts and identify the evidence gap.

Bilingual visibility scoring record

For each pair, retain the task brief, approved wording and reviewer decision, prompt and answer languages, run conditions, evidence links, eligibility by metric, and the brand and branch matches. In the monthly summary, show planned and scorable counts by language, common pair counts, outcomes on both sides, and the review action those observations support. Name the question set and date range in the caption so the next reader can tell exactly what the reported difference covers.

AI Visibility is published by Lunasol, which offers GEO and AEO services. The eight-pair example is constructed for this guide. The published source comparison is identified separately. About AI Visibility

You’ve reached the end of this article. Read as plain text

Search the blog

Summarize with AI

Choose your assistant. Copy the article and summary request, then paste it into a new chat.

The full article is included, so your assistant can read it without opening this website. Nothing is submitted automatically.

View or copy the summary requestOpen assistant

Summaries are generated by your chosen service and may miss details. Its account requirements and terms apply.