Define what language means in this report
Write a report contract before exporting data: the property, date range, search type, dimensions and the meaning of a language group. A group based on Arabic pages measures performance of those pages. It does not directly measure the language spoken by each searcher. Country and query wording are different observations and should not silently replace page language.
Maintain a reviewed address-to-language mapping, with a separate bucket for unknown or shared destinations. URL prefixes can help classify pages when the site follows them consistently, but verify exceptions. Pair translated resources with a content identifier so you can compare their editions without pretending they are the same URL.
Use one row definition and disjoint groups
Choose a consistent row definition, such as date and page, and retain any other dimensions used in the export. Two rows can be added only when they represent distinct observations under the same reporting definition. Record the export scope and time so overlapping downloads are identifiable before they enter the final sheet.
Assign each page row to one language group. If a shared selector is included in both Arabic and English groups, their combined total duplicates that row. Keep such a page in a shared group or apply one documented classification. Do not append an all-pages export beneath its Arabic subset and then sum both as new traffic.
Keep aggregation and ratios explicit
Search Console distinguishes property-level and page-level aggregation. Its documentation explains that chart and table totals can differ for this reason. Keep the official property total alongside the page-language breakdown, with their definitions visible; do not force them to agree by removing unexplained rows or treating every difference as a collection error.
Calculate a combined click-through rate from the combined clicks divided by the combined impressions for the chosen consistent dataset. Do not average the Arabic and English percentages without their denominators. Avoid combining average positions from incompatible exports. Label preliminary or incomplete periods and compare equivalent scopes rather than presenting the newest partial period as a finished result.
Illustrative example: two page groups
Suppose a fictional page-level dataset contains 30 clicks and 600 impressions for Arabic pages and 20 clicks and 400 impressions for English pages. The groups do not overlap and use the same period and dimensions. Their combined page-level figures are 50 clicks and 1,000 impressions, giving a 5 percent click-through rate. These are invented teaching numbers, not a customer report.
A second file containing the same Arabic rows must not add another 30 clicks. Check its scope and row keys, then retain one authoritative copy according to your documented import rule. If a property-level chart shows a different impression total, preserve that separate measurement and investigate aggregation before claiming the language breakdown is wrong.
Publish a reproducible comparison
Before sharing, verify that every included row has one classification, the exports use the same filters and duplicate imports have been resolved. Save the mapping, scope and calculation with the report. Show Arabic, English and any shared or unknown group; hiding uncategorized pages makes the coverage look more complete than it is.
Explain what changed without attributing every movement to translation quality. Different page availability, query demand and reporting periods can affect comparisons. Search clicks are not unique people, purchases or confirmed indexing of every page. Keep the bilingual content inventory beside performance data so the next investigation starts with actual resources and consistent measurement.