A brand appearing in an AI answer is only the beginning of the source review. The surrounding explanation may rely on different pages when the buyer asks in Hebrew instead of English. Nir Levi’s Israeli GEO pilot examined this distinction using six technology-buying scenarios, two languages and two rounds of Perplexity answers. The prompts placed the buyer in Israel and asked for trade-offs and sources without naming the tracked vendor.
Across 12 language-pair comparisons, 0 had identical exported source-host sets. Mean overlap was 34.7%. The preselected category brand appeared in 12 of 12 English answers and 12 of 12 Hebrew answers. These counts come from six fixed questions, each asked twice per language on 11 October 2026 in Perplexity Search using the GPT-6.1 Sol model label.
Source-host overlap is the number of shared hostnames divided by the number of distinct hostnames across the two answer exports. Perplexity’s Copy output is the measured record; grouped on-screen sources may contain additional links.
The study compares sets of unique source hostnames in the copied answers. Shared hosts are divided by all distinct hosts in each language pair. This gives a descriptive overlap measure: two answers can name the same brand while their supporting exported sources have limited overlap. It does not measure the quality of a product or the percentage of all buyers who encounter it. Keeping the calculation tied to its actual denominator makes the finding usable.
A translation review usually focuses on the business’s own pages. A source review adds another layer: which outside comparisons, documentation pages or local explanations appear around the business in an answer? In the first website-builder pair, for example, both answers discussed Wix while their exported source sets differed. This directs the next investigation toward specific source pages and buyer requirements rather than a blanket decision to translate every page.
Choose a question where the source map reveals a clear information need. Ask a subject specialist to identify which parts of the answer are accurate, incomplete or outdated. Then assign an owned-page improvement with supporting documentation and a review date. Preserve the original answer and compare a later observation under the same recorded settings. The team now has an explicit information task and a repeatable way to inspect what happened next.
What surprised me is that the brands stayed the same while the sources changed so much. For Israeli businesses, it means translating content is not enough. We need to research Hebrew and English separately, especially given the Hebrew market’s much smaller size compared with English.
The complete study, paired prompts, evidence downloads and Nir Levi’s commentary are available at https://nirlevi.com/en/research/israeli-geo-hebrew-english/
For businesses targeting Israel, Nir Levi’s research provides a practical argument for examining Hebrew and English separately. The source map, rather than a translated keyword list alone, can help define the next GEO content brief.
Explore GEO and AEO consulting in Israel with Nir Levi at nirlevi.com and bring a buyer question to your next review.
