Why an Israeli GEO Baseline Needs Repeated Questions: Findings from Nir Levi’s Pilot

An AI answer is easy to capture and easy to overinterpret. A company sees its name once and wants to know whether its visibility has improved. Nir Levi’s Israeli GEO pilot addressed the measurement problem by asking the same six buying questions in Hebrew and English, then repeating each wording in a fresh Perplexity conversation. The resulting 24-answer record allows a direct comparison of same-language repeats as well as language pairs.

What the pilot recorded

The study contains 12 same-language repeat comparisons: six English and six Hebrew. Exported source-host sets were identical in 0 of those 12 comparisons. Mean source-host overlap between repeats was 47.2%. The paired-language overlap, calculated separately, averaged 34.7%. All observations were collected on 11 October 2026 using Perplexity Search, the GPT-6.1 Sol model label and fresh incognito conversations. The two measures describe this saved panel under those conditions.

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.

A repeat should preserve the original question

Changing a prompt between checks can change the buying task. Adding a brand name, asking for a different number of options or introducing a new geographic condition creates a different observation. The pilot therefore saved the wording before collection and retained every completed answer. Its two rounds used the same questions and reversed the within-pair language order. This provides a clear record of what was held constant and what was compared.

Look at several outcomes separately

A repeat review can inspect whether the tracked brand remains present, whether an official-domain link appears in the export and whether the wider source set changes. Those outcomes describe different parts of the answer. A company with access to its own commercial records can separately examine referred visits and received enquiries. Combining everything into a single score too early makes it harder to identify which part of the process actually changed.

Extend the record before claiming a trend

The completed pilot is a same-day snapshot. Its next useful extension is the same question panel on later dates, with the product surface and settings recorded again. When a business changes a page, keep that intervention in a separate change log. This gives an analyst the evidence needed to investigate a possible relationship without erasing other explanations. Model updates, source updates and repeat variability should remain visible in the record rather than being silently attributed to the campaign.

Nir Levi’s perspective

Qualified enquiries and revenue matter most. Mentions and citations show that we are moving in the right direction, but clients want business results, not just visibility. Alongside that, I want to show the client the size of its market and its share, so the numbers have context and proportion.

Read the original research

The complete study, paired prompts, evidence downloads and Nir Levi’s commentary are available at https://nirlevi.com/en/research/israeli-geo-hebrew-english/

Put the finding to work

Nir Levi’s GEO consulting can use this structure to turn a one-off screenshot into an ongoing review process. Start with a defined business question, retain the evidence and agree on the next measurement before interpreting a change as progress.

Explore GEO and AEO consulting in Israel with Nir Levi at nirlevi.com and bring a buyer question to your next review.