Same brands, different sources
Six buying questions in Hebrew and English, with the findings, evidence and Nir Levi’s interpretation for Israeli businesses.
Read the guide and evidence ↗Tests & working tools
What supports the recommendation?
When evaluating a recommendation, I want to understand its evidence and how to test it. This collection brings together tests, examples and tools from my site. Each serves a different purpose: an experiment on this website, a worked decision or a worksheet for your own observations.
Documented test on this website
Can an AI agent complete a form?
What happens when a form receives text that is too short, and can it recover? The article documents a test on a local copy of this website, the defect and its fix. Codex performed the checks; this was not a client project.
Published: 10 September 2026
Read the guide and evidence ↗Worked example with search data
Which page deserves the next investment?
A B2B keyword research example using tool data, search result comparisons and a way to select a team task. It demonstrates a decision process without claiming an increase in client traffic or inquiries.
Published: 17 September 2026
Read the guide and evidence ↗Tool for your own observations
Record mentions and sources
Build a question set and retain answers, sources and repeated observations. The worksheet separates mentions, citations and recommendations. It is a working template, not a pre-collected research dataset.
Read the guide and evidence ↗These are resources published on my own website. They are not independent endorsements or client project results. Each guide records its methods, sources and writing assistance.
Before using a finding
Define the question
Establish what was tested, on which site or market and under which conditions. A screenshot alone is not enough to support a conclusion.
Examine the measurement
Check who ran the test, when, with which tool and how many repeats. In AI answers, distinguish a mention, a source citation and a recommendation.
Decide what applies elsewhere
A result from one task can suggest something to investigate on your site. It does not establish that the same change will produce the same result elsewhere.
