Insights
What we are learning from sampling six engines every day, including the results that complicate our own product.
Methodology
September 2026
Asking the same question five times in one hour returned three different brand sets. Here is the run, the spread we measured, and why a single number would have been the wrong output.
Engines
September 2026
One is a summary stitched onto search results, the other is a conversational assistant. Brands routinely do well in one and badly in the other, which is why blending them would flatten the only useful signal.
Engines
September 2026
A data provider that returns Bing web results does not return the Copilot answer. The distinction is easy to miss and it quietly invalidates a whole row of a tracker.
Influence
August 2026
Across our sampling, the things that changed answers were rarely on-brand pages. Third-party sources the engines already trust did most of the work.
Research
August 2026
How often each engine cites at all, how many sources it shows, and how much that has swung over successive model updates.
Methodology
July 2026
Personalised answers are what your buyer sees, but they are unmeasurable. Why an incognito-equivalent, fixed-market sample is the only defensible trend line, and what it cannot tell you.
Want the underlying runs? The methodology explains how each sample is taken, and we will share the data with researchers on request.
