The research record is the starting point
Published ranking order comes from the study's underlying research data. The article should identify the measures available for that comparison, such as recommendation coverage and average position. A position in the table is specific to that research question and date.
The publishing team explains the evidence. It does not select a preferred winner or change recorded AI responses. Category comparisons use a read-only aggregation of current published study evidence when no active ranking snapshot exists.
Read coverage with its denominator
A count such as six recommendations out of seven usable responses describes the breadth of support in that response set. A percentage without its denominator can obscure a small sample. Coverage does not measure market share, customer satisfaction, service reliability, or the probability of getting a booking.
When an average position is supplied, read it alongside the number of responses in which the provider appeared. A high position in one answer and broad support across many answers tell different stories.
A shortlist needs additional context
An AI system can favor a familiar brand, repeat outdated information, or omit a suitable provider. Similar answers can also reflect shared source material rather than independent agreement. Use the provider discussion, source links, and limitations alongside the table.
Providers with no appearance in a study have not necessarily failed an assessment. Their absence only tells you about the responses collected for that study.