Omni Eclipse
Omni Eclipse
Book Your Audit →Speak with Team

Methodology

A ranking is only worth citing if you can check how it was made, including the parts of it that are weak. This is that page.

What is measured

Visibility is the share of AI answers in which a brand is named. Not whether the brand's website was linked, and not whether it ranked in Google. Named, in the answer text, by an assistant responding to a question a buyer would actually ask.

That distinction matters more than it sounds. Being used as a source and being recommended are different outcomes, and a brand can do a lot of one without the other. Omni Eclipse is a case in point: our pages are cited as sources in 48 of these answers and we are named as a recommendation in 8.

The prompts

Each index runs a fixed set of buyer prompts written the way people ask them, not the way keywords are written. For the agency index that is 39 prompts covering how to choose an agency, agency versus tool versus doing it in-house, what to look for and what to avoid when hiring, cost, and the same questions by country.

Prompts are fixed for the life of an index period. Changing the questions mid-period would change the result without changing anything real, so new prompts start a new period.

The engines

Every prompt is run against ChatGPT, Google AI Overviews, Google AI Mode, Perplexity and Gemini. Answers are collected on a rolling schedule rather than in one sitting, because these systems change their answers day to day and a single snapshot would be a measurement of one afternoon.

What counts as a mention

A brand is counted when it is named in the answer body. Aliases count: a brand with more than one common name is matched on all of them. A domain appearing only in a citation footnote does not count as a mention, because that is a source, and sources are counted separately.

Counting is per answer, not per occurrence. A brand named four times in one answer counts once, so a verbose answer cannot inflate a score.

Where this method is weak

Answer engines personalise and vary. Two identical prompts minutes apart can return different brands, which is why the index reports a share across hundreds of answers rather than a rank from a single run. Treat a two-point difference between adjacent brands as noise.

The brand set is not the whole market. It covers the brands we track, which is a large sample of the category but not a census, so absence from the index means we did not track that brand, not that it never appears.

Prompt selection is a judgement. We chose questions a buyer would ask, and a different reasonable set of questions would produce somewhat different numbers. The prompt list is published so anyone can disagree with it specifically.

It measures being named, which is not the same as being chosen. No index can tell you which agency does better work.

Corrections

If you are listed and believe a figure is wrong, tell us and we will check it. Corrections are logged below with the date and what changed, rather than being edited in silently. An index that quietly changes its own history is not worth citing.

No corrections have been issued yet. This is the first published period.

Published periods