How We Built the Index

People ask how the Authority Index actually works, usually because they want to know whether to trust it. Fair. A database is only as good as the discipline that built it. So here is the whole pipeline, no hand-waving. Four stages: collect, classify, score, benchmark.

1. Collect

We pull public posts from a defined set of 53 top B2B creators, the people whose content consistently performs and who publish enough to give us a real sample of their range. Only public posts. No private analytics, no follower data beyond what is publicly visible, no DMs. If you could not see it by visiting their profile, it is not in the Index. The corpus now sits at 29,599 posts, spanning years of publishing.

2. Classify

This is where a pile of text becomes an instrument. Every post is tagged across 100+ fields: intent (what the post is for), format, hook type, CTA type, cognitive load (how dense the writing is), word count, and structure. We use a language model to classify at scale, but every field is a constrained choice from a defined taxonomy, not a free-text opinion. The model is not asked "is this good?" It is asked "which of these six intents is this?" That constraint is what keeps the classifications consistent enough to count.

3. Score

A post's raw engagement is close to useless on its own, so we do not lead with it. Every post gets a performance band relative to the creator who published it. Top-of-range for that person is "Exceptional." This is the single most important step in the build, because it lets a 5,000-follower account and a 500,000-follower account live in the same dataset and be honestly comparable.

We also weight engagement properly. Reactions, comments, reposts, and saves each carry real, separate weight, so a save-heavy visual post is not penalised for getting fewer comments.

4. Benchmark

Once every post is classified and scored, the aggregates become possible: average engagement per intent, per format, per word-count band, per cadence, the percentile distribution of engagement rates, the fingerprint of exceptional posts. These refresh continuously as the corpus grows, so the numbers you see are current, not a snapshot from last year.

The honest limits

No dataset is neutral. Ours is B2B creators, so it runs hotter than LinkedIn as a whole. It is public engagement, not impressions or conversions. And classification, however constrained, is still interpretation at the edges. We say all of this out loud on purpose. A tool that hides its limitations is asking you to trust it blindly. This one shows you the seams.