Why I Built Authority Lab

For years I sold advice I could not source.

That is the least flattering sentence I could open with, so let us get it out of the way early. A good part of my consultancy work was telling capable people how to show up on LinkedIn. Post at nine. Open with a hook. Tell a story. Close with a question. I said those things in rooms where people were paying me to say them, and I said them well, because I had read what everyone else had read and I could deliver the whole set with a straight face.

What I could not do was answer the follow-up.

I had got into the habit of asking other people first. Where does that come from? How do you know? The answer was always some version of "it worked for me once". That answer is a coincidence with good marketing, dressed as a method.

Then I turned the question on my own material and got the same thin thing back. A handful of posts that had done well. A story I had built around them afterwards. A lot of borrowed confidence from people running the same trick one rung up the ladder. I had watched two posts with the identical correct structure land in completely different places, and I had never once been able to say why.

A rule for everything. A receipt for nothing.

So I did the annoying thing. I stopped arguing about it and went to check.

The receipt rule

The rule I ended up building a company around is small enough to fit on one line. If a claim about writing cannot produce its receipt, it does not get to be a rule. Not a heuristic, not a best practice, not a thing everybody knows. It stays an opinion, held by a named person, until somebody measures it.

The Authority Index is what that rule looks like once you build it. It began as a database of real LinkedIn posts from strong B2B creators, each one read and classified across more than a hundred fields: intent, format, proof type, opening line, cognitive load, how it closes, how far it develops. The analysis cohort settled at 29,599 posts from 53 creators, and every claim on this blog is one cut of it.

One decision inside that matters more than all the others. Performance is measured against the writer, never against the crowd. A post with two hundred reactions means nothing on its own. It means something once you know it did three times what that same person usually does. Compare people to each other and you are mostly measuring audience size, which nobody can rent overnight. Compare a person to themselves and you are measuring the writing.

The first time it caught me

A few weeks in, the database told me I had been wrong for a month.

My scoring ranked posts mostly by comments, because I value comments. I like a thread. I like being argued with. What I had actually done was write my own taste into the machine, and quietly punish every post that gets saved and reshared instead of debated, which is most good visual work.

The day I wired in the fuller picture, the average post score moved from 1,833 to 3,106. Overnight, on the same posts. Nothing about the corpus had changed. Only my measurement had, and my measurement had been carrying my preferences around like contraband.

That incident is the whole product in miniature. The thing exists to catch me while I am believing something false, in public, with money on it. Confirmation was never the job.

What it cost

Money, mostly. Some of it stupidly.

One night a collection run went wrong. I had capped it inside the tool that pulls the data, the tool ignored the cap, and it started hoovering up an entire comment history one page at a time. I killed it when I saw it and wrote the damage into my log as $17.90, which felt survivable, and I moved on with my evening.

Four days later I opened the actual invoice. The run had cost $27.99. Killing it had stopped the data arriving and had not stopped the meter, so my careful log was simply wrong, and one rejected tool had swallowed most of that month's budget on a mistake I had already filed as handled.

The rule that came out of it runs before every collection job now: set the spending ceiling at the platform level, never at the tool's, because the tool has no reason to care. The lesson underneath is the same one the scoring bug taught. My notes are not evidence. The invoice is evidence. Everything I write down about my own work is a claim waiting for its receipt.

The five quiet months

The part nobody puts in a launch post is the silence.

I stopped publishing for about five months while I built this. No newsletter, no clever thread about what I was learning along the way, nothing to show anyone. Just collection running overnight, classifications failing, two benchmark tables refusing to agree with each other, and a growing suspicion that I had picked an extremely hard way to answer a simple question.

When I came back, the first account I ran the finished thing against was my own. Not a client's, not a flattering case study. Mine. It scored me 71.3 out of 100, showed me which pillar I was weakest on, and I published the number that week.

I would like to report that I took it well. I took it accurately, which is different, and better.

What Authority Lab actually is

It is a thinking tool. It reads what you have published, scores it against the Index, and shows you which part of your writing is costing you the most, with the reason attached and the measurement it came from. Then it stops, and leaves the writing to you.

It will not write your posts. There are plenty of tools that will, and they all produce the same beige middle, because they are all averaging the same internet. The writing is the one part that has to sound like a person who has actually been somewhere. A machine that hands you finished sentences takes the only bit that was ever yours.

So it does the other job instead. It hands you the receipt and gets out of the way.

What I do differently now

I hold my own claims to the standard I hold everyone else's, which is slower and considerably less fun. Before I say a thing about writing, I go looking for the receipt, and when there is not one I say "I think" out loud, like an adult. Twice now that habit has forced me to publish a correction to something I had already told thousands of people. Both times it cost me an afternoon and bought back the only thing this whole project runs on.

The rest of this blog is me reporting what is in the Index, including the parts that contradict what you were told, and including the parts that contradicted me.

If a claim cannot show a receipt, it is a preference. Yours is allowed to be one. A rule is not.