The Citation Flywheel Starts at Zero.

Publishing more content is a treadmill: the day you stop running, you stop moving. A citation flywheel is different. Original data gets lifted, each lift produces a brand mention, mentions raise your entity weight, and heavier entities get retrieved more, which makes the next citation cheaper than the last. Turns compound. Treadmills do not.

The catch is the physics. Flywheels are heaviest at zero, and zero is where mine currently sits. I will show you the whole loop anyway, because the loop is the strategy.

Why is content volume a treadmill?

Because every article you publish competes with every article everyone publishes, and generative tools have made everyone's marginal article free.

When supply is infinite, the unit of competition stops being the article and becomes the thing an article cannot fake: a figure that exists nowhere else. An answer engine assembling a response does not need your fourth take on a topic. It needs a number it can attribute, and average takes are now free to generate without you.

You cannot out-publish the internet. You can out-originate it.

How does the flywheel actually turn?

Five bearings, in sequence. Jim Collins described the flywheel effect (opens in a new tab) as consistent pushes on one heavy wheel compounding into momentum. This is that, for being cited.

Originate. Produce data only you can produce. Mine comes from Orbyt: the Orbyt Intelligence compensation dataset, 15 million data points across 3,445 roles and 81 cities, generated by a product I built and run solo.

Address. Give the data a permanent, citable home. Mine is deposited on Zenodo under DOI 10.5281/zenodo.19653006 (opens in a new tab), CC BY 4.0, with my ORCID attached. A DOI outlives any URL and makes me the origin of those figures, not one more page repeating them.

Resolve. Make the who behind the data unambiguous. My identity anchors to Wikidata (opens in a new tab) and ORCID 0009-0005-2615-3624, registries I do not control, so a machine deciding whether to cite me can first decide who I am.

Package. Serve every claim as a capsule that survives being lifted without its page. The full implementation is Built to Be Quoted; the data infrastructure underneath it is its own discipline.

Instrument. Point a probe at the engines and count. Without this step the flywheel is a belief, and beliefs do not compound.

Where does the loop close?

Off-site, which is the part most on-site optimization ignores.

A citation is not just traffic. It is a brand mention placed inside the corpus and the context future models learn from. The strongest external evidence I trust finds brand mentions correlating with AI visibility roughly three times more strongly than backlinks. Mentions raise entity weight. Entity weight raises retrieval. Retrieval raises the odds of the next lift.

That is the compounding edge: every turn of the wheel makes the next turn cheaper. It is also why the wheel is so heavy at the start. An unmentioned entity does not get retrieved, and an unretrieved entity cannot earn the mention. The first turns are pushed by hand: the dataset shared where practitioners already look, the figure quoted in someone else's argument, the product whose own users repeat its numbers.

What does turn zero look like?

Like nothing is working. That part deserves to be said plainly.

My citation scoreboard reads zero. Every run, published. The dataset exists, the DOI resolves, the identity graph is clean, the capsules are liftable, and no engine has cited me yet. A flywheel on turn zero is indistinguishable from a wheel that will never spin, and only the instrument can tell you which one you built.

So why keep pushing? Because the investment survives either answer. The dataset is real product surface. The identity work compounds across every channel. The instrument tells the truth weekly, refreshed by the same pipeline that keeps my figures honest. And the alternative, the treadmill, is guaranteed to stop the day I stop.

A flywheel you stopped pushing too early is indistinguishable from one that never worked. That is the bet, and I am taking it in public.

What to do Next

Find your one dataset. Every operating company generates data nobody else has: benchmarks, distributions, failure rates, costs. Pick the single set a journalist, an analyst, or an engine would actually want.

Give it an address stronger than a blog post. A DOI costs nothing on Zenodo. A dated, versioned, licensed dataset with a named author beats a landing page for citability, permanently.

Then instrument before you evangelize. Decide what counts as a citation, build the probe, record zero honestly. The companies that win this loop will be the ones still pushing on turn four, and the only way to reach turn four is to know, with numbers, that you are past turn three.

Treadmills reward the sprint. Flywheels reward the grip. Originate, address, resolve, package, instrument, and keep your hand on the wheel.

Related reading:

• I Built the Scoreboard. It Says Zero. the instrument this flywheel is judged by
• Built to Be Quoted. the passage engineering on each bearing
• Every Fable Has a Moral. Mine Has Data. why the data is the moral
• Prose Is Paint. Data Is Load-Bearing. the data layer the wheel spins on