Figures in this case study are illustrative and drawn from an anonymised engagement. They are shared with permission in aggregate form.

This engagement is the counterexample to the previous one. Nothing needed pruning. The content was excellent — genuinely the best reference material in a crowded category, written by the engineers who had built the thing. It ranked respectably. And in eighteen months it had earned 11 referring domains, appeared in zero AI answers about its own subject, and generated a branded search volume indistinguishable from noise.

The constraint was not the content. It was that nothing about the company established it as a source worth citing.

What “publisher standing” actually means

Three things were missing, and they compound.

No verifiable authors. Every post was bylined “The Team”. There was no person attached to any claim, no author page, no profile anywhere that a system could resolve to an identity with a history in this field. A byline that cannot be resolved is not a byline.

No original data. Every piece explained something. None of them knew something that was not already knowable. There was no reason for anyone else in the category to link to them, because there was nothing in them that could not be sourced elsewhere.

No entity footprint. The company existed in its own marketing and almost nowhere else. No structured data connecting it to its category, no consistent presence across the handful of databases and directories that a knowledge graph actually reads, no sameAs pointing anywhere real.

What we shipped

One benchmark, run properly. We spent seven weeks designing and running a reproducible benchmark of the eleven tools in the category, including the client’s own, on a workload the category argues about constantly. We published the methodology, the raw data, and the harness — including the two benchmarks where a competitor beat the client.

That last part is the part people flinch at, and it is the part that works. Publishing a result that is inconvenient for you is the cheapest credibility signal available, and it is the reason the study got picked up by people who had no incentive to be kind.

Author entities, built once. Three engineers were promoted to named authors with real biographies, author pages, and sameAs pointing at profiles they actually maintain — conference talks, a package registry, a public repository. This is a two-day job that most companies never do, and it converts every future post from an anonymous artefact into something attached to a person with a record.

Distribution to five places, not five hundred. We did not run an outreach campaign. We identified the five newsletters and two forums where this specific audience argues about this specific problem, and we made sure the study reached the people who write them, with the raw data attached and no embargo games.

What moved, and when

Nothing moved for eleven weeks. The study published in week eight of the engagement; the first meaningful citation appeared in week nineteen. That lag is normal and it is the single hardest thing to hold a client through, which is why we say the number out loud in the first call rather than discovering it together in month four.

By month six: 47 referring domains, branded search up 6.2x off a tiny base, and — the metric the client cared about most — their name appearing in AI-generated answers to the category’s central question, sourced to the study.

What we would do differently

We should have built the author entities in week one rather than week nine. The study went out under a company byline because the author pages were not ready, which meant the single highest-authority artefact the company had ever published did nothing to establish the individuals behind it. We reissued it with proper attribution later, and it did not have the same effect. Entity work is cheap and slow-acting: do it before you need it.