PYC Notes — Phil Yarrow

Why one operator beats a stack of vendors

Wed, 15 Jul 2026 00:00:00 GMT

Growth work is usually bought as a stack: an SEO agency, a dev shop, a content team, an automation contractor, a paid-media buyer. Each is competent. Each has a contract. And the work quietly dies in the spaces between them.

The thing that stack runs on — and the thing it keeps losing — is context. Not the deliverables, but the understanding of how the search infrastructure, the automation that feeds it, and the reporting that proves it all fit together, and why each was built the way it was. That understanding is the real asset. And every time it crosses a boundary between two companies, it gets compressed — lossily.

Every handoff is a lossy compression step

Watch a single handoff up close. The strategist writes: these twelve pages must link to each other, in this exact pattern, because that structure is what tells a search engine they’re one subject and not twelve stray articles. The ticket that reaches the dev shop says: add internal links between these pages. So they do — a tidy “related articles” strip in the footer, identical on all twelve. Every link is present. The pattern, which was the entire point, is gone. Nobody catches it for a quarter, because the ticket is closed and the links are there. The what shipped; the why evaporated at the boundary.

That’s one handoff. Nobody in it was incompetent. But each boundary drops a little of the why, and the errors don’t cancel out — they compound in the same direction, away from the original intent. So four vendors isn’t four invoices. It’s three lossy compression steps between a decision and the shipped change, and the signal that arrives at the end is a degraded copy of the one that left.

What you’re actually buying: latency and fidelity

When one operator holds all of it, there are no boundaries to lose context across. The person who designs the topical map is the person who builds the automation that ships it and the person who reads the results back into the map. Nothing is translated between vendors because nothing is handed off. A decision made on a Friday ships by Monday, because there’s no queue between deciding and doing.

Strip the romance off it and it’s two engineering variables. Latency — how fast a decision becomes a shipped change. And fidelity — how much of the original intent survives the trip. A vendor stack is high-latency and lossy by construction. One brain is low-latency and near-lossless, because the deciding and the doing happen in the same head.

Why that compounds

It is also why the work keeps building instead of resetting. A prestige-marque automotive specialist we’ve run search and automation for has stayed on one retainer for five-plus years; organic is now around two-thirds of the business. Five years is long enough that a vendor stack would have churned two or three times — and each churn is a total context wipe, a new team relearning the why from documentation that never captured it. One operator kept the context intact, so year five compounded on year one instead of starting it over.

That’s the quiet mechanism under the compounding: it is an asset that only grows while someone holds its context continuously. Break the continuity and you don’t pause the compounding. You reset it.

The honest limit

This has a ceiling, and it’s worth naming. One operator doesn’t scale to a hundred-page-a-week content operation or a ten-person design sprint. When the value is raw volume of hands, a stack wins, and you should buy one.

It wins when the value is coherence — when the thing being built is a system whose parts have to agree with each other, for years, or the whole thing drifts. Compounding search authority is exactly that kind of system: the map, the automation and the reporting all have to point the same way, and a single misaligned decision costs more than an extra pair of hands would ever save.

So the trade is simple. You give up the ability to add headcount fast. You get a system where nothing is lost in translation — because there is no one to translate to — and an asset that compounds precisely because its context never leaves the room.

See what one owned system looks like over five-plus years in the field reports, or book a teardown to get the roadmap for yours.

What topical authority actually is — and the half nobody publishes

Wed, 15 Jul 2026 00:00:00 GMT

Ask most people what topical authority is and you’ll hear “cover a subject completely.” That’s half right, and the missing half is the interesting one.

Coverage is the price of entry. The complete guide walks the whole method — map the subject as entities, cover it as one linked network, let historical data accrue. All true, all necessary. But you can do every bit of it and still not be trusted, because a search engine — and increasingly a language model — isn’t only asking did they cover everything? It’s asking can I rely on what they said? Completeness gets you considered. Credibility gets you cited. They are not the same thing, and the second is the one almost nobody builds on purpose.

Here’s the counterintuitive part. The most reliable signal of credibility is the one thing nearly all SEO advice tells you to bury: honest coverage of where you’re weak.

The page that tells you not to buy

Take a glazing repair business. The obvious content plan is every page that makes the sale — misted-unit repair, broken-pane replacement, emergency call-outs. But a genuinely authoritative glazing site also publishes the page nobody in sales asks for: repair vs replace — when it isn’t worth fixing. A page that will, some of the time, tell the reader to walk away.

That page looks like a hole in the funnel. It’s the opposite. It is the page that makes every other page believable. A source that only ever says “yes, buy this” reads — to a human and to a model — like a brochure. A source that will tell you when not to buy has just demonstrated that its recommendations track reality rather than revenue. That is exactly the quality a language model is trying to detect when it decides whose answer to lift into an AI overview.

Why it works, mechanically

There’s a real mechanism under this, not just good manners. A search engine reads your site as a set of statements about entities, and it weighs how those statements behave across the whole network. A network that covers the profitable questions and the inconvenient ones — the failure modes, the cheaper alternative, the case where your own service is the wrong call — looks like expertise. A network that covers only what converts looks like marketing. And the honest page is, not coincidentally, what Google’s own information-gain thinking rewards: the page that adds the thing the other ten results left out because it was off-message.

The soft spots also make the strong pages credible by association. Anyone can claim to be the best at the thing they sell. The site that also documents the limits of the thing it sells is making a different, harder-to-fake claim — that it understands the subject well enough to know where its own offer stops. That is the claim authority is actually built on.

What it changes about how you build

It changes the content plan. The instinct is to cover the core — the pages that convert — and stop there. But the outer section, the honest and the unprofitable and the actually, in your case, don’t, is not padding around the core. It is the part that makes the core rank at all, because it is the part that proves you’re a source and not a sales page.

So cover the subject completely, yes. But leave the soft spots in — visibly. In a search landscape increasingly mediated by models trained to work out who’s trustworthy, the willingness to say “this one isn’t for you” is quietly becoming one of the hardest ranking signals to fake — because faking it means arguing against your own sale.

The full method is in the guide, and the field study shows it holding across sixteen months of real data. This note is the half the method only mentions in passing.

Why organic search is flat for a year — then hooks

Tue, 14 Jul 2026 00:00:00 GMT

There’s a chart I show almost every prospect: sixteen months of organic clicks for a glazing repair business. Flat for twelve months — noisy, unremarkable, the kind of line that makes a nervous client ask what they’re paying for. Then, around month thirteen, it breaks upward and doesn’t stop. The last quarter is roughly five times the same quarter a year earlier.

Everyone reads that shape as “be patient, it compounds.” True — and useless, because it asks the client to fund a year of pure faith. The useful thing is the reason the curve is a step and not a ramp. Once you know it, the flat year stops being an act of faith and becomes something you can actually watch move.

You’re buying rank you can’t see

Here’s what’s happening in the flat year. Your rankings are improving — measurably, the whole time. In that glazing account, average position climbed from around 46 to around 24 across the window. That’s real, monotonic progress. But position 46 and position 24 earn almost exactly the same number of clicks: roughly none. Clicks don’t accrue smoothly with rank. They’re a threshold function of it — the click-through curve is nearly flat across page five, page four, the bottom of page two, then rises steeply only as you cross onto page one.

So for a year you are accumulating rank in the region where rank doesn’t yet pay. The work is landing; it’s just landing below the waterline. Then a cluster of your pages crosses the page-two-to-page-one threshold at roughly the same time — because they were built together and matured together — and the click line doesn’t ramp, it hooks, because a whole network crossed the paying line at once.

That’s the compounding, mechanically. Not “trust me, it’s accruing.” It’s rank you can watch accruing on a chart, converting to clicks all at once when it finally crosses the line where clicks live.

Why it’s an asset and paid isn’t

This is the real difference from paid, and it’s sharper than “rent versus own.” Paid is a tap: the thousandth pound buys about the same click as the first, and the day the card declines, the traffic is zero. Rank below the fold is an asset under construction — position you’ve earned and now hold, whether or not it’s paying clicks yet. And unlike the tap, it makes the next page cheaper: a page landing on an established, credible network crosses the threshold faster than the first page did, because the engine already trusts the source. Paid gets more expensive as you scale into competition; organic gets cheaper as your authority deepens.

The honest version

Two caveats, because the fairy-tale version of this gets sold far too often.

First, it is not immune to algorithm updates. A core update can knock a well-built site back a notch — in one recent quarter, three of the accounts I watch dipped through one. The difference is that coverage-built rankings tend to recover and keep climbing where thin ones get reset. Durability is not invulnerability, and anyone who promises the latter is selling something.

Second, the asset only compounds while someone keeps holding it — keeps shipping, keeps the context intact. Stop feeding the network, or hand it between vendors who each relearn it from scratch, and the compounding doesn’t pause. It decays. “Once it’s built, it’s yours” is true only for as long as you keep building.

So the sale was never “fund a mystery for a year.” It’s “fund a leading indicator you can watch.” Position moves first; clicks are the lagging confirmation. A programme that’s working shows it in rank long before it shows it in traffic — which means the flat year isn’t flat at all, if you’re looking at the right line.

The field study shows the curve — and the position line underneath it — with sixteen months of real Search Console data, including the accounts still below the waterline.