Every Channel Has a Golden Age

In October 1994, AT&T ran the first banner ad on HotWired. It said "Have you ever clicked your mouse right here? You will." Its click-through rate was 44%. Today, a display banner earns about 0.1%, a decline of roughly a thousandfold, which makes banner ads the most precisely documented collapse in marketing history.

This isn't a new observation. Andrew Chen named it the Law of Shitty Clickthroughs back in 2012; Gabriel Weinberg's Traction said "over time, all marketing channels become saturated" in 2015. What I want to add is the discipline the observation demands and almost nobody practises: every channel judgement you hold has a date on it, and most founders are running numbers from a year that already ended.

The same curve, five channels

Google search ads launched self-serve in 2002 with a five-cent minimum bid. By 2005 the average click cost 38 cents. In 2025 it costs $5.26, roughly fourteen times the 2005 price for the same real estate.

Facebook repriced faster. CPMs averaged 21 cents in early 2011; by early 2012 they'd already risen 41% year on year, and the 2025 median sits around $13.48. Forty-five times the 2011 price. Note the shape of that curve: the repricing began the moment the channel publicly proved out — years before it matured. The proof and the golden age's end are the same event.

Cold email's version is the one this book keeps returning to. A 12-million-email study in 2019 measured 8.5% reply rates on outreach. The strict-cold average today is 0.45%, and even the friendliest platform averages fell from 5.1% to 3.4% in the last two years alone.

Search itself is closing. In 2019, half of Google searches ended without a click to the open web; by early 2026 it's 68%, and of every thousand searches only 276 now land on an independent site. AI Overviews cut click-through on the top organic position by a third in early 2025; by December the measured cut was 58%. Nine months, near-doubled squeeze.

And LinkedIn organic, the channel every founder was told to bet on in 2021: measured reach fell 50% year on year by February 2025, then another third across 2025, declining for 98% of users. The golden age lasted about three years, which is roughly standard.

Why windows close

Three forces, always the same. Competition arrives and bids the price toward the channel's true value, because arbitrage advertises itself. The platform, once it owns the audience, takes its margin back, throttling the free reach it used as bait. And users adapt: the pattern that earned 44% in 1994 earns 0.1% now not because banners got worse but because humans built banner blindness. Saturation isn't a moral failure of marketers. It's the system working.

Which produces the rule founders hate: by the time a channel has a conference, a certification and a "definitive guide", its golden age is over. The public proof is the closing bell.

Spotting a window while it's open

Openings have signatures, and they're mostly price signals.

A price gap against a comparable audience: X ads reaching US tech buyers at $2–7 CPM while LinkedIn charges $34 for a heavy overlap of the same people, with X's click prices actually falling 9% last year because advertisers fled faster than users did.

Fresh inventory with clumsy tooling: ChatGPT's ad CPMs reportedly fell from around $60 to $25 within months of launch as inventory opened, while the self-serve manager is still in beta and most advertisers haven't shown up. Awkward tooling is a feature; it's the queue-length filter.

Formats the incumbents dismiss: every golden age on the list above was, at its start, something serious marketers considered beneath them. Banner ads were a gimmick, AdWords was for people who couldn't afford real media, Facebook was for students, cold email was spam, LinkedIn posting was cringe. The dismissal is the moat.

The timestamp discipline

So the practice is boring and quarterly. Write your channel theses with dates: "X ads reach our buyers at a fifth of LinkedIn's price, as of July 2026." Re-underwrite each quarter: is the number still true, or am I spending on a memory? Kill channels on evidence, enter channels on price gaps, and hold the portfolio loosely, because the portfolio is the strategy (that's a later chapter of the book).

The exercise for this one: write down your two main channels and, next to each, the year its performance numbers were last independently true. If you can't name the year, that's the finding.

This is a chapter of Right Channel, written in public. If you're sitting inside a window I haven't named, you have no reason to tell me, and if you tell me anyway I'll take it as proof the thesis is landing. Corrections and counter-data welcome as always.