if you woke up january 14th to your rpm down 50-90%, you weren't alone.

google's own status page shows it started jan 13 at 6pm UTC and wasn't "resolved" until jan 15 at 9:57pm. that's 52 hours. their words: "systemic decline in Ad Exchange (AdX) match rates and delivery specifically impacting Google demand sources."

here's what that means in english: google's own advertisers stopped bidding on your inventory. for two days. globally.

germany down 64%. spain down 90%. US sites reporting 50-80% drops. they called it fixed. check your dashboard. is it back to normal?

didn't think so.

but here's the thing. january wasn't the problem. january was the symptom.

sad to know. my adsense went from $1.5k/month to $218 over the past year and i thought it was just me. bad content. wrong niche. not enough traffic. so i did what any reasonable person would do. i spent a week pulling data from alphabet earnings calls, cloudflare's 2025 report, and every forum thread i could find (well might not something every reasonable person would do, but I belong to that category, I guess).

turns out it's not me. it's not you either. the game changed and nobody sent us the memo.

the 90/10 split

google's network revenue (adsense, admob, ad manager) is now only 10% of their total ad revenue. first time ever. the other 90% goes to youtube, search, ai overviews.

jason kint called it out in august: "i've been watching google's ad revenue mix shift from network to owned properties for over a decade. it just hit 90%."

look at that chart. 2013 we got 27%. every quarter since, it's been bleeding. slow enough that you don't notice until you check your year-over-year and wonder what happened.

you're not imagining things. it's structural.

the bot problem

this one took me down a rabbit hole.

cloudflare's 2025 year-in-review says ai crawling increased 15x this year. not 15%. fifteen times. and here's the part that made me close my laptop and go for a walk: some ai companies crawl your site 70,000 times for every 1 visitor they send back.

anthropic's crawl-to-refer ratio is 70,900:1. they're training on your content and sending you nothing.

that china traffic flooding your ga4? i dug into this. bots from lanzhou and singapore are triggering measurement calls without actually visiting your site. your pageviews go up. your CTR crashes. your rpm dies. and you sit there wondering why your "growing" site is making less money.

one publisher in the adsense forums said his rpm dropped 62% while cloudflare showed 6x more "traffic" than adsense counted. the math doesn't lie.

bots don't click ads. they just inflate your denominator.

january wasn't a bug

let me connect the dots.

publishers were already reporting 70-80% cumulative revenue declines since mid-2025. december's core update crushed traffic for content sites during what should've been the most lucrative month of the year. one guy reported christmas week down 85% from normal.

then january 14th hit and everyone noticed at once. but the crash didn't start in january. january just made it visible to people who weren't paying attention.

google called it a technical issue. maybe it was. but when your "fix" doesn't restore revenue to pre-crash levels, that's not a bug. that's a new baseline.

ai content gets rejected

found a documented case while researching. site with 700+ articles applied for adsense. rejected. owner ran the top 10 articles through an ai detector. 7 came back as 100% ai-generated.

they removed those articles. reapplied. immediately approved.

google can detect ai content. they're using it in approval decisions. they're just not telling anyone publicly. something to keep in mind if you're scaling with chatgpt and wondering why your approval keeps getting denied.

what the high-rpm publishers are actually doing

i spent time analyzing publishers who are still hitting $50+ rpm. wanted to understand what separates them from the rest of us.

three patterns kept showing up:

1. aggressive bot filtering. not just the obvious stuff. they're blocking ahrefs crawlers, semrush bots, entire chinese ip ranges. these crawlers create impressions without clicks. every fake impression tanks your CTR, and low CTR means lower bids from advertisers. one guy said blocking seo tool crawlers alone improved his rpm by 15%.

2. traffic source matters more than traffic volume. same site, same content, wildly different rpm by country:

netherlands: $143
germany: $99
UK: $95
USA: $69
brazil: $9

that's not a typo. netherlands is paying 15x what brazil pays. if you're optimizing for raw pageviews without looking at where they come from, you're leaving money on the table.

3. quality over scale. found a publisher making $1,350/month on 27,000 pageviews. that's $50 rpm. meanwhile there are sites with 500k pageviews making less because they're chasing volume with low-intent content.

the formula isn't more traffic. it's better traffic + fewer bots + commercial intent.

what actually works now

i'll save you the trial and error i went through.

stop chasing pageviews. advertisers pay for intent. "how to code a for loop" attracts people who will never buy anything. "best crm for construction companies 2026" attracts people with budgets. guess which one gets $50 cpm.

block bot traffic aggressively. cloudflare free tier is a start. block chinese ip ranges. block ai scrapers that take your content and send nothing back. your analytics will look worse. your revenue will look better.

diversify now. not eventually. now. google is actively moving advertiser spend to owned properties. the deal they made with publishers twenty years ago is being unwound quarter by quarter. if adsense is your only revenue stream, you're building on sand.

the uncomfortable part

ai overviews show zero ads while using your content to answer queries. think about that.

you write the article. google crawls it. their ai summarizes it. user gets the answer without clicking. google keeps 100% of whatever they eventually monetize that interaction for.

old deal: we crawl you, send traffic, you make money. new deal: we crawl you, answer it ourselves, thanks for the training data.

the 90/10 split isn't the end state. it's the trajectory.

#CaseStudy #Research

techwizardrino

Posts (5)