1. Placement and density (documented, fastest, most abused)
Google's own optimization guidance documents the mechanism: ad placement, formats and the number of units change earnings, and its experiments tooling exists to test exactly that Google AdSense Help2026-08. The arithmetic is on the impression RPM page: page RPM = impression RPM × units × fill. The bill: each added unit earns less than the last, slows the page, and pushes toward the policy line on ad-heavy layouts. Density is a throttle, not a strategy.
2. Content mix toward demand (structural, compounding)
Advertisers bid on audience value, so shifting your publishing calendar toward higher-demand topics within your niche raises blended RPM without touching layout. The demand ordering and its sources are on average RPM by niche. The bill: it is slow, and chasing demand outside your site's authority can cost the traffic that multiplies the RPM.
3. Geography mix (real, hard to steer)
High-bid markets pay multiples of low-bid ones for the same pageview. Networks price this openly: Raptive's published entry rules require 50% US/CA/UK/AU/NZ traffic at its 25k tier Raptive Help2026-08. You steer it through topic selection and SEO targeting, slowly. The bill: deliberately excluding audiences is usually worse business than serving them.
4. Timing and seasonality (free, only needs patience)
Q4 pays the year's best rates and January the worst, visible in the dated Ezoic Ad Revenue Index Ezoic Ad Revenue Index (dated series)2026-08. Schedule content launches and layout experiments so that peak inventory is live before November, and never judge an experiment across a quarter boundary. See why did my RPM drop for the calendar.
5. The network switch (the biggest single step, when eligible)
Moving from self-serve to a managed network changes the revenue share, the demand stack and the optimization layer in one contract. Published shares: AdSense pays 80% after the advertiser platform fee Google AdSense Help2026-08, Mediavine publishes 75% base rising with scale Mediavine Help2026-08, Raptive publishes 75% Raptive Help2026-08. What is not published anywhere is your uplift; treat every projected percentage as marketing until your own test proves it. Thresholds and terms: the network table; decisions: the head-to-heads.
The no-evidence pile
| Claimed hack | Published evidence |
|---|---|
| Posting time of day changes RPM | not published |
| Word-count minimums raise RPM | not published |
| Refreshing publish dates lifts ad rates | not published |
| One weird ad size that networks hide | not published |
None of these has a named, dated source behind it. Spend the effort on the five levers above; the redline on this site marks published ceilings, and these hacks have not published so much as a floor.