We flipped every ad switch and measured the screen.
Ad-quality watchdogs flag sites where ads cover too much of the screen. We used two small test sites, one on Mediavine and one on Raptive, changed ad settings (one at a time on Mediavine, stacked step by step on Raptive), and had a robot reader scroll real articles on a phone and a desktop to measure how much of the screen was ad.
Where the "35%" and "50%" lines come from
Advertisers don't pick websites one by one. They buy through systems that grade sites first, and a bad grade means fewer advertisers bidding, which means lower pay. One of those graders, DeepSee, publishes its red lines. A site lands in its "High Risk" group if either is true:
• Ads cover 35% or more of the screen on average while you scroll. Picture a phone: more than a third of what you see, on average, is ads.
• Ads cover half the screen or more at some moment, on most visits. One bad screenful is enough.
DeepSee says only about 1% of the sites it tracks cross each line. Other graders, like Jounce and OpenSincera, score sites too but don't publish a number. So we used DeepSee's lines as the yardstick. Read DeepSee's rules →
Why the spikes matter more than the average
How the grader looks at a site. DeepSee's robot visits a site again and again. Each visit is a "crawl". During a visit it takes snapshots of the screen and measures how much of each one is ads. Its two red lines read those snapshots in different ways:
• The average line (35%) is named "Average Snapshot Density": all the snapshots averaged together. Screens that show only the article pull the number down, so a few ad-heavy moments get diluted.
• The peak line (50%) trips if ads fill half the screen "at least once during most crawls" (DeepSee's words). Nothing dilutes it. One bad screen on a visit counts, and if that happens on most visits, the site is flagged.
What our test showed. Averages ran from about 3% to 18%, far below 35%. Peaks sat much closer: the typical worst screen was 23–45%, and some runs went past 50%. Three changes caused those over-50% moments: full-screen ads, the top banner, and Raptive density on Max. That's why peaks, not averages, are the thing to watch.
What we don't know. DeepSee doesn't publish how many visits it makes, what screen size it uses or how far it scrolls. "Most crawls" suggests a spike on 1 visit in 6 wouldn't flag a site on its own. 2 in 6 (Max density) is getting close, and real readers see more variety than our 12 test views.
Key takeaways
Even the most aggressive setting (Raptive density on Max) averaged 18% of a phone screen. The risk is short spikes, not the average.
Turning it off cut desktop coverage by about 60%. But that bar earns about 19% of ad income on comparable sites: roughly $50 → $40 RPM. Expensive.
They didn't move the average, but each produced a one-off peak over 50% (1 of 6 runs). If a watchdog cares about peaks, these are the cheap ones to drop: full-screen ads earn about 2–2.5%.
Turning it off cut phone coverage by about 40%, but ads loaded fell from 30 to 19. Rough phone RPM estimate: $40 → $26.
Max loaded about 35% more ads (best case ~$40 → $54 RPM) but pushed 2 of 6 phone runs past 50%, which is DeepSee's red line. Very Low cut ad loads but barely beat simply turning off large format.
The sidebar slide-in and the first-screen speed setting showed no measurable change in coverage. Leave them alone.
Our recommended setup
Averages never came near the 35% line, so cutting ads across the board buys almost nothing. The smart move: keep the ads that earn, switch off the few that cause spikes.
✕ Full-screen ads: off. Caused a spike past 50% and earn only about 2%.
✕ Top banner: off. It caused the worst moment in the whole test (59%).
✓ Sticky bottom bar: keep. It's the biggest coverage lever, but it earns about 19% and coverage stays low with it on.
– Slide-in sidebar, first-screen speed mode: leave at default. No measurable effect.
✕ "Turn on new features automatically": off, so new spike-makers don't switch on quietly.
✕ Full-screen ads: off.
✓ Density: Optimized / Medium, never Max. Max pushed 2 of 6 phone visits past 50%.
✓ Large-format ads: keep. Turning them off cut about a third of phone ads and barely moved the peaks (the typical worst screen stayed around 40–44%).
✕ Auto-optimize: off. It's the only way to stop Raptive raising density on its own. With it back on, the live density on our test site crept from 22% to 24%: a hint, not proof.
? Frame ads: maybe off on desktop. Desktop coverage fell from 9.9% to 6.3% with no fewer ads loaded, but only 4 page views. Test it first.
How sure is this? About 12 page views per setting on two small sites. That's enough to see which settings cause the big moves, not exact dollars. A few weeks of earnings on a real site is the final word.
How to read the charts
- Blue bar = average share of the screen covered by ads while scrolling the article. Lower is better for readers.
- White dot = the typical worst moment on a page (median of the peak across runs).
- Orange tick = a single run that spiked past 50%.
- Dashed lines = the levels one watchdog uses as red flags: 35% average or 50% at any moment.
- Ads loaded = how many ads loaded per page view.
- Est. RPM = a rough guess at earnings per 1,000 page views, explained below.
Mediavine test site
Mediavine has no density dial, only on/off switches. We flipped one switch at a time and put it back before the next one.
Phone
typical peak 27% · 37 ads loaded
typical peak 28% · 33 ads loaded (-12%)
typical peak 23% · 28 ads loaded (-26%)
typical peak 27% · 29 ads loaded (-21%)
typical peak 27% · 30 ads loaded (-18%)
typical peak 28% · 31 ads loaded (-17%)
Desktop
typical peak 11% · 27 ads loaded
typical peak 14% · 33 ads loaded (+23%)
typical peak 9% · 30 ads loaded (+12%)
typical peak 13% · 33 ads loaded (+22%)
typical peak 14% · 34 ads loaded (+24%)
typical peak 12% · 33 ads loaded (+23%)
Raptive test site
Raptive has on/off switches plus a density dial. Here the changes stack: each row keeps the changes above it. Auto-optimize was turned off so the settings would stick.
Phone
typical peak 39% · 30 ads loaded
typical peak 45% · 30 ads loaded (-1%)
typical peak 44% · 19 ads loaded (-36%)
typical peak 35% · 22 ads loaded (-29%)
typical peak 43% · 41 ads loaded (+36%)
Desktop
typical peak 29% · 36 ads loaded
typical peak 29% · 36 ads loaded (+0%)
typical peak 29% · 30 ads loaded (-18%)
typical peak 28% · 23 ads loaded (-37%)
typical peak 30% · 48 ads loaded (+32%)
About the dollar estimates
- The test sites get almost no traffic, so we start from a typical RPM for real sites on each network: ~$50 on Mediavine and ~$40 on Raptive.
- Mediavine rows: when a switch removes a whole ad unit, we take off what that unit earns on comparable real sites. The sticky bottom bar earns about 19%; full-screen ads about 2%.
- Raptive rows: we scale by how many ads loaded, as if every ad sold for the same price. That overstates the gain from adding ads (extra ads usually sell for less) and may overstate the loss from removing big formats (those often sell for more).
- We counted bids but didn't record their prices, so these are ballpark figures, not measured revenue. Only a few weeks of real earnings after a change can settle it.
What this test can and can't tell you
- Each setting was measured on 3 articles × 2 passes × phone and desktop: about 12 page views per setting. That is enough to see big effects; differences under about 3 points are noise.
- Ads change from visit to visit, so a single run can spike. That's why we show both the typical peak and any one-off spikes.
- We measured coverage, not revenue. "Ads loaded" is only a rough hint at revenue. The real cost of a change shows up in a few weeks of earnings data.
- The watchdogs' own scoring is private. Matching their published thresholds is a good sign, not a guarantee.
- No ads were clicked. Both sites were put back to their original settings after the test.
Measured with a headless browser (390×844 phone, 1440×900 desktop) that scrolls each article in steps and records the share of the visible screen covered by ad slots at every stop.