We recorded a 23x revenue spread, then found most of it was a missing import

This finding was written up in a weaker form first, and the weaker form was wrong. The correction is left in view because the way it broke is more useful than the number it originally produced.

What we recorded first

Ad revenue per thousand page views, joined to page path across one property for twenty eight days, ran from Rs 47.29 on the best page to Rs 2.08 on the worst. A 23x spread inside a single site, on one template, under one ad account. It was written up as evidence that what a page is about matters far more than where the ads sit.

What the check found

The claim attached to that spread was that both pages carried the same two ad slots, so placement could not explain the gap. Nobody had opened the file. Reading both pages out of version control, the top page carries two ad slots and the bottom page carries none.

The worst earning page on the site was not a monetisation puzzle. It was a missing import. It returned Rs 2.08 per thousand views because it was serving almost no ads, and it entered the analysis as though it were serving two.

The number that survived

Drop that page and hold slot count genuinely constant, at two slots on every remaining page, and the spread is Rs 47.29 down to Rs 10.25. That is 4.6x, and it is real: same template, same slot count, same ad account, same window, same country mix.

Three things move page revenue

Page revenue depends on whether the slots exist at all, how many ads render per view, and what advertisers pay for the ones that do. Each was measured across the same page set, and the ranges are worth having, but read the warning under the table before using them.

These three are SEPARATE RANGES, not a decomposition of the 4.6x, and multiplying them is wrong. Each is a maximum against a minimum taken across the page set, and the two extremes of one need not fall on the same pages as the extremes of another. 3.3 times 1.8 comes to 5.9, which is larger than the 4.6x it would supposedly explain, and that arithmetic is the tell.

Only the third is a defect. The first two are properties of the content, and they point the same way: pages that hold attention serve more ads and attract better paying ones. Ad click-through on the deepest page ran 2.11 percent, against 0.00 percent on a single-answer calculator in the same window.

What the correction cost

The original write-up concluded that depth of engagement was a far larger lever than geography: 23x against a measured 4.6x for the gap between Indian and Anglosphere ad rates on the same property. With the defect removed, the two levers are the same size. That conclusion did not survive its own check, and the version that did is less convenient.

A second thing the check overturned: the subject we had assumed was the weak monetiser was the strongest one on the site. It had been called low value on the strength of published vertical benchmarks. On this property it held the top nine revenue slots outright. The benchmarks did not describe this site, and its own measured data was available the entire time.

How to check your own

Pull revenue, ad impressions and views per page rather than per site. Compute ads per view first, because a low figure there is either a missing slot or a page nobody scrolls, and those need opposite responses. Then read the slot count out of the source for your best and your worst page before writing any sentence that depends on it. That step takes half a minute, and it is the one that gets skipped.