The subject that earns the most per visit is not the subject that sells for the most
930 words · about 4 min
Two questions get the same answer everywhere: which subject should I write about to earn well, and which one should I write about to sell well. They are treated as one question. They are not. Across the 11 subjects where we measure both, the rank correlation between earnings per thousand pageviews and the multiple a buyer pays is 0.036 — no relationship worth the name. Finance earns the most of any subject and sells at 35 times monthly profit, mid-table. Food and drink earns below the middle and sells at 42 times, the highest we measured. Information sites earn near the bottom and sell at 41.
- How it was measured
- Two tables, joined. Earnings per thousand pageviews come from 123 display-advertising businesses whose traffic and revenue a broker verified before sale, across 1.17 billion pageviews. Sale multiples are the median for display-advertising sales in each subject, from 2,661 completed sales, keeping only subjects with at least 15 sales so a median means something. 11 subjects appear in both tables. The rank correlation is Spearman's, computed over those 11. Eleven points is a small basis for a correlation and we say so: what the table below shows directly is that the two orderings are different, which needs no correlation at all.
- Last checked
Listen to this Playing read from this page, nothing added
If you are picking a subject to write about, you are really asking two questions:
1. What will this earn me each month? 2. What will it be worth if I sell it?
Everyone answers those as though they were one question. They are not the same question, and the numbers say so.
The two rankings
We measure both. Earnings per thousand pageviews — the figure the trade calls RPM — from 123 businesses a broker checked before they sold. Sale price as a multiple of monthly profit, from 2,661 completed sales.
Here are the 11 subjects that appear in both, sorted by what they earn:
| Subject | Earns per 1,000 views | Sells for | Sales measured |
|---|---|---|---|
| Finance | $29.73 | 35x | 22 |
| Home & Garden | $25.43 | 36x | 51 |
| Culinary | $20.59 | 36x | 40 |
| Pet care | $18.97 | 40x | 46 |
| News & education | $18.13 | 29x | 35 |
| Food & drink | $17.93 | 42x | 28 |
| Technology | $15.51 | 30x | 54 |
| Entertainment | $8.41 | 26x | 23 |
| Information & databases | $8.41 | 41x | 18 |
| Hobbies | $7.07 | 38x | 27 |
| Gaming | $6.48 | 32x | 20 |
Read down the first column and then the second. They do not track each other.
- Finance earns the most of any subject — $29.73 per thousand views, more than four times gaming. It sells at 35 times monthly profit, which is middling. - Food and drink earns below the middle at $17.93, and sells for more than anything else on the list at 42x. - Information sites earn near the bottom at $8.41 — and sell at 41x, second highest. - Gaming earns least at $6.48, and sells at 32x. Middling again.
The rank correlation between the two columns is 0.036. Zero would mean the orderings are unrelated. 0.036 is not meaningfully different from zero.
What that means in money
Two sites, same traffic, same monthly profit. One is about finance, one is about food.
The finance site earns more per visitor — considerably more. But when they sell, the food site fetches 42 months of profit against the finance site's 35. On the same profit, that is a fifth more money at the exit.
Neither fact is a secret. They have simply never been put in the same table, because almost nobody holds both — the earnings come from verified traffic and revenue, the prices from completed transactions, and they are different records.
Why the two would not track each other
We measured the difference; we did not measure its cause. So this section is reasoning, not evidence, and it is separated from everything above for that reason.
RPM is set by advertisers. It is high where the audience is worth reaching — someone about to move money, or to buy a boiler. It has nothing to do with how the business behaves after the sale.
A multiple is set by buyers, and a buyer is pricing risk and durability: how volatile the traffic is, how dependent on one search term, how much work the site takes each week. A recipe site with a decade of traffic and a mailing list is a calmer asset than a finance site whose rankings move with every update, even though the finance site earns more per visit.
That story fits the table. It is not proven by it.
What we are not saying
We are not saying the correlation is provably zero. Eleven subjects is a small basis for a rank correlation, and the honest range around 0.036 with 11 points is wide. What the table shows directly — that the top earner is mid-table on price, and the top-priced subject is mid-table on earnings — needs no correlation to see.
These are display-advertising sites. Sale multiples for the same subject under a different business model are not measured here.
A median is not your site. Each row summarises between 18 and 54 sales. Your own multiple depends on your own numbers, and the spread inside each subject is wider than the gaps between them.
What to do with this
1. Decide which question you are actually asking. If you are building income, the first column is your column. If you are building something to sell in three years, the second one is. 2. Do not use RPM as a proxy for value. It is the most available number and the most quoted, and on this evidence it tells you nothing about the exit. 3. Check your own multiple, not the median. Our free tool prices your site against the completed sales in your subject. Nothing you type leaves your browser.
If you are earlier than that, the more useful figure is what the door costs: we have measured what it takes to reach Mediavine's $5,000 bar, which is about 27,000 pageviews a month at the median — and between 14,000 and 64,000 depending on which row of the table above you are on.
Check this yourself
- What a thousand pageviews earns, by subject — 123 businesses, as a CSV - What businesses sold for, by kind — 2,661 completed sales, as a CSV - Archived with DOIs: 10.5281/zenodo.22280271 and 10.5281/zenodo.22280368
Both tables are published in full. The join above is one median per subject from each, and you can redo it in a spreadsheet.
Tracked as RPM-RANK-DOES-NOT-PREDICT-PRICE. If a number here looks wrong, tell us and we will check it against the source.