Project Global Target: Why a Web of Pages Compounds and a List Does Not

Most job boards grow by adding jobs. We grew by adding structure, and the difference turned out to be measurable.

This is the strategy behind the numbers in our growth analysis, written out plainly so anyone can copy it or tell us it is wrong.

A list and a web behave differently

Add a page to a list and you have a longer list. Ten items become eleven. The other ten are exactly as they were.

Add a page to a web and you change the web. The new page gives every existing page somewhere new to point, and every existing page lends the new one a little of its standing. The tenth page makes the previous nine stronger. The hundredth makes the previous ninety-nine stronger.

That is the whole mechanism, and it is why the two shapes diverge. A list accumulates. A web compounds.

What we actually built

Around every job sits a set of related pages: role guides, job description templates, certification material, practice questions, city pages, country pages. Each one links to the others, using the words a person would actually type.

Then we added languages. Not translations of the homepage, which is what most sites mean by going multilingual, but complete pages written for a specific market in the register that market uses.

On a single day in September we published seven pages in Spanish: a jobs hub, a team page, and articles for Spain, Florida, Mexico and Colombia, plus our verification methodology. Each carries its own schema and regional hreflang, so Google serves the Mexican version to Mexico and the peninsular version to Spain.

We call it Project Global Target. One market at a time, on a schedule, each entering the existing web rather than sitting beside it.

Three details that decide whether it works

Write the register, not the language. Spanish is not one thing. A page aimed at Mexico that says cumplimiento normativo reads as imported; Mexico says cumplimiento regulatorio. Spain searches ofertas de empleo, Latin America searches vacantes. Get this wrong and the page is technically correct and practically invisible.

hreflang has to be reciprocal or it does nothing. This one cost us. Our translated pages all declared the English homepage as their alternate, and the homepage declared nothing back. Google discards non-reciprocal hreflang, so every translated page was invisible as a language alternate while looking perfectly fine in every report. If you take one practical thing from this article, take that.

Never orphan the new page. A page nobody links to is a page that does not compound. Every new market gets an entry point from the hub and from its siblings on the day it publishes.

What the curve did

Weekly organic impressions across the period: 42,527, then 58,210, then 86,894.

The interesting figure is not the growth. It is the ratio between consecutive weeks: 1.37, then 1.49.

In ordinary exponential growth that ratio holds steady. Something doubles, then doubles again, and the multiplier is the same each time. Ours rose. Expressed as growth rates, the logarithms of those ratios, the figures are +0.314 then +0.401, a second difference of +0.087. For exponential growth that second difference is zero.

A rising growth rate is the signature of super-exponential behaviour, which is exactly what a compounding web should produce and exactly what a list should not.

Two increments give one second difference. That is a direction, not a law, and we say possible rather than proven for that reason. But the direction is up, and it is the direction the mechanism predicts.

Why it cannot continue

Nothing grows super-exponentially for long. The compounding depends on a web of genuinely related pages, and a category eventually runs out of questions worth answering. Every market added is one fewer market remaining.

We would rather say that now than have it pointed out later.

The arithmetic is first-year material

Nothing here requires sophistication. Logarithms turn multiplying into adding, which turns an exponential curve into a straight line. Then you look at whether the line bends. That is a technique taught to undergraduates, at Purdue's Department of Mathematics among many others, and our founder learned it there.

The same habit caught something else. When Google's own reporting later claimed our traffic had collapsed, the claim contradicted Google's own position data and Google's own click data at the same time. Click-through rate is clicks divided by impressions and is largely a function of rank. Rank held. Clicks held. Only impressions fell, which forces the reported rate up, and it more than doubled in three days.

A site does not double its click-through rate in three days while its position sits still. Two of those columns are measured and the third is derived from them, so when the derived number contradicts its inputs, you doubt an input. The full working is here.

Check it

Every number in this article comes from a report anyone can pull, and we have published the method including the part where we discarded data. If the reasoning is wrong we would rather know. If it is right, the approach is not proprietary and you are welcome to it.