We Paid for Placement on a Site With Great Traffic. The Traffic Was Refreshing Itself.

A vendor sent us a media kit with a traffic chart that went up and to the right. Monthly visits pulled from one of the well-known estimation tools, screenshotted, cleanly formatted, sitting above a rate card that looked fair for the reach being claimed.

The level was believable. The shape was not.

Real publications have ragged traffic. December sags. A story catches and the month spikes. Somebody's algorithm changes and a quarter looks bad for reasons nobody at the publication can explain. This chart climbed in a smooth line and then held flat, the way a thermostat holds a room.

So we stopped reading the deck and went to look at the site.

What we found

The article pages carried a script that reloaded them on a timer. Open a piece, leave the tab alone, and the session count kept moving without a human doing anything. A large share of the remaining sessions arrived from a narrow set of low-quality sources that no reader gets to by accident.

Then we checked the things a real publication cannot fake cheaply. Almost nobody was searching for the publication by name. Several of their recent articles were not in the index at all, which means those pages could not be found by anyone who was not sent a direct link. There was no audience. There was a number, and the number had been built.

We had already placed client coverage there. That is the part I dislike telling, and it is the reason any of this is worth writing down.

Why reach is the easiest number in the media kit to fake

Third-party traffic estimates are models. They infer visits from panel data, clickstream samples and statistical fill, and they are reasonable at comparing large, well-known domains. On a small site they get thin, and thin models bend. The site owner controls the exact signal being sampled, and the site owner is the only person in the transaction with a reason to move it.

Set that against how a media buy actually gets decided. Reach arrives before the invoice.

Everything that matters more, whether the piece was read, whether it was indexed, whether it still shows up when someone searches the client, arrives weeks after the money has moved. So the industry buys on the one metric the seller controls, and audits the metrics the seller cannot touch only when something has already gone wrong.

Five checks we run before we place anything now

Ask for read access to the analytics rather than a screenshot. Genuine publishers hesitate, ask what you need, then give you a scoped view or share their screen on a call. A vendor with nothing real to show sends another image.

Look at branded search volume for the publication's own name. Nobody types the name of a site nobody reads. This is the single hardest signal to manufacture at low cost, and it separates real small publications from expensive empty ones better than any traffic estimate.

Take five recent articles and search a distinctive sentence from each in quotes. If the page does not come back, it is not indexed, and a placement there is worth nothing to the search result the client is actually judged on. This check takes four minutes.

Read the traffic shape across two years rather than the headline number. Straight lines are built. Real audiences are lumpy, seasonal and occasionally embarrassing.

Ask for one specific outcome from one previous client. Not a logo wall. One placement, one company, one thing that happened afterwards that somebody can confirm.

What this costs us

We lose work over this. When a client puts our plan beside a competing agency's, theirs has more placements and larger reach figures on the same budget. Ours is shorter and more expensive per placement, and defending it takes a meeting the other agency never had to sit through. Some clients take the bigger number, and I understand why. It is a better-looking document.

That is a real cost, and I would rather name it than pretend that doing the careful version also wins the pitch. It usually does not.

What it buys is narrower and slower. Coverage that is indexed is still working a year later, because it keeps appearing when someone searches the client's name. That search result is what most of our reputation work is eventually measured against, so a placement that cannot be found is not cheap coverage. It is a donation.

The check nobody wants to run

There is an uncomfortable version of this exercise available to almost every marketing team, which is to run those same five checks backwards across the placements you bought last year. Not to assign blame. To find out how much of your reported reach was ever real.

Most teams will not do it, because the reporting has already gone to the board and nobody wants to reopen a number that was accepted. That reluctance is precisely why the market for manufactured traffic keeps working. It is not a sophisticated fraud. It survives on the fact that reach is reported once, early, by the party being paid, and almost never checked afterwards by anyone.

Check it once. If your vendors are clean, you have lost an afternoon and gained a set of numbers you can actually defend. If they are not, you would rather find out from your own audit than from a client asking why they cannot find the article you told them about.