We audited our own website and found half the traffic was fake

In September we ran the same audit on ourselves that we run for a new client. We started where anyone starts, with the headline figures, and they were excellent.

Over ninety days, users went up 35 percent. Sessions up 17 percent. Over the most recent twenty-eight days, page views have gone up 132 percent. Engagement rate at 94.6 percent. Key events, which is what Google Analytics calls conversions, are running at a rate of 88 percent.

If you saw that on a client report, you would think the quarter had gone well.

The numbers underneath

Then we looked at where the people came from.

Of 677 users in twenty-eight days, 350 were in the United States. The single largest group after that was Singapore, with 240. Then Brazil with 18, Canada with 17, Bangladesh with 16, Argentina with 13, Pakistan with 13, India with 12, and Ukraine with 11.

We are a marketing agency in Burlington, Massachusetts. We do not serve Singapore. And Singapore, specifically, is where one of the large cloud providers runs a data centre region, which is where automated traffic tends to come from.

Roughly half our audience was not a potential customer. That was the first thing.

The second was the device split. Of 988 sessions, 934 were desktop and 55 were mobile. That is 94 percent desktop. Real local small business traffic runs somewhere between 40 and 60 percent mobile, because people search for local businesses on their phones. A 94 percent desktop split is not a browsing pattern. It is a crawling pattern.

The third was the pages. Our homepage served 1,516 pageviews to 306 users, which is five pageviews per person on a homepage. Our locations page served 293 page views to 10 users, which is twenty-nine views each. Nobody reads a locations page twenty-nine times. Something was walking through the links.

The fourth was the referral traffic. Referral users were up 1,165 percent over the quarter. The two largest sources were a site called trafficheap.cc, which sent 180 sessions, and syndicatedsearch. goog, which sent 89. The first of those is a bought-traffic service. Our growth was, in part, purchased.

Bot traffic is annoying. The measurement problem was worse because it was our own doing.

In Google Analytics 4, you tell it which events count as conversions by marking them as key events. Somewhere along the way, page_view had been marked as one.

That means every single pageview was being counted as a conversion. In twenty-eight days that produced 4,764 key events. It is why the conversion rate read 88 percent and why every channel looked like it was converting beautifully.

Here is what actually happened in those same twenty-eight days. One form submission. Four forms start. Forty-two clicks on anything trackable.

And while we were counting pageviews as conversions, the event that genuinely mattered was not being counted at all. Over ninety days our site recorded 125 tel_click events, which is people tapping the phone number. For a local services business, that is the conversion. It was sitting there, untracked, worth more than everything on the dashboard, and invisible.

Once you strip the noise out, the real picture is smaller and much more useful.

Around 300 to 350 real United States visitors a month. About 35 of them are arriving from organic search. Roughly 42 phone taps a month. One or two genuine inquiries.

And one finding we would never have seen through the noise: organic search, the smallest channel on the report, was by far the best. Those visitors stayed three and a half minutes and came back nearly three times each. Every other channel averaged under a minute.

Organic was also the only channel that had gone down, by 6.8 percent.

So the honest summary of our quarter is this. The channel that worked was shrinking. The channels that grew were bots and bought traffic. And the dashboard said everything was fine.

Five things, and the first four took an afternoon.

We unmarked page_view as a key event and marked the things that are actually conversions: phone taps, form submissions, chat completions, and bookings.

We added referral exclusions for the junk traffic and stopped accepting bought traffic from anywhere.

We built an audience for the towns we actually serve so we can see local visitors separately, and we started applying a United States filter to every report we look at.

We wrote down six honest baseline numbers, and every decision from here gets measured against those rather than against the flattering ones.

Then the bigger one. We had location pages for thirty-seven towns. Most were near-identical, most covered places we do not really serve, and they were competing with each other for the same searches. We are retiring thirty-three of them and rebuilding around the ten towns in our actual patch. Four of those ten already existed and needed rewriting. The other six had never been written at all, which tells you something about how that list was put together in the first place.

You do not need an agency for this. Open your own analytics and check four things.

One, where are they? Reports, then demographics, then country. If a large share is somewhere you do not sell to, and especially if one foreign country is unusually dominant, you are looking at automated traffic.

Two, what device? Reports, then Tech, then Device category. If you are a local business and the desktop is above 70 percent, something is off.

Three, what counts as a conversion? Admin, then Events. Look at the Mark as key event column. If page_view or scroll is toggled on, every number you have been reporting is a pageview count wearing a different name.

Four, what is your actual inquiry count? Not sessions. Not engagement rate. The number of people who called or filled in the form last month. If you cannot get to that number in under a minute, that is the thing to fix first.

If three of those four give you an uncomfortable answer, you are roughly where we were in August.

Every agency says it reports honestly. It costs nothing to say.

This is what it looked like when it cost us something. Our numbers were better before we checked them. We could have fixed the settings quietly and let the chart keep going up.

We would rather show you the version where we were wrong, because that is the only version that tells you anything about how we will report on your account.

If you want the same check run on your site, our free 3-point growth leak review is exactly this, done for you. Hand-reviewed. No obligation. No sales sequence. You keep the findings.