The marketing numbers that lie: bot traffic, vanity metrics and what to watch instead

Here is a scenario worth sitting with. A company opens its analytics one morning and active users are down 44% month over month. Pageviews down 30%. The obvious reading is that something broke, or the market moved, and the obvious reaction is to spend money fixing it.

The correct reading, in that particular case, was that the numbers had finally started telling the truth. Almost all of the traffic that disappeared had never been human.

How to tell whether your traffic is real

Open your analytics and add two columns to your country or city report: average engagement time and engagement rate. Then look at where the volume actually comes from.

The tell is unmistakable once you have seen it. Real markets show minutes - two minutes, fourteen minutes, occasionally twenty. Bot traffic shows zero seconds, and it clusters in a specific set of cities: Ashburn, Boydton, Des Moines, Cheyenne, Council Bluffs, Phoenix, San Jose. Those are not markets. They are the addresses of the world's largest data centres.

If the top ten cities driving your "users" are data-centre towns averaging under a second on page, then your headline traffic number is measuring crawlers, uptime monitors, scrapers and scanners. It is not a business metric. It is weather.

Why the number moves for reasons that have nothing to do with you

This is the practical danger. Bot volume is spiky and arbitrary. Turn on a bot-protection rule at your CDN and "traffic" collapses overnight - that is a win being reported as a catastrophe. Someone points a new scraper at you and traffic doubles - that is noise being reported as growth. Either way you are making decisions against a number that no customer influences.

There is a second, more legitimate reason your measured traffic can fall: privacy consent. If you run a compliant consent banner in Europe, visitors who do not opt in are not measured at all. Your analytics is now reporting a subset by design. Measured traffic went down; actual traffic did not.

The four numbers worth watching instead

1. Search Console impressions and clicks. This is the most underused dashboard in most companies. It is measured server-side by Google, so it is immune to bots inflating it, ad blockers suppressing it and consent banners gating it. If your analytics says traffic collapsed but Search Console shows impressions at an all-time high, you have a measurement story, not a demand story.

2. Engaged sessions, not sessions. An engaged session requires real attention - time on page, multiple pages, or a conversion. Bots essentially never produce one. Switching your reporting from "users" to "engaged sessions" removes most of the fiction in a single click.

3. Average engagement time by market. This is your reality check. It tells you which countries contain people. A market with 30 users at fourteen minutes each is worth more of your attention than one with 15,000 at zero.

4. One real conversion, measured end to end. Not a click on a button - the completed thing. A signup that exists in your database. A booking. A qualified enquiry.

The conversion point is where most measurement quietly fails

Almost everyone tracks the click on "start free trial". Far fewer track that the account was actually created, because the click happens on the marketing site and the account gets created somewhere else - a different app, a different codebase, often built by a different team at a different time.

That gap is tolerable when you are doing content and SEO. It becomes expensive the moment you spend on advertising, because ad platforms optimise toward whatever you tell them a conversion is. Tell Google Ads that a button click is a conversion and it will get very good at buying you button clicks - from people who never sign up. You will spend the budget and the campaign will report success.

Before any paid spend, the honest test is: can you point at one number that only increases when a real person completes the thing you care about, and has it actually fired at least once? If the answer is no, the measurement work comes first. It is not glamorous and it will delay the launch by a few days, and it is cheaper than the alternative.

What good looks like

• Traffic reported as engaged sessions, with bot-heavy sources understood and excluded from the story you tell
• Search Console as the independent check on organic demand
• Engagement time read by market, so you know where the humans are
• Exactly one primary conversion, defined as a completed action, verified end to end
• Everything else treated as diagnostics, not scoreboard

AI92 reports along these lines because the alternative is a dashboard that looks busy and tells you nothing. The number that matters is the one that only moves when a person decides something.