Short link clicks against Google Analytics

A campaign reports 2,708 clicks on the short link and 2,410 sessions in Google Analytics. Neither number is broken, and neither tool is lying. They are measurements taken from two different places, and the gap between them has named parts.

Bitly documents the same thing about its own numbers, pointing at bots, browser plug-ins and email security scanners that expand links. This post is the version of that answer with the arithmetic shown.

One period, two numbers

Requests received

3,120

Known crawler

−244

Link prefetch

−121

No user agent

−47

Counted clicks

2,708

Sessions on the destination

2,410

298 clicks counted at the redirect never became a session on the destination page.

A redirect hit and a page view are different events

A shortener counts a request it answers. An analytics tool on the destination counts a page that loaded and ran its script. Everything that happens between those two moments is a source of difference, and all of it goes one way.

A visitor who taps the link and closes the tab before the destination renders is a click and not a session. So is a visitor whose ad blocker or tracking protection stops the analytics script, and so is a visitor on a connection slow enough that they gave up. This is why the redirect-side number is generally the larger one and why it cannot be corrected by tuning the smaller one.

uhm.co writes the click before the 302 is sent, so the count does not depend on the destination page at all. That property is what makes the comparison meaningful in one direction: a click with no session tells you something about the destination, while a session with no click means the visitor arrived by some other route.

Clicks are not sessions

A session is a container. Analytics tools group several page views by the same person within a window into one session, and a return visit later in the day may or may not open a new one depending on the tool's rules.

uhm.co applies no session window at all. Two clicks a minute apart from the same person are two clicks. The only deduplication in the pipeline guards against its own retries, so a single real visit is never reported as two because of a fault on this side.

Comparing a click count with a session count therefore compares a count of events against a count of groups. Some of the gap is simply that.

Bots reach the redirect first

Anything that fetches a URL reaches the shortener, because the shortener is the URL. Crawlers, uptime monitors, mail security scanners and browser prefetch all send real requests that no human made, and they arrive at the redirect whether or not the destination page ever loads its analytics script.

uhm.co filters three categories and names each one in the dashboard: known crawlers, browser prefetch, and requests with no user agent at all. The filtered count sits next to the total rather than being quietly subtracted from it, which is the only way a customer can check the number rather than take it.

A fourth category, traffic from datacenter address ranges, is defined but is not being applied today. It is named here because it appears in the product's own documentation, not because it is filtering anything.

The day boundary is a choice, and the two tools chose differently

uhm.co buckets clicks into days in UTC. Google Analytics reports in the property's configured time zone. For a campaign that runs in Turkish evening hours, that alone moves clicks across a date line and makes a day-by-day comparison disagree even when the totals for the week match.

Compare totals over a range with a margin at both ends before concluding that anything is wrong.

What to do with the gap

The useful question is not which number is right. It is whether the gap is stable. A gap that holds steady across campaigns is the normal distance between two vantage points. A gap that suddenly widens is a finding: a broken tag on the destination, a campaign hitting an audience with heavier ad blocking, or a burst of automated traffic that the filter reasons will name.

The exact rules this product follows, including what is filtered and why it will never match another tool, are written out on the methodology page.