Link analytics: the six numbers worth reading, and the rest

Open any link's stats page and the biggest number on it is clicks. It is also the one that will mislead you most often, because it counts requests and you are trying to count decisions.

Six numbers on that page do better.

Unique visitors

Distinct visitor identifiers within a day. The identifier is a hash of a rotating salt, the visitor's address and their user-agent string, and the salt advances every 24 hours, so the same person on Tuesday and Wednesday is two unrelated identifiers.

That design has a consequence worth stating: uniques is a per-day number and cannot be summed across days. Seven days of uniques added together is not weekly reach, it is the sum of seven daily reaches. If someone reports a monthly unique figure built by addition, it is wrong.

The number is also an estimate. It comes from a probabilistic counter that keeps a sketch instead of a list, which is deliberate — there is no list of your visitors anywhere, so there is nothing to leak, subpoena or sell. Accuracy at normal volumes is well inside the range where the decision would not change.

Scans, separately from clicks

A visit arriving through a QR code is marked as a scan. That single separation is what lets a link live on a poster and in an Instagram bio at the same time and still tell you which one worked.

Every gc.mw QR encodes the short link with a marker attached, so the attribution is structural rather than inferred from a referrer that phones do not send.

Bots, in their own column

Known bot user agents and requests the network edge scored as automated are counted as bots and kept out of every audience breakdown. Country, city, device, browser, referrer and campaign tables are human traffic only; the bot number sits beside them as its own total.

Read it as a health signal. A link whose bot share jumps overnight has usually been posted somewhere it is being scraped, or is being probed. A link with a large bot share from the first hour is usually in an email being opened by corporate security scanners.

Hour of day

Twenty-four buckets, in UTC. The most underused chart we ship.

A campaign link with a tight spike an hour after each send is an email working. A print link with an even line across the day is a code people scan when they happen to walk past it, which tells you the placement matters more than the timing. A link whose traffic peaks at 3am local is almost always automated, whatever the bot column says.

Referrer type, before referrer host

Referrers are grouped into search, social, email, ads, QR, referral and direct. The grouping is more useful than the raw host list, because the raw list fragments — one social network can produce four or five hostnames depending on the app and the country.

Direct is the bucket that needs interpreting. It contains people who typed the link, people who scanned a code, people arriving from apps that strip the referrer, and everyone on a link opened from a document. "Direct" almost never means "typed it in".

Variant, when you are testing

If a link carries geo rules, device rules or an A/B split, every click records which branch it took — geo:IN, device:ios, ab:A. The variant breakdown is the only honest way to know whether a rule you wrote is serving anyone.

A rule with under a couple of per cent of traffic is usually not worth its maintenance. A country high in the country breakdown with no rule pointing at it is the next rule to write.

What to ignore

Number	Why it misleads
Total clicks, all time	Includes bots' near neighbours, repeats and years of drift; useful only against itself
Uniques summed over a range	Daily identifiers cannot be added; the salt rotates every 24 hours
Click-through rate from a shortener alone	We see the click, never the impression that preceded it
A single day's country split	Small samples produce dramatic-looking geography that reverses next week

Reading it in practice

Start with uniques, not clicks. Check the bot column for anything sudden. Look at the hour chart to understand the shape of demand. Use referrer type to find where the traffic actually came from, and only then open the host list.

Everything here applies to every branded link on any plan; what changes with the plan is how far back you can look and how much detail the tables carry. On Free you get a seven-day series and the top few referrers; from Pro upwards the full breakdowns and a year of history; on Business and Team, three years and period-on-period comparison.