If you came from a cookie-based tool, the first thing you will notice is that nobody ever comes back. Somebody who reads your site every morning appears as a new visitor every morning. That is not a bug and it is not a gap in the data — it is the whole mechanism, working, and it is worth understanding on day one rather than week three.
The key changes every night
A visitor is recognised within a day by a value derived from their request, mixed with a secret that is specific to that day. At midnight UTC the secret is replaced and the old one is destroyed. From that moment, nothing in the system — including us — can connect today's visitor to yesterday's, because the thing that would have made the link no longer exists anywhere.
That is a stronger statement than "we do not link them". A promise not to do something can be withdrawn, or compelled, or broken by a mistake. A secret that has been destroyed cannot be used by anybody, including by somebody who asks us very firmly.
What you still get, within the day
Rather a lot, which surprises people who assume anonymity costs everything:
- Exact daily visitors. Not an estimate and not a sample — the real count, for that day.
- Whole visits. A visit ends after thirty minutes without activity, so somebody who reads three pages over an hour with a coffee in the middle is one visit, not two.
- Entry and exit pages, and the path between them. The journey through a single visit is intact.
- Where they came from. Referrer and campaign, attached to that visit.
What stops at the day boundary is the connection between visits. Within one, the picture is complete.
Reading the numbers without being misled
The rule is to compare like with like. A daily visitor count compared against other daily visitor counts is exact and trustworthy, and a trend built from it is as good as any trend you have ever had.
Where people trip is summing days into a month and calling the result monthly visitors. It is not — it is the sum of daily visitors, which counts a five-day-a-week reader five times a week. That number is still useful as long as you know what it is, and it is consistent, so the month-on-month shape is real even though the absolute figure is not what a cookie-based tool would have shown you.
When you genuinely need the other thing
Sometimes the honest answer is that this is the wrong tool. If your business runs on cohort retention — what proportion of January's signups were still active in March — then you need to recognise a person across months, and no amount of careful anonymous measurement will produce that number.
The useful distinction is between analytics and product analytics. Who is visiting your marketing pages, where they came from and what they read is a question about traffic, and it does not need identity. What a signed-in customer did across their first month is a question about an account you already have — and the place to answer it is in your own database, where you are allowed to know who they are because they told you.
Splitting the two questions is usually better than buying one tool that does both badly. It also puts the identified half behind your own login, which is a much easier thing to explain to anybody who asks.
The trade is set out in full on cookieless analytics, including the two other things it costs you, and why analytics tools never agree covers the other reasons two tools report different numbers for the same site.