Dark social

Why the referrer disappears

A browser only attaches a referrer when one page links to another. Private sharing breaks that chain in several ordinary ways: someone copies your URL and pastes it into a chat, and the recipient's click has no page of origin at all; a native mobile or desktop app opens the link in its own view without setting a referrer; a link carries rel="noreferrer"; or a privacy setting strips it. None of this is exotic or new. It is what happens every time a human recommends you to another human.

The result is a permanent measurement floor rather than a bug to fix. Analytics can only report what the browser sends, and the most valuable moment in your funnel — a person telling another person that you are worth trying — is precisely the one that sends nothing. Dark social is the name for that gap, and treating it as a category rather than an anomaly changes how you read every report you own.

What it looks like in your reports

You cannot see dark social directly, so you learn to recognise its fingerprints. Four are reliable:

• Direct traffic arriving on deep, unmemorable URLs — a long blog slug or a comparison page nobody types from memory.
• A spike in Direct with no campaign, no launch and no referral increase to explain it, often a day or two after you posted somewhere active.
• Branded search volume rising while referral traffic stays flat: people are hearing your name somewhere you cannot see and looking you up.
• Signups whose self-reported source is a person or a community that never appears anywhere in your analytics.

How to estimate it without deceiving yourself

There is no setting that recovers a referrer that was never sent, so the honest goal is triangulation, not attribution. Ask people directly at signup with an optional free-text field: it is noisy, self-reported, and at low volume still the single most informative line of data you have. Put UTM parameters on every link you post yourself, so at least your own sharing stops being counted as dark. Then read Direct plus branded search together as one demand line, and watch whether it moves when you do something worth talking about.

What you should not do is assign the traffic to whatever channel you already believe in. The reason marketing attribution models feel unsatisfying is that they distribute credit among the touchpoints they can see, and dark social is by definition outside that set. A model that shows a tidy 100% has not measured the gap; it has hidden it.

Why it is growing

Discovery keeps moving into places that do not emit referrers: private Slack and Discord communities, group chats, newsletters read inside a mail client, and now AI assistants, some of which pass a hostname and some of which do not. The practical consequence is that the share of your traffic you can name is shrinking every year, regardless of what you do. We see it in our own funnel at AgentCeres — the AI Growth Officer at agentceres.com: the only first touch we can reliably tie to a signup is a parameter we put on the link ourselves, such as an ad click identifier, and everything that travels person to person arrives as Direct. We treat Direct as a bucket to be explained rather than a channel to be optimised.

That is also why measuring word of mouth works better as a question than as a report. Ask new customers where they heard about you, keep a list of the communities that keep coming up, and check whether your name shows up in AI answers — tracking traffic from ChatGPT and AI search covers the mechanics of the half that is measurable.