Dark social is the sharing and recommendation activity your analytics cannot attribute — links pasted into private messages, names mentioned in meetings, advice traded in closed communities, and brands recommended inside AI chat answers. The influence is real; the trail is not.

Why does dark social show up as direct traffic?

Copying a link usually strips tags and referrer data, many messaging and native apps pass no referrer at all, and spoken recommendations pass nothing. When the buyer later types your address into a browser, most analytics platforms file the visit as direct, mixing it with habitual repeat visitors.

Are AI chat answers a dark social channel?

They behave like one. When an assistant recommends a brand inside a chat, the influence event happens without any trackable signal, and later visits look like direct or branded search. How engines pass referrers when users do click through varies by product and keeps changing, so today's behavior is an observation rather than a rule.

How can B2B teams estimate dark social?

Triangulate: add a free-text how-did-you-hear field to signups and deals, watch branded search trends as a rough shadow of invisible recommendation, listen in the communities buyers use, and scan AI answers to your buyers' questions on a schedule to see whether and how you appear.

What is dark social

Guide · Glossary & Definitions · 4 min read · last verified 2026-07-27

Reviewed before publication Editorial board Independent commercial review

In shortDark social is the sharing and recommending analytics cannot attribute — private messages, communities, meetings, and now AI chat answers. Why it files under direct traffic, how it distorts attribution, and how to triangulate it.

Dark social is the sharing and recommending your analytics cannot attribute: links pasted into private messages, names dropped in meetings, advice traded in closed communities, and — most recently — brands recommended inside AI chat answers. The term covers every path a buyer takes toward you that leaves no readable trail, and in business-to-business markets that appears to be a large share of the paths that matter.

Why the channel is called dark

Web analytics attributes a visit by reading where it came from — a referrer header, a tagged link, a recognizable source. Dark social breaks all three mechanisms at once. When someone copies a link from your site and pastes it into a chat, the tags and referrer are usually stripped in transit. Many messaging apps, mail clients, and native mobile apps pass no referrer at all. And a recommendation spoken aloud in a meeting passes nothing whatsoever, because there was never a link to begin with.

The traffic does not disappear; it gets mislabeled. A buyer who hears about you in a private channel and later types your address into a browser lands in the bucket most platforms call direct. Analysts have long observed that this bucket is a junk drawer — genuinely loyal repeat visitors mixed with the untraceable output of every private recommendation you have ever earned. Reporting built only on trackable sources therefore behaves like the person searching for keys under the streetlight: not because the keys are there, but because that is where the light is.

Where B2B recommendations actually happen

Ask practitioners where their best deals began and the answers tend to cluster in places no pixel reaches. A head of engineering asks a former colleague what she used at her last company. A private Slack or Discord community thread compares vendors candidly precisely because no vendor is watching. A conference conversation ends with a name scribbled in someone's notes. An internal memo argues for a shortlist, and the shortlist travels by email between people who never visit your site until the decision is nearly made.

What these moments share is that they are conversations between trusted peers, held in spaces designed to be private. That privacy is not a measurement bug to be engineered away; it is the reason the recommendations carry weight. A channel you could fully track would, by that fact, likely be a channel buyers trusted less.

The newest dark channel: AI answers

A growing share of early-stage research now appears to happen inside AI assistants. A buyer describes a problem, asks what kinds of tools exist, and receives named recommendations inside the chat itself. If your brand appears there, the influence event has already occurred — yet nothing reached your analytics. Should the buyer later search your name or type your address, the visit files under branded search or direct, indistinguishable from habit.

How assistants pass traffic when a user does click through varies by engine and changes as products evolve; today's referrer behavior is an observation, not a guarantee. What seems stable is the structural point: the recommendation and the click are separated in time and often in channel, so the recommendation goes unrecorded. Why AI referral traffic is undercounted examines the mechanics in detail. The practical conclusion is that AI answers belong on the dark social list — a word-of-mouth surface, except the mouth is a model.

What dark social does to attribution

Attribution models can only divide credit among touches they recorded. When the decisive influence was invisible, its credit does not vanish — it gets silently reassigned to whatever was visible, most often the final click. The predictable distortion: channels that capture existing demand look stronger than they are, while the activity that created the demand looks weaker or absent. Teams then fund the harvest and starve the planting. What is attribution modeling unpacks why every model carries this blindness, and How to measure AEO ROI shows what outcome measurement can look like when clicks are not the unit of proof.

Measuring the unmeasurable, imperfectly

You cannot instrument a group chat, but you can triangulate:

Add a free-text "how did you hear about us?" field to signup and deal intake, and actually read it. Self-reported answers are fuzzy on details yet frequently surface channels no tool recorded.
Watch branded search volume over time. People rarely search a name they have never encountered, so the trend often serves as a rough shadow of invisible recommendation.
Sit in the communities your buyers use, as a listener. Direction, not data.
Measure the AI answer surface directly: ask the engines your buyers' actual questions, record whether you appear, what is said, and which sources are cited, then repeat on a schedule. This is the approach Magrios takes — treating presence in answers as an observable to track rather than inferring it from click residue.

How to measure brand awareness without surveys extends this toolkit.

Influence what you cannot track

The uncomfortable summary is that much of your best marketing will never appear in your analytics, and the newest channels are making that more true rather than less. The response is not resignation but a shift of effort: be worth recommending, be easy to describe accurately in a sentence, and be present with citable evidence where the recommenders — human and machine — go looking. Dark social rewards brands that accept the dark part and work on the social part.