Direct traffic registrations are four different problems wearing one label

Direct is the biggest row on the channel report. Two thousand four hundred registrations, more than paid search and paid social put together, and the only thing anybody says about it in the campaign review is that brand is strong.

Direct traffic registrations get that treatment because the label sounds like an explanation. It is not one. Direct is what your reporting says when a session arrived carrying nothing, and at least four unrelated things produce exactly that condition. Two of them are good news, one is a defect you can fix this month, and one is a marketing channel you are not measuring at all.

What direct means in a registration file

It means absence. Google's default channel definitions, current in 2026, make that explicit: the direct group requires the source to be exactly (direct) and the medium to be (not set) or (none). No positive evidence is involved.

That matters because absence has causes, and the causes leave different fingerprints. A returning attendee typing your url and a lost campaign tag both produce the same empty column, and they are separable if you cross tab the rows against things you already store.

Four sources, one label

Bookmarked and typed returns. Somebody who came last year, has the url, and goes straight there. Genuine direct, and the closest thing to a brand measurement in the file.

Untagged email. A send that went out without parameters, or a plain text version whose link was pasted rather than built, or a message forwarded onward with the tracking stripped by an intermediate client. The click is real and the campaign value never existed.

App and messaging clicks. A link opened from a desktop mail client, a team chat, or a messaging app. Many of these send no referrer at all, so nothing arrives even when the destination is right. This is the slice that is really a channel, and instrumenting it deliberately is B9's subject.

Redirect loss. The tag existed at the click and was thrown away by a redirect, a form step or a payment return before the record was saved. These are the ones that make a paid campaign look like brand strength.

Three cross tabs that pull them apart

None of this needs a new system. It needs three queries against the registration file and one against your email send log.

New against returning. Match each direct registration's email address to prior edition files. A returning registrant arriving direct is plausible. A first time registrant arriving direct is not, because somebody who has never attended has no reason to know the address, so a marketing touch happened and its evidence was lost.

Device. Split direct registrations by device. Mobile skews heavily towards clicks that came out of an app or a message, because that is where those links get opened. Desktop skews towards bookmarks and mail clients.

Hour of day against send times. Plot direct registrations by hour across the campaign, then overlay your email send times. Email produces a sharp, short spike. Bookmarked returns produce a flat weekday daytime pattern that looks nothing like it.

Working the split of 2,400

Take the 2,400 and run the three cross tabs.

The prior edition match returns 1,010 returning registrants arriving direct. Not all of those are bookmarks, because a returning attendee also gets your emails, so cross reference against the send list before claiming them.

The hour of day analysis is where the email slice appears. Direct registrations run at a baseline of about 12 per hour across the campaign. In the four hours following each of three sends, they run at 62 per hour. That excess of 50 per hour across twelve hours gives roughly 600 registrations arriving direct within a send window, and 610 of those rows also match an address on the send list. Those are email, whatever the report says.

Device and referrer take the next slice. Across all 2,400 direct registrations, 1,340 are mobile, which is well above the 46 per cent mobile share of the file as a whole and is the first sign that something other than bookmarks is at work. Of the rows left after the send window matching, 700 are mobile with no referrer and no send list match, arriving in clusters rather than evenly. That is the app and messaging group.

The redirect slice is measured rather than inferred: the hop test on your own links says a particular redirect drops the query string, and 250 direct registrations have a landing url that only that redirect produces.

That leaves 840 as bookmarked and typed returns, and the four add back to 2,400. The pie chart that said direct was 2,400 now says brand is 840, email is undercounted by 610, there is a 700 registration channel nobody is measuring, and 250 registrations are a bug.

Why does email keep landing in direct?

Three separate reasons, and only one of them is your fault.

The first is untagged links, which is a process failure and is fixed by building sends from the same link builder as everything else.

The second is forwarding. A colleague forwards the invitation, the recipient clicks, and depending on the client the parameters may survive while the identity of the original recipient does not, so the registration is attributed to a send it did not come from. Shared inboxes do the same thing at larger scale, and on shows selling to procurement teams a single monitored address can sit behind a dozen registrations.

The third is that the engagement evidence you would use to catch the first two has been getting weaker. Apple announced Mail Privacy Protection in June 2021, describing it as helping users prevent senders from knowing when they open an email, and masking the reader's IP address so it cannot be linked to other online activity or used to determine location. Open data stopped being a reliable signal from that point, which matters here because opens are what teams reach for when a click arrives with nothing attached.

The referrer will not rescue you either. Chrome moved its default referrer policy to strict-origin-when-cross-origin in version 85 in 2020, so a cross origin click carries the origin and never the path or query string. You can see that a visitor came from a webmail domain. You cannot see which campaign.

Which of the four can you actually fix?

The redirect slice, immediately, and it is usually the smallest. A query string rule in one config file, and 250 registrations start arriving with their source attached.

The untagged email slice, within one campaign cycle, by moving every send to a link builder and adding a check that no send goes out with an untagged link in it. That is 610 registrations moving from a label that means nothing to a channel with a cost attached.

The app and messaging slice is a build. Share links with a distinct medium, per person invite codes, and a way of spotting registrations arriving in bursts from one employer. That work is B9's.

The bookmarked returns are not a problem at all, and once the other three are handled, the direct row becomes a number worth reading. A direct row that is 840 registrations of returning attendees is a brand measurement. A direct row of 2,400 mixed causes is a filing cabinet.

The one number to watch after that is first time registrants still arriving direct. In this file, 1,390 of the 2,400 are new to the show, and every one of them found you somehow. Track that count on its own line each edition. It should fall as the other fixes land, and when it stops falling, whatever remains is the size of the acquisition you are getting from routes that leave no evidence at all.

Where the split stops

The four categories overlap at the edges, and the boundaries are decisions rather than facts. A returning attendee who clicked an untagged email is both a return and an email registration, and the method above assigns them to email because the send window match is stronger evidence than the prior edition match.

The hour of day method also degrades when sends are frequent. A show sending twice a week has no clean baseline hours left to compare against, and the excess calculation stops separating anything. On those shows, the send list match does most of the work and the timing analysis becomes a check rather than an estimate.

The last limit is that this whole exercise reallocates labels without recovering evidence. It gives you a better read on where registrations came from and it does not put a campaign value back on a row. Fixing the causes for the next edition is the only durable outcome, which is why the split belongs in the same review as the wider unattributed decomposition that B7 sets out, and why a self referral problem should be ruled out first through the cross domain check in B4.

Run the hour of day query this week. Count direct registrations by hour for the last thirty days, mark your send times on the same chart, and read the four hours after each send. If direct registrations triple in that window, you have found several hundred registrations that belong to email, and you have found them without changing a single line of tracking code or waiting for the next acquisition and attribution release cycle.