Half of ecommerce orders are "direct." 91.7% of those customers know better.

A shopper hears about a brand from a friend, buys it a week later, and the dashboard files the sale under "direct". According to a new study
of 3,303,402 post-purchase survey responses across 3,237 online
stores, that is what happens to more than half of ecommerce orders,
and the customers involved know precisely how they found the brand.
The analytics simply has no column for the answer.

The Zigpoll Attribution Index, published by survey and attribution
platform Zigpoll, asked customers how they heard about the brand
immediately after purchase and matched each answer to the source the
store's own analytics recorded for that same order. 51.8% of orders
arrived with no usable source.

The comfortable explanation is that those customers cannot remember.
They can. Among untracked orders, 8.3% of customers said they did not
know how they had found the brand. Among tracked orders, the figure
was 7.3%. The two groups are equally certain; only one of them was
recorded.

What the untracked customers said is the finding marketers will
recognise. They named word of mouth 32% more often than tracked
customers did, and YouTube 40% more often. Measured from the channel's
side, a word-of-mouth sale goes untracked 58.7% of the time and a
YouTube sale 60.2%. A Facebook sale, which arrives with a click
identifier attached, goes untracked 40.1% of the time.

"Every brand has had the argument about what 'direct' means, and it
always ends the same way: someone says brand strength, someone says
broken tracking, and the meeting moves on because there was never a
way to settle it," said Jason Zigelbaum, founder of Zigpoll. "Now
there is, and the answer is worse than either. Those customers were
sent by a friend or a video. The report gave the credit to whatever
they clicked last. So when the budget review asks each channel to
prove itself, the channels creating the most demand are structurally
the least able to answer, and they are the ones that get cut. Word of
mouth doesn't get a seat in that meeting. It should."

The timing matters. Brands are setting 2027 budgets on last-click
reporting over the next eight weeks, and peak season is when the gap
between where demand is created and where it is captured gets widest.

Share of sales with no tracked source, by the channel the customer
named:

- YouTube — 60.2%
- Word of mouth — 58.7%
- Press and editorial — 56.4%
- TikTok — 56.3%
- Creator and influencer — 54.6%
- Podcast — 51.8%
- Instagram — 50.8%
- Search — 50.1%
- Facebook — 40.1%

The study also puts a first number on a channel that leaves no trace
at all. The share of shoppers naming an AI assistant as how they found
a brand rose from 0.23% in December 2025 to 1.10% in August 2026, a
4.8-fold increase in eight months and the fastest-moving line in the
dataset. It is still roughly one stated discovery in ninety. But a
shopper who asks an assistant and types the brand name two days later
arrives as "direct", which means the channel will grow inside the
blind spot rather than beside it.

The practical test is a cheap one. A brand that asks the customers in
its own direct bucket how they found it will get one of two answers.
If they cannot say, the tracking is fine and the bucket is brand
equity. If they say a friend or a video, the brand is allocating
budget from a report that cannot see the channel doing the work.

The Attribution Index is published quarterly, with tables for nine
retail verticals published alongside the aggregate at
https://www.zigpoll.com/attribution-index. All figures are free to
reproduce with attribution, and Zigpoll will provide channel or
category cuts to journalists on request.

Methodology: Findings are drawn from 3,303,402 responses to "how did
you hear about us" questions, collected between 18 September 2025 and
18 September 2026 across 3,237 ecommerce stores using Zigpoll. Only
responses attached to a completed order were included, so that a
customer's stated answer could be compared with the source recorded
for that same order. Responses were grouped into channels by an open,
rule-based mapping; answers that could not be confidently classified
were reported separately rather than assigned to a category, and
represent 12.7% of the total. No channel figure is reported unless at
least 50 distinct stores contributed to it. Time series use complete
calendar months only. Stores using Zigpoll are not a random sample of
ecommerce, and the figures describe that population rather than the
market as a whole.