Does "How Did You Hear About Us" Actually Work?

The question analytics can't answer

Every attribution stack has the same blind spots: the podcast mention, the Slack recommendation, the conference conversation, the ChatGPT answer that named your product. No pixel sees any of it. And that's why the humble "how did you hear about us?" field keeps getting recommended as the fix, usually by people who then move on without explaining how to actually run one.

So does it work? Yes, with an asterisk the size of a billboard. Self-reported attribution catches channels nothing else can see, and it lies constantly. The teams that get value from it design for both facts at once.

Where people lie (not on purpose)

Memory is the problem. Someone hears about you on a podcast in March, sees your LinkedIn ad in April, and signs up in May. Ask them how they heard about you and they'll say "LinkedIn," because it's the touch they remember, or "Google," because that's where they typed your name. The podcast that started everything gets nothing.

The pattern is predictable: self-reported answers over-credit memorable, recent and searchable touches, and under-credit the quiet first ones. Big flashy brand moments get over-reported. Nobody ever writes "a comparison table I skimmed six weeks ago." So treat the field as a channel detector, not a precision instrument. It tells you a channel exists and roughly breathes. It doesn't rank your budget.

Designing the field

The implementation choices matter more than teams expect:

Free text beats a dropdown for discovery. A dropdown can only confirm channels you already suspected. The whole point is catching "my coworker Dana won't shut up about it" and "some AI tool recommended you." Run free text at least quarterly. If you need clean data for a dashboard, use a dropdown with an "other" text option, and actually read the others. The others are the entire payoff.

Put it on the highest-intent form. Demo request or signup, not the newsletter. You want it at the moment of conversion, and volume there is small enough that reading responses stays a coffee-break job, not a data project.

Make it optional, or expect fiction. Required fields get keyboard mashing. In our experience an optional field converts fine and answer quality jumps, because the people who answer chose to.

Store it next to the UTM data. The answer belongs on the contact record, in its own field, right beside the captured UTM parameters. The comparison between the two columns is where the insight lives. A form tool that captures both at submit makes this automatic.

The 90-day experiment

Here's the honest way to evaluate it. Run both systems side by side for 90 days: UTMs capturing what links measured, the field capturing what humans remember. Then put them in one table by contact.

Expect three buckets:

1. Agreement. utm_source=google, answer "Google." Comforting, tells you nothing new.
2. The field sees what links can't. UTM says direct, answer says "the Lenny podcast" or "a friend sent it." This bucket sizes your dark social and audio channels. For most B2B teams it's 20-40% of responses, which is the number that justifies the field's existence.
3. The links see what memory can't. UTM says the June newsletter, answer says "just Googled it." Memory compressed the story. This bucket teaches you the field's error rate.

The two systems disagree because they measure different layers: links record behavior, the field records what stuck in someone's head. Both are true. A channel that's big in bucket two deserves budget your click data would never approve.

One caveat generic advice skips: keep the wording stable. Changing "how did you hear about us" to "what brought you here today" mid-year changes the answers you get, and your trend line quietly becomes fiction. Pick a phrasing, freeze it, and let the data compound.

It works. Just never alone, and never as the tiebreaker against click data. It's the witness testimony to your UTMs' security footage: less trustworthy per data point, and the only source that saw certain rooms.