Is asking 'how did you hear about us' reliable?

Partly, and in a specific way: it tends to surface channels tracking cannot see — private recommendations, communities, AI assistants — while over-crediting whatever is memorable or recent. Both are tendencies rather than laws. The method is best treated as testimony: genuinely informative, systematically biased, and strongest when read next to software attribution rather than instead of it.

Should the attribution field be open text or a dropdown?

Open text generally produces truer answers, because dropdowns inflate the options they list and erase the ones they omit. The cost is categorization work afterward. A common middle path is open text on the form plus a probing question on the first call, with raw answers preserved so they can be recategorized as new channels emerge.

What is the difference between self-reported and software attribution?

Software attribution records clicks with precision and stays silent about untracked influence; self-report covers all influence with the imprecision of human memory. They fail in opposite directions, which is why the useful move is comparison: agreement raises confidence, and divergence points at dark channels worth investigating.

What should we do when buyers answer with an AI assistant's name?

Take it seriously as a channel sighting — analytics undercounts AI-driven discovery for structural reasons, so the form is often where that influence first becomes visible. Keep the raw answers, watch whether the mentions grow, and consider measuring the other end of the event: what the assistants are actually saying about your category when buyers ask.

What is self-reported attribution

Guide · Glossary & Definitions · 4 min read · last verified 2026-07-28

Reviewed before publication Editorial board Independent commercial review

In shortSelf-reported attribution asks buyers directly where they found you. It hears channels software never sees and flatters the memorable at the same time — both tendencies, not laws. How to design the field and read the answers.

Self-reported attribution means asking buyers directly where they found you — an open question on a signup form, a line in a demo request, or a question posed at the start of a first call — and treating their answers as evidence about which channels bring you business. It is the human-memory counterpart to software-based tracking, and it hears what no tracking pixel can record.

The channels only a human can report

Click-based tracking records what happens inside instrumented sessions and stays silent about everything else. A meaningful share of business-to-business discovery appears to happen in that silence: a name dropped in a meeting, a recommendation traded in a private community, a link pasted into a group chat — the territory usually called dark social — and, most recently, an AI assistant naming a vendor inside an answer. When a buyer types a colleague kept mentioning you into a form field, that sentence often carries channel information no analytics platform possesses. Asking is currently among the few methods that surface these paths at all, which is the strongest argument for the practice: not that memory is accurate, but that the alternative to an imperfect witness is frequently no witness.

Memory is a biased witness

The same answers over-credit whatever is easy to recall. Buyers tend to compress a months-long journey into the one source they remember best — often the most recent touch, the most distinctive one, or simply the easiest to spell. A podcast with a memorable host may collect credit that a quietly effective piece of search content deserved a share of; the reverse can happen too, and the direction of the distortion is hard to know from inside the data. Both properties of the method — it hears the invisible, it flatters the memorable — are tendencies rather than laws. Treating self-report as reliable is a mistake; treating it as worthless is the same mistake in the other direction.

The field itself is a research decision

How you ask shapes what you learn. A dropdown menu inflates whatever options it lists and erases whatever it omits — buyers pick the nearest plausible item rather than fighting the form — while an open text box yields messier answers that are usually closer to what actually happened, at the cost of categorization work afterward. Placement matters as well: a question asked after the buyer has committed to signing up usually draws more considered answers than one adding friction mid-form. Even the wording nudges outcomes, since asking where did you first hear about us and what made you sign up today are different questions that buyers often answer differently. None of this is a reason for paralysis; it is a reason to treat the field as designed research rather than form furniture, and to keep the design stable once chosen so answers stay comparable over time.

Two flawed witnesses beat one

Software-based attribution is precise about clicks and silent about everything unclicked; self-report speaks to everything and is precise about nothing. Read together, they check each other. Where both point at the same channel, confidence is warranted. Where they diverge, the divergence itself is the finding: a channel that looms large in self-report while barely registering in tracking is a candidate dark channel worth investigating, not a data-quality embarrassment to reconcile away. Teams that keep the two systems side by side, without forcing agreement, assemble a fuller channel picture than teams that crown either one.

When the form says the name of an assistant

Assistant names — ChatGPT, Perplexity, and their peers — have begun appearing in how-did-you-hear fields, and for many teams that field is currently the clearest sighting of AI influence they get, because AI referral traffic is undercounted in analytics for structural reasons. The behaviour is present-tense and unstable: which assistants buyers consult, and whether those consultations produce measurable clicks, both keep shifting as the engines change. Self-report catches this influence after it has already happened. Measuring what assistants actually say about your category addresses the step before, while the recommendation is being formed — that is the ground Magrios covers, capturing what the assistants answer so the form field's testimony can be checked against what buyers were being told at the time. The two views describe opposite ends of the same event, and neither substitutes for the other.

Making the answers usable

A few working practices keep the method honest. Preserve the raw text forever and categorize on a regular cadence, so recategorization stays possible as new channels emerge. Resist collapsing vague answers — a buyer who writes Google might mean an ad, an organic result, or an AI-generated overview, and only a follow-up question distinguishes them. Ask again on the first sales call, where a conversation can probe the journey a form cannot. And report the results as what they are: testimony, valuable and biased at once, strongest when read alongside the tracking data it disagrees with.