Plenty of guides tell you to add a "How did you hear about us?" question to your forms. Far fewer show you what comes back, and that gap matters, because the answers arrive as messy human sentences, not tidy data.
This guide walks through the kinds of answers real forms produce, how to sort them into channels, and what to do with the vague ones. Every example below is invented for illustration, but each one is typical of a pattern that appears on almost every B2B form.
What the Answers Look Like
The first surprise for most teams is how specific people are. Given an open text box, buyers name colleagues, podcasts, communities and posts. The second surprise is the mess: typos, half-memories and one-word replies live alongside the gold. Both are normal. Here is a set of invented but typical examples with the channel each would belong to.
Channel Word of mouth
Also worth noting the buyer changed companies: your product travelled with them
Channel Podcast
Vague on the show, clear on the channel. Classify the channel, park the show
Channel Organic search
Genuine search-led discovery. The comparison habit is useful context for sales
Channel LinkedIn organic
Names the platform and the topic. Not the same channel as a LinkedIn ad
Channel Word of mouth
Private share. This journey was invisible to analytics from start to finish
Channel Unclassifiable
Keep it in the data as its own category. The share of these tells you how early to ask
A dropdown forces buyers into your list of channels. An open text box lets them tell you about channels you did not know you had.
Sorting Answers Into Channels
Classification is simple to describe and easy to do badly. Three habits keep it honest.
Fix the channel list first. Decide your channels before you start reading, and write one line describing what belongs in each. Without the written list, the classifier drifts: the same answer lands in different channels depending on the day and the person sorting.
Classify what is written, not what you hope. The temptation is to nudge ambiguous answers toward whichever channel you are currently excited about. Resist it, or the data becomes a mirror. When an answer genuinely supports two readings, mark it unclassifiable rather than guessing.
Keep the original text. The channel label is for counting; the sentence is for learning. "My old boss used you at her last company" classified as word of mouth still carries a second insight: your product travels with people between jobs. Throw away the text and you keep the count but lose the stories.
By hand, in a spreadsheet, this is entirely doable at low volume and a genuinely good way to learn your own patterns. As volume grows, hand-sorting becomes the bottleneck, which is where automated classification with human review of the uncertain cases earns its keep. Either way, the method is the same one described in our complete HDYHAU guide.
The Mess Is the Message
Teams sometimes see the typos and half-memories and conclude survey data is too unreliable to use. But the mess is what honest recall looks like. Inside it are the names of the colleagues, shows and communities actually driving your pipeline.
Clean data that misses the real channels is worth less than messy data that names them.
Frequently Asked Questions
How do you categorise HDYHAU survey responses?
Define a short list of channels that matter to your business, usually eight to fifteen, such as word of mouth, podcast, organic search, LinkedIn, events and communities. Read each answer and assign the channel the buyer is actually describing, keeping a category for unclassifiable answers. Consistency matters more than the exact list: the same kind of answer must land in the same channel every time.
What do you do with vague answers like "online" or "Google"?
"Google" usually means organic search, but read it in context: some people say Google when they mean an ad, others when a colleague sent them a link they then searched for. Classify what you can defend, put the rest in an unclassifiable category, and track that category's share. If it grows, your question needs better timing or wording.
How many responses do you need before the data is useful?
Patterns usually emerge sooner than teams expect. Even a few dozen answers will show your two or three dominant channels clearly. What needs more volume is trend analysis, comparing quarters or campaigns. Start classifying from the first answer and let the confidence build with the count.
Should one answer be allowed to name two channels?
Yes, because journeys genuinely span channels: "heard the podcast, then a colleague confirmed you were good" is two real influences. Record both rather than forcing a choice. The first mention is usually the origin story; the second is the validation that closed the gap.
James Kevan is the co-founder of First Signals, which classifies open text attribution answers automatically and routes the uncertain ones to a human, so the mess becomes a channel report you can trust.
Related guides: HDYHAU Surveys for B2B · How to Measure Word of Mouth · The B2B Dark Funnel
© 2026 James Kevan / firstsignals.ai. Share freely with attribution.