What did Refine Labs' 12-month study find?
Buyers and software gave very different answers. Refine Labs, a B2B demand generation agency, ran a 12-month test that captured both kinds of attribution for the same form submissions and published it as The Attribution Mirage, a post by CEO Megan Bowen (published 15 October 2024, updated 19 August 2026). The charts in the post label it a 2023 analysis of Refine Labs' own HubSpot data (the page's summary line says 2025), so these are Refine Labs' numbers about Refine Labs' own pipeline.
The method fits in one sentence. Refine Labs made a free-text "How did you hear about us?" field mandatory on its "Book a Strategy Call" form, with no leading text or suggestions, then reviewed each answer next to HubSpot's software attribution for the same submission.
| Measure (Refine Labs' data) | Software attribution (HubSpot) | Self-reported (buyers' own words) |
|---|---|---|
| Conversions credited to web search | 78% (direct and organic) | 12% |
| Conversions credited to dark social (social media, podcast, word of mouth, community) | Not given | 85% |
| Closed/won: web search (direct, organic, paid search) | 79% | 3% |
| Closed/won revenue credited to dark social | Not given | 98% |
| Podcast and community | Not measured at all | Named by buyers |
Refine Labs adds that buyers who mention its podcast convert to revenue at the highest rate. Three caveats come from the post itself:
- Percentages count mentions. Refine Labs divides each source's mentions by total mentions, and notes that one answer often names several sources.
- The samples get small. The text says 620 conversions were in the sample, but the two charts are labelled n = 128 high-intent leads and n = 30 high-intent revenue opportunities.
- Its channels shape its answers. Refine Labs says its go-to-market strategy is based on active social media and podcast presence, and social media and podcast were the two sources its buyers named most. A company that sells through partners or search may see a different split.
Why does software attribution credit search so often?
Software can only record the visits it can see. HubSpot's help page on Original and Latest traffic source properties (last updated 4 August 2026) says the Original Traffic Source property shows "the first known web source" through which a contact interacted with your business. The possible values include organic search, paid search, referrals, AI referrals, organic social, paid social and direct traffic. The same page warns that ad-blockers can block HubSpot's tracking cookies and affect those values.
Refine Labs walks through the gap with an example. A marketing leader asks peers in a Slack community what they use, gets three vendor names, searches Google for them, clicks two organic results and one ad, and books demos. The vendors' software credits SEO and paid ads. The buyer would say the Slack thread.
Refine Labs lists where these unrecorded conversations happen and calls them dark social:
- Social networks such as LinkedIn, Facebook, TikTok and Instagram
- Content platforms such as YouTube, Spotify and Apple Podcasts
- Communities and groups, including Slack communities and LinkedIn groups
- Internal company chat in Slack or Teams
- Direct word of mouth: phone calls, texts and DMs
- Meet-ups and events, in person or virtual
None of these leaves a click on your site until the buyer decides to visit. In Refine Labs' example, that visit came from a Google search for names a peer had suggested, which is why the free-text answer and the software source belong side by side. If founder posts on LinkedIn are part of how you sell, our guide to LinkedIn personal branding for founders covers the posting side.
How do you add a free-text "How did you hear about us?" field?
Add one required, open-text question to the form that signals buying intent, and keep the software source on the same record. These steps follow Refine Labs' method; the tool settings come from each vendor's help pages, checked on 20 October 2026.
Pick one high-intent form
Use your demo request form, booking page or signup form. Refine Labs used its "Book a Strategy Call" form. Start with one form so every answer means the same thing.
Ask in plain words, with no options
Use the question "How did you hear about us?" as an open text field. Refine Labs recommends no drop-down menus or suggestions, so buyers describe the source unaided. Leave the placeholder text empty too, so you don't hint at an answer.
Build it in HubSpot
HubSpot's form editor lets you drag in a field and connect it to a new property; multi-line text is a supported property type. Toggle on Required field so the form can't be submitted without an answer.
Or build it in Calendly
Calendly lets you add up to 10 custom questions per event type. Add the question in the event type's invitee form, tick Required and choose the Multiple lines answer type. Calendly saves the answers on its Calendar page, and you can export them from there.
Keep the software source next to it
Export the answer together with whatever your CRM or analytics recorded as the first source. In HubSpot that is Original Traffic Source and its two drill-down properties, which sit on the same contact record.
How do you code "How did you hear about us?" answers?
Give every answer one or more short codes from a fixed list, and keep the raw text. Refine Labs' charts group answers into organic or direct, social media, podcast, word of mouth, community and other or unknown. Our table below starts from those groups and splits out sources worth seeing separately; the codes and the five rules after it are our suggestion, not Refine Labs'. The example answers are illustrations, not real responses.
| Code | Use it when the answer names | Example answer (illustrative) | Follow-up worth asking |
|---|---|---|---|
| SEARCH | Google, Bing or "searched for" | Googled tools for this | What made you search? |
| AI-ASSISTANT | ChatGPT, Claude, Perplexity, Gemini | ChatGPT suggested you | What did you ask it? |
| SOCIAL-LINKEDIN | A LinkedIn post, comment or profile | Saw the founder's posts | Whose post? |
| SOCIAL-X | A post, reply or profile on X | A thread on X | Which account? |
| PODCAST-VIDEO | A podcast, episode or YouTube video | Heard you on a podcast | Which show? |
| WORD-OF-MOUTH | A colleague, friend, customer or investor | A friend uses you | Who referred you? |
| COMMUNITY | Slack, Discord, Reddit, forums, groups | Mentioned in a Slack group | Which community? |
| REVIEW-SITE | G2, Capterra or a comparison page | Read reviews | Which site? |
| EVENT | A conference, meetup or webinar | Met you at a meetup | Which event? |
| OUTBOUND | Your email, DM or call | You emailed me | Which message? |
| OWN-CONTENT | Your blog, newsletter or templates | Your blog post | Which page? |
| UNKNOWN | Blank, "web", "don't remember", junk | idk | Ask again on the call |
Five rules keep the codes useful:
- Allow several codes per answer. "Saw a LinkedIn post, then googled you" gets SOCIAL-LINKEDIN and SEARCH. Count responses and mentions separately, because Refine Labs' percentages are per mention.
- Never overwrite the raw answer. Store the code in its own column.
- Code once a week, in one sitting, with the same table, and add a code only when a source keeps appearing.
- Check each other. Once a month, have a second person code the same 20 answers and write a rule for every disagreement.
- Spot-check any AI coding. Refine Labs notes that free-form answers can now be classified automatically. If you let a model do it, give it this table and review a sample every week.
How do you compare self-reported and software attribution?
Put both on one row per conversion, then compare the totals once a month. Refine Labs recommends measuring attribution on two levels: software-based attribution for capturing demand, self-reported attribution for creating it. It calls the combination hybrid attribution.
Our suggested format is one sheet with these columns. The two rows are made-up examples to show the format:
| Date | Company | Answer (raw) | Code(s) | Software first source | Stage today |
|---|---|---|---|---|---|
| Example | Company A | Saw your post on LinkedIn, then searched | SOCIAL-LINKEDIN, SEARCH | Organic search | Demo held |
| Example | Company B | My co-founder recommended you | WORD-OF-MOUTH | Direct traffic | Proposal sent |
At month end, build a second table with one line per source group: the count and share of self-reported mentions, the count and share of conversions the software credited, and the gap between them. Look for three things:
- Sources only buyers name. Podcasts, communities and referrals with no software equivalent, as in Refine Labs' data.
- Search inflation. Conversions the software credits to organic search or direct traffic where the buyer names something else first.
- Stage quality. Which codes reach a proposal or a closed deal, once you have enough of them.
Outbound leads are easier to trace, because you can tag them when they are created. Siela's AI SDR saves each lead to its built-in CRM with the reason it was picked, and each run gets its own CRM group, so you can see which run found a lead, even before that lead fills in a form.
What can go wrong with self-reported answers?
Answers are memories, so treat them as evidence rather than records. Refine Labs describes the field as capturing how prospects recall becoming aware of a company. Watch for four problems:
- Last step remembered. A buyer who heard you on a podcast and later searched your name may write only "Google". Ask on the first call what made them search, and add the second code.
- Junk in a required field. A mandatory field can collect answers like "web", "online" or a few random letters. Code them UNKNOWN and track that share each month; if it climbs, reword the question.
- Small numbers. Refine Labs' closed/won chart rests on 30 opportunities. With a few dozen answers a month, one referral or one podcast episode can swing the percentages, so report counts next to every percentage.
- Leading the answer. A dropdown, a placeholder example or a list of channels next to the field suggests answers. Refine Labs avoided all of these.
Our advice: do not cut a channel because of one month's data from either method. Use the comparison to decide what to test next, then watch whether the answers move.
A 30-day plan to start self-reported attribution
A team of one to five people can run this in about an hour a week. The plan is our suggestion for applying Refine Labs' method; it is a method any team can run with the tools it already has.
Week 1: add the field
Add a required, open-text "How did you hear about us?" field to your highest-intent form or booking page. Check that the answer lands in your CRM next to the software source.
Week 2: code the first answers
Copy the coding table above, code every answer so far and write down any answer that didn't fit. Add codes only for sources that appear more than once.
Week 3: add the call question
On every first call, ask what made the buyer search for you or visit your site. Add the answer as a second code on the same row.
Week 4: compare and pick one test
Count mentions by code and conversions by software source. Write a one-page summary: top three self-reported sources, top three software sources, the biggest gap, and one channel to test next month.
Keep the field running after the first month. The comparison gets more useful as answers build up and as you can follow coded leads to closed deals.
Frequently asked questions
What is self-reported attribution?
Self-reported attribution is the source a buyer gives in their own words when you ask how they heard about you, usually in a free-text form field. Refine Labs pairs it with software attribution: software shows where a buyer first interacted with your website, while the buyer's answer shows where they first heard your name. Refine Labs calls the two together hybrid attribution.
Should "How did you hear about us?" be a required field?
Refine Labs made its field mandatory on its Book a Strategy Call form for its 12-month study. A required field can attract junk answers such as "web" or a few random letters, so code those as unknown and track their share. If the share keeps rising, reword the question or ask it on the first call instead.
Why not use a dropdown for "How did you hear about us?"
A dropdown only offers the channels you already expect, and it suggests answers to the buyer. Refine Labs recommends an open text field with no drop-down menus or suggestions, so buyers describe the source unaided. Free text also keeps detail a dropdown loses, such as which podcast, which colleague or which LinkedIn post.
What did buyers name most in Refine Labs' study?
In Refine Labs' chart of 128 high-intent leads, social media was the source buyers named most, followed by podcast. HubSpot's software attribution credited organic or direct traffic most often. Refine Labs says podcast and community are not measured at all by software attribution, and that buyers who mention its podcast convert to revenue at the highest rate.
Can AI code "How did you hear about us?" answers?
Yes, with checks. Refine Labs notes that free-form answers can now be classified automatically. Give the model a fixed coding table, ask for one or more codes per answer, keep the raw text in its own column, and have a person review a sample every week, because short answers like "Google" or "a friend" often need context from the sales call.
Sources
- Refine Labs: The Attribution Mirage (Megan Bowen) (published 15 Oct 2024, updated 19 Aug 2026)
- HubSpot Knowledge Base: Understand Original and Latest traffic source properties (updated 4 Aug 2026)
- HubSpot Knowledge Base: Create and customize forms (updated 9 Oct 2026)
- Calendly Help: How to add or remove invitee questions (updated 24 Aug 2026)