Self-reported attribution: the only B2B attribution that survives 2026

TL;DR

Multi-touch attribution measures the channels that happen to be trackable. Self-reported attribution measures the channels that actually drove the decision. In 2026 those are not the same set.

• 70% of the B2B buyer journey is anonymous (6sense, 2025).
• 84% of content is shared through dark channels that produce no intent data.
• Chris Walker’s data across 40+ SaaS companies: what buyers report and what attribution software measures differ systematically.
• AI-engine referrals from ChatGPT, Claude and Perplexity often arrive as direct or unknown in GA4.
• The fix is a required free-text How did you hear about us field on every high-intent form. This article is the implementation playbook.

Introduction

The B2B marketing attribution conversation has been broken for five years. We kept fixing the tooling. We added more touchpoints to the multi-touch model. We bought Dreamdata, HockeyStack, Factors.ai. We argued about W-shaped versus U-shaped weighting at conferences.

Then 70% of the buyer journey became anonymous. Then 51% of software buyers started research in ChatGPT (G2, 2026). Then AI-engine referrals stopped passing referrer headers. The tool stack measures less of the journey every quarter, and the marketing team that depends on it makes worse budget decisions every quarter.

This article is the operator playbook for the alternative. Self-reported attribution is not new (Chris Walker has been arguing for it since 2022, 100+ B2B companies have implemented it since), but the implementation is still botched more often than not. The wording matters. The form placement matters. The analysis pipeline matters. The political work of getting your team to trust the data matters most.

This is for the CMO, RevOps lead or growth marketer who has read the same Chris Walker LinkedIn post three times and now wants the build.

Why multi-touch attribution stopped working

Three structural breaks compounded between 2021 and 2025.

The anonymous-journey break. 6sense’s B2B Buyer Experience Report 2025: 70% of the buyer journey is anonymous; 75% of buyers use backchannel research before contacting sales; 81% have decided before reaching out. Multi-touch attribution can only measure tracked touchpoints. If 70% of the journey is not tracked, the attribution model is, structurally, a model of the 30%.

The dark-social break. Sangram Vajre and Chris Walker popularised the term in 2021. The thesis has only sharpened. 84% of content is shared through DMs, Slack, podcasts and similar channels that do not produce intent data. The shares that drive vendor decisions are precisely the ones the tracker cannot see.

The AI-engine break. ChatGPT, Claude and Perplexity referrals often arrive in GA4 as direct or unknown. The buyer typed your URL into the address bar after reading the LLM answer. Multi-touch attribution credits Direct; the deciding channel was the LLM citation. In G2’s 2026 data, 51% of software buyers start research in an AI chatbot; that channel is invisible to multi-touch.

The combined effect is that multi-touch attribution increasingly tells a story about a shrinking subset of the journey. The story is internally consistent and externally wrong. As Adam Robinson of RB2B argues:

“Most marketers would say I am an idiot, but I do not believe in attribution. It forces you to focus on the wrong things.”

The position is provocative and the underlying observation is correct. Attribution that cannot see the deciding touchpoints forces budget toward the touchpoints it can see, which makes the deciding touchpoints even harder to fund.

What self-reported attribution is, exactly

Self-reported attribution is a measurement method in which the buyer answers a question (typically How did you hear about us?) on a form. The response becomes the primary attribution signal, captured in the CRM at the same first-class status as tracked attribution.

Three things define a real self-reported attribution implementation versus a half-built one.

The field is required. Not optional. The optional version has 12 to 25% response rate. The required version has 100% response rate. The data quality difference is not subtle.

The field is free text. Not a dropdown. Dropdowns bias the answer; the buyer picks the closest option even when the closest option is wrong. Free text captures the actual brand mention or peer name or podcast title.

The field is on the highest-intent form. Demo request, pricing inquiry, contact sales. Not the newsletter signup, not the gated PDF, not the webinar registration. The high-intent form is the one where the buyer has the patience and the motivation to answer well.

Implemented this way, the field captures the buyer’s perception of what actually drove the decision. Implemented any other way, it captures noise.

The question wording that gets honest answers

The wording matters more than any other implementation detail. The wrong wording produces nothing.

Good wording, in order of effectiveness:

“How did you first hear about us?”

“What is the main reason you reached out to us today?”

“Where did you first see Falora mentioned?”

Wording that fails:

“How did you find us?” (too vague; buyer types Google)

“Marketing source?” (jargon; buyer skips)

“Please select your attribution source:” (insulting; buyer leaves)

The best wording acknowledges the human reality: the buyer remembers a moment, a person, a brand mention. They do not remember a marketing source. The question should let them describe the moment.

Add a second optional question for the buyers who want to elaborate:

“Anything else you want us to know about how you discovered us?”

This optional second field captures the longer tail: the specific podcast episode, the LinkedIn poster, the peer in Slack, the ChatGPT query. The text in this field is often the most strategically valuable data in the form.

Where to put the field

The form placement matters almost as much as the wording.

On the demo-request form, after company and email, before scheduling. The buyer is committed and motivated. Response quality is highest.

On the pricing-inquiry form, after the requested usage volume. Similar logic; the buyer is committed.

On the contact-sales form, immediately after the message field. The buyer has just typed a paragraph; one more sentence is no friction.

Not on the newsletter signup. Low intent, low quality. Wastes the question.

Not on the gated PDF download. Low intent, mostly false answers.

Not on the webinar registration. Low intent and the answer is always Webinar.

Restrict the implementation to the three to five highest-intent forms. The data quality on those forms is what matters.

The categorisation pipeline

Free-text responses need to be aggregated to be useful. The categorisation pipeline is the unglamorous middle step that turns 200 free-text responses per quarter into a dashboard your CFO can act on.

Three categorisation approaches, in order of sophistication.

Manual weekly review. A RevOps analyst spends 30 minutes per week categorising new responses into a defined taxonomy (search, peer referral, podcast, community, LinkedIn organic, paid social, etc.). Works for volumes under 50 per week.

Keyword-rule auto-categorisation. A set of rules maps response text to categories. “Google” or “search” maps to Search. “Pavilion” or “Exit Five” maps to Community. “ChatGPT” or “Claude” or “AI” maps to LLM. A human reviews the unmatched 10 to 20% weekly. Works for volumes 50 to 500 per week.

LLM-assisted categorisation. A Claude or GPT call classifies each response into the defined taxonomy with confidence scores. A human reviews low-confidence cases. Works at any volume.

The taxonomy itself matters. Start with 8 to 12 categories. Add granularity over time. Do not start with 40 categories; you will spend the first quarter reorganising the taxonomy instead of using the data.

A useful starting taxonomy for B2B SaaS:

1. Search (organic)
2. AI engines (ChatGPT, Claude, Perplexity, Gemini)
3. Peer or word-of-mouth
4. Community (Pavilion, Exit Five, RevGenius, Slack groups)
5. Podcast
6. Third-party publication or newsletter
7. LinkedIn (organic or post)
8. Review site (G2, Capterra)
9. Event (in-person or virtual)
10. Paid advertising
11. Direct contact from founder or team
12. Other

How to use the data without breaking your team

The political work is harder than the technical work. The marketing team has spent two years building dashboards in Dreamdata or HockeyStack. The self-reported data will tell a different story. The team will feel exposed. Manage it.

Three rules for using self-reported attribution data without losing the team.

Frame it as additive, not replacement. Multi-touch attribution still has uses (campaign-level optimisation, A/B testing, sequence performance). Self-reported attribution answers a different question (where did the buyer first encounter us, what convinced them to take action). Run both. Use self-reported for strategic budget allocation, multi-touch for tactical optimisation.

Show the gap explicitly. Build a dashboard that shows tracked attribution and self-reported attribution side by side, by channel, by quarter. Let the team see the gap. The gap itself is the most valuable insight.

Move budget gradually. If self-reported attribution shows Podcasts at 22% of high-intent conversions and tracked attribution shows Podcasts at 3%, do not immediately move 19 points of budget. Move 5 points, measure, iterate. The team needs to see the new allocation work before they trust the methodology.

The end-state is a quarterly board review that includes both attribution views and explicit acknowledgement of the gap. The CMO who runs that review has more credibility with the CFO, not less.

What the data typically shows in B2B SaaS

Across the deployments we have observed, three patterns recur.

Tracked attribution overcounts paid and direct. Both channels are the residual when everything else is unidentifiable. Self-reported attribution typically shrinks both by 30 to 50% and reveals the actual upstream drivers.

Tracked attribution undercounts community and word-of-mouth by 5 to 10x. Buyers report peer referral and community-driven discovery at rates that tracked attribution never sees, because the deciding touch happened in a Slack DM or Pavilion conversation.

ChatGPT, Claude and Perplexity now appear at 8 to 22% of self-reported attribution in the deployments we measured in Q1 and Q2 2026. Two years ago this number was 0%. The discovery-shift to AI engines is real and self-reported attribution is currently the cleanest way to measure it. This is why GEO investment compounds.

A 14-day implementation plan

For a team starting from a multi-touch-only baseline, here is the 14-day implementation.

Days 1 to 3: form audit and field design. Identify your 3 to 5 highest-intent forms. Design the question wording. Design the optional follow-up field. Build the CRM field as a first-class attribution source.

Days 4 to 7: deployment. Add the field to all selected forms. Test the form submission flow end-to-end. Ensure the response is written to the CRM correctly and is visible in your reporting tools.

Days 8 to 10: baseline data and taxonomy. Start collecting responses. Build the 8 to 12 category taxonomy. Decide on the categorisation method (manual, rule-based, or LLM-assisted).

Days 11 to 14: dashboard. Build the tracked-vs-reported comparison dashboard. Share with the marketing and revenue team. Document the methodology in writing.

By day 14 you have working self-reported attribution. The strategic insights start compounding from quarter two onwards as the response volume grows.

Frequently asked questions

Does self-reported attribution work for enterprise sales? Yes, with a small adjustment. The first sales call captures the response (the AE asks the question and types the answer into the CRM). The form-based capture supplements with self-serve inbound. The two together produce a complete picture.

What about the legal-basis question under GDPR? A self-identified attribution response on a B2B demo form is legitimate-interest data with explicit context. No additional consent is required beyond the standard demo-form consent. The data is also high-utility, which strengthens the LIA.

What if the buyer answers I do not remember? Capture it as Unknown and let it be a category. The Unknown rate is informative; it should be 5 to 15% in a healthy implementation. A 30%+ Unknown rate suggests the question wording is wrong.

Should I show the response options to the buyer? No. The dropdown bias is the single biggest implementation failure. Always free text. Categorise on the back end.

Can I trust self-reported attribution from anonymous form fills? Yes, with sample-size caveats. Individual responses are noisy; the aggregated quarterly pattern is reliable.

Conclusion

The attribution conversation is not a tooling problem. It is a measurement-philosophy problem. Multi-touch attribution measures the channels it can see. Self-reported attribution measures the channels that actually drove the decision. In 2026 those are not the same set, and the gap is growing every quarter.

The implementation is two weeks of work. The strategic payoff is a budget allocation that compounds with the buyer behaviour shift instead of against it.

If you want help designing the question wording, the categorisation pipeline and the dashboard for your specific motion, book a 45-minute attribution review with Falora.

Sources

• 6sense, B2B Buyer Experience Report 2025
• G2, 2026 Software Buyer Behavior Report
• Chris Walker on LinkedIn (self-reported attribution thesis, 2022-2026)
• Geisheker, Is Marketing Attribution Dead? Dark Funnel 2026
• Cognism, Dark Social Explained
• Sangram Vajre, GTM Partners. The GTM Operating System
• Adam Robinson, RB2B

Related reading on Falora

• GEO for B2B SaaS: how to get cited by ChatGPT
• The signal-based selling playbook
• The anatomy of a GTM engineering system
• The outbound agency cost autopsy
• GTM engineering vs growth marketing

About the author

Jeroen De Broyer is co-founder of Falora. He has implemented self-reported attribution at scale in 11 European B2B SaaS scale-ups and has spent the last decade reconciling tracked and reported data as a sales operator. He writes on LinkedIn.