Definition
Self-reported attribution is a measurement method that asks customers directly how they heard about a company, usually with a 'How did you hear about us?' question at signup, checkout, demo request or onboarding. The answer is stored next to the tracked source, so marketers can compare what people say influenced them with what their analytics recorded.
The method is old, but AI assistants made it newly relevant. When ChatGPT, Claude or Gemini recommends a product, many people never click the citation. They type the brand name into a browser or search for it on Google, and the visit lands in Direct or Organic Search. Even Google's GA4 AI Assistant channel, added in May 2026, only catches visits that arrive with a recognizable referrer. The rest becomes AI dark traffic.
Two 2026 data points show the size of the gap. In a June 2026 study with Graphite, n8n found that GA4 last-touch attribution credited AI search with about 1% of conversions while its post-conversion survey credited about 9%, and that roughly 90% of those AI-influenced conversions never clicked a citation link (Graphite). Organic and paid search showed similar shares in both sources, which suggests the survey was not simply noisy. In October 2026 Ahrefs reported that ChatGPT and Claude together account for 21.7% of its signups across free and paid tools, ahead of YouTube and second only to Google, and said referral traffic alone would have 'badly undersold' AI (Ahrefs).
A workable setup has a few parts:
- Ask after the conversion, or as an optional field, so the question does not cost signups.
- Name the assistants individually (ChatGPT, Claude, Gemini, Perplexity, Copilot, Google AI Overviews or AI Mode) rather than one 'AI' option, and keep a free-text 'Other'.
- Allow more than one answer, since buyers often hear about a brand in several places.
- Optionally ask which prompt or question led them to you. Ahrefs recommends this because it shows which products and categories to invest in.
- Report the response rate next to every share, and keep the wording stable. Adding AI options mid-year creates a break in the series, not growth.
The weaknesses are well known. Memory is imperfect, people credit the last thing they remember, and many users say 'Google' when they mean an AI Overview. A pick-list also shapes the answer: an option that is not offered rarely gets chosen. Treat the share as a direction, not a precise count, and triangulate it with AI-referred traffic, branded search demand and visibility data.
For GEO teams, self-reported attribution closes the loop that visibility tracking opens. Prompt tracking shows whether and how assistants recommend you, through metrics such as AI share of voice. Survey answers show whether those recommendations turn into customers. Together they support a zero-click attribution model that credits AI influence without pretending every recommendation produces a click, and they fit into broader AI search analytics reporting.
Examples of Self-Reported Attribution
- A workflow automation startup adds ChatGPT, Claude and Perplexity to its post-signup survey and finds 8% of new users name an assistant, while GA4's AI Assistant channel credits less than 1% of signups.
- A skincare brand's post-purchase survey collects free-text answers like 'Gemini said it was fragrance-free', which the team classifies into channels with a language model and then uses to brief product page updates.
- A B2B sales team adds 'Where did you first hear about us?' to its discovery call script and logs answers in the CRM, where 'ChatGPT recommended you' becomes the third most common response in one quarter.
- A marketing lead adds AI assistant options to the signup form in Q2 and flags in the quarterly review that the AI share cannot be compared with Q1, when those options did not exist.
