Florian S. Kluge
Glossary
What is self-reported attribution?
Self-reported attribution means asking new users or customers how they found you, usually with a "How did you hear about us?" question at signup, and counting those answers alongside the click data your analytics records.
In plain terms
Analytics sees the visit that brought someone to your site. It doesn't see the ChatGPT answer, the Slack thread or the podcast that sent them there. A buyer who read about you in an AI answer and then typed your URL can show up as direct traffic, with no source identifying that answer. Asked, the same buyer might say "ChatGPT."
Tally's founders saw this gap in public. In a June 2025 update, they wrote that they were seeing over 2,000 new users signing up via AI tools, "and those are just the ones we could track." They said their onboarding survey showed the number was a lot higher.
Where it goes wrong
The answers are memories, not measurements. Researchers call the problem recall bias: people misremember or leave out details, and the error can be worse when the event was longer ago. Someone who heard about you twice may name only the most recent or most memorable source.
Your answer options also shape the answers. If "AI assistant" isn't on the list, nobody picks it. If "Google" is, you can't tell whether they searched a problem or searched your name after a colleague mentioned you.
How it complements click data
| Signal | What it sees | What it misses |
|---|---|---|
| Click data | The tracked visit before signup, with page and referrer | Anything that happened without a click |
| Self-reported answer | Where the buyer believes they found you | Exact pages, and anything they forgot |
Keep the two columns side by side. Never add them together: the same customer often appears in both. Store the answer on the account so you can see which answers turn into paying customers, not just signups.
Why it matters for a SaaS founder
The decision it informs is whether a channel your analytics can't see, such as AI answers or word of mouth, deserves budget. One question at signup costs almost nothing. If you get a handful of signups a week, read the answers yourself. A dashboard can wait. My article on whether SEO is bringing paying customers shows both signals in one report.
Example
Shipcheck is an invented 25-person CI testing tool. Its analytics shows 4 signups last month with a ChatGPT referrer. Its signup question shows 19 people who answered "ChatGPT or another AI assistant," and 6 of them now pay. The numbers are invented. In this example the survey captures more reported discovery, while the click data records specific visits. That won't always be the pattern.
Frequently asked questions
Should "How did you hear about us?" be open text or multiple choice?
A short list with an "Other" text field works well. The list makes answers countable. The text field catches sources you didn't think of, including the specific AI tool.
Can I add self-reported and tracked numbers together?
No. Many customers appear in both. Report them as separate columns and accept that the true number sits somewhere you can't see exactly.
Is self-reported attribution reliable enough to make decisions?
Reliable enough to spot a channel you were ignoring, not to calculate an exact return. Use it next to click data and billing, never alone.