Home / Self-Reported Attribution

Self-Reported Attribution: Why “How Did You Hear About Us?” Beats Your Analytics

Analytics tracks what it can see. An increasing share of B2B buying happens where it cannot: inside AI assistants, in private communities, on podcasts, in conversations. A single question on your booking form recovers most of it.

Why this matters more than it did two years ago

Self-reported attribution has always been useful. It has become necessary, because a genuinely new acquisition channel appeared and most analytics stacks cannot see it.

AI assistants now recommend vendors. Someone asks ChatGPT or Claude which agency to use, gets three names, and goes to one of those websites. Depending on how they arrive, that visit may land in your analytics as direct traffic, as a referral from a domain you do not recognise, or not at all.

We know this happens because we have asked. Four of our own sales calls in a two-week window came from ChatGPT, and only one of them was visible in the clickstream. The other three were invisible until somebody asked the prospect how they found us.

What your analytics genuinely cannot see

SourceVisible in analytics?Why
Google organicYesreferrer is preserved
Paid campaignsYesUTM parameters
ChatGPTSometimesappends utm_source=chatgpt.com on links it surfaces, but not on every path
Claude, Gemini, PerplexityRarelyoften arrives as direct or an unfamiliar referrer
Podcast or newsletter mentionNothe listener types your name
Private community or SlackNono referrer survives
Word of mouthNothere is nothing to track

Everything in the bottom half of that table arrives as "direct", which is analytics shorthand for "we do not know". For many B2B companies direct is the largest single bucket, and treating it as people typing the URL from memory is wishful.

The question to ask, and where to put it

Put it on the booking form or the demo request form, not on the first touch. You want it from people who converted, because those are the answers that matter.

Ask it as an open text field rather than a dropdown. A dropdown can only return options you thought of, and the entire point is to discover the ones you did not. Nobody would have had an "AI assistant" option in 2024.

The wording that works: "How did you first hear about us?" The word first matters, because without it people report the most recent touch rather than the one that started it.

Make it optional. A required field reduces completion, and across our panel 88% of lead form opens are already abandoned before you add anything.

Then ask the same question verbally on the call and record the answer, because the written one is often two words and the spoken one is a story.

How to bucket free text without losing the signal

Free text needs tidying, but tidy it late rather than early.

Keep the raw answer in a field that never gets overwritten. Add a second, normalised field for reporting. When somebody writes "chatgpt", "chat gpt", "ChatGPT comparison", or "asked AI", those are one bucket in the report and four different stories in the raw data, and the raw data is where you learn what prompt they ran.

Review the uncategorised pile monthly. New channels show up there first, which is the whole point.

Reconciling it with last-click

The two will disagree and that is expected. Last-click tells you the final step; self-reported tells you the cause. A buyer who first heard you on a podcast, searched your name three weeks later and booked, is a podcast lead in self-reported data and a branded search lead in analytics. Both are true. Only one tells you where to spend.

Use self-reported attribution to decide what to invest in and last-click to optimise the mechanics of what you already run. Trying to make them agree wastes time and usually ends with somebody deleting the more inconvenient one.

What it is worth

Concretely: it is the only way to know whether AI assistants are sending you buyers. There is no Claude webmaster tool, no Perplexity analytics, and ChatGPT’s UTM parameter only survives some paths. If you want to know whether the content you are publishing is being cited into a sale, the form field is the measurement instrument. There is not a second one.

It costs an afternoon. We have delayed it ourselves longer than we should have, which is how we know what the delay costs: every analysis of which channel produces revenue has an estimate in it where a number should be.

Frequently asked questions

What is self-reported attribution?

Asking buyers directly how they first heard about you, usually as an optional open-text field on a booking or demo form, and storing the answer alongside the deal. It captures channels that leave no digital trace.

Should the field be free text or a dropdown?

Free text. A dropdown only returns options you already thought of, and the purpose is to discover the ones you have not. Normalise into buckets at reporting time, keeping the raw answer intact.

How do I know if AI assistants are sending me leads?

Self-reported attribution is the only reliable method. ChatGPT appends utm_source=chatgpt.com to some links, but Claude, Gemini and Perplexity usually arrive as direct traffic, and no AI assistant provides analytics to the sites it cites.

Does self-reported attribution replace analytics?

No. Use self-reported data to decide where to invest and last-click data to optimise what you already run. They answer different questions and will disagree, which is normal.