okaihub

Attribution for AI Founders: Where Did That Signup Come From?

Multi-touch attribution is overkill for a small team. The practical method for knowing which channel produced a signup, including AI-assistant traffic.

Published 2026-09-13 · 3 min read

Marketing attribution software solves a problem a pre-seed founder does not have: allocating credit across dozens of touchpoints and channels at scale. The problem a small founder actually has is simpler and harder — knowing which of the three or four things they are doing produced the people who signed up and stayed.

This is the practical version, which is mostly about setting things up before the traffic arrives rather than analysing it after.

Why simple attribution is enough

  • You have few channels. Three or four, not thirty. You do not need a model to allocate credit across them; you need to tell them apart.
  • Multi-touch modelling needs volume. Attribution models are statistical and require enough conversions to be meaningful. A pre-seed product does not have it, and the model output would be noise.
  • The decision is coarse. "Do more of channel A, stop channel B" does not require precise credit allocation. It requires knowing which channel produced activated users.

The three-layer method

  1. 1

    Layer one: UTM parameters where they survive

    Append utm_source, utm_medium and utm_campaign to every link you control — your site, your social profiles, your listings where the directory preserves them, your emails. This is the precise layer, and it works everywhere the link is passed through unmodified.

  2. 2

    Layer two: referrer where UTMs are stripped

    Many platforms and most AI assistants strip query parameters. For those, the HTTP referrer is the signal. Build a saved analytics segment grouping sessions by referring domain, and map each referrer to the channel it came from.

  3. 3

    Layer three: just ask

    "How did you hear about us?" as a single optional field at signup, or in your first conversation with a user. Crude, self-reported, and frequently more accurate than either technical layer for a small product.

Handling AI-assistant traffic

Traffic from ChatGPT, Perplexity and similar cannot be tagged, because you did not place the link and the citation strips parameters. Referrer is the only method:

  • Build a segment for the assistant referrer domains. chatgpt.com, perplexity.ai, claude.ai, copilot.microsoft.com, gemini.google.com and the others.
  • Expect undercounting. Some AI traffic arrives with no referrer at all and lands as direct. Watch the direct bucket for unexplained rises that correlate with being cited.
  • Track the landing pages it arrives on. Which pages AI traffic reaches tells you which content is being retrieved — that is the actionable output, because it shows what to write more of.
  • Do not confuse a crawler hit with a referral. A retrieval crawler fetching your page is a candidate event; a user arriving via a citation is the referral. They appear in different logs.

Attributing the channels you cannot tag

ChannelHow to attribute it
Community participationReferrer if the platform passes it; otherwise ask. Most community traffic arrives with no UTM
A newsletter mentionUsually a clean referrer. Ask the author to use a tagged link if they will
Word of mouth / directInvisible technically. The "how did you hear about us" field is the only route
A directory listingReferrer by domain; UTM only if the directory preserves your link parameters
Organic searchSearch console shows the queries and landing pages directly — the best-instrumented channel you have

What to actually record

  • Source of every signup, best-effort, from whichever layer caught it. A source on 70% of signups is useful; a perfect model on 5% of the volume is not.
  • Activation by source. The source that produces signups is not necessarily the one that produces activated users, and only the second matters.
  • A weekly count per channel, so you can see a trend rather than judging on a single day.
  • The questions people ask at signup, which reveal the channel and the intent better than any tag.

The trap: measuring what is easy

Analytics tools make it trivial to see pageviews, sessions and bounce rate, and hard to see which channel produced a paying user. Founders therefore report the easy numbers and mistake that for understanding. The discipline is to define the conversion you care about first, then build attribution backward from it — even if that means a manual spreadsheet and a signup question rather than a dashboard.

Attribution is not about perfectly crediting every touchpoint. At pre-seed it is about knowing which two or three things you are doing actually produce people who stay — so you can do more of those and stop the rest.

More in Measuring growth