Marketing Attribution for New Channels: A Practical Approach

Every few years a channel shows up that nobody can measure properly. Social was that. Podcasts were that. Right now it is AI search.

The pattern is always the same. The channel is clearly doing something, the reporting is terrible, and the people who wait for clean attribution show up three years late.

Marketing attribution for new channels is a solvable problem. It just is not solvable with the dashboard.

Why New Channels Break Attribution

Attribution systems assume a click with a referrer. New channels usually violate that in one of three ways.

The interaction happens off your site entirely. Somebody reads an AI answer that cites you and never clicks.

The referrer is stripped or wrong. A lot of app traffic and AI traffic lands as direct, which means your best performing channel shows up as an unknown.

The gap between exposure and action is long. Somebody hears you on a podcast in March and searches your name in July. No system connects those.

The Three Methods That Actually Work

Ask people. A single "how did you hear about us" field on your intake form outperforms every attribution model available for small businesses. It is imperfect and biased and it is still the best data you will get on channels that do not report.

Make it a free text field or a short list. Do not make it required.

Watch branded search volume. When a channel is working but not attributable, people search your name. Branded search in Search Console is the most reliable lagging indicator of unmeasurable marketing.

If branded queries climb after you start something new, that thing is working, whatever the dashboard says.

Run a holdout. Turn the channel off in one market or for one period and watch what happens. This is the only method that establishes causation, and almost nobody does it because turning off something that might be working feels dangerous.

Two weeks of honest data is worth more than a year of modeled attribution.

What I Use for AI Search Specifically

Direct traffic with no referrer that lands on deep pages. Somebody arriving directly at a specific blog post did not type that URL from memory.

Assistant referrals where they exist. Some AI platforms do pass a referrer, and it is worth segmenting them out even at low volume, because the trend matters more than the number.

The "how did you hear" answers. When people start writing "ChatGPT" in that field, you have your answer, and the volume there tends to lead the analytics by months.

I went deeper into which of these signals actually mean something in what actually matters about AI search, and why most of the surrounding advice is noise in GEO vs SEO.

The Mistake to Avoid

Do not build an elaborate measurement framework before the channel has volume.

I have watched teams spend a quarter designing attribution for a channel producing four leads a month. The correct instrument at that volume is asking the four people.

Build the measurement when the channel is big enough that the measurement changes a decision. Before that, a spreadsheet and a conversation is the right tool.

How to Talk About It With a Boss or Client

The honest framing is that you are running an experiment with a defined budget, a defined window, and a defined signal you will look at.

"We will spend this much for three months and judge it on branded search and intake form answers" is a defensible position. "We cannot measure it yet so we are not doing it" is how companies end up four years behind.

Set the timeline expectation up front too, because new channels almost always have long latency. The Latency Ladder exists for exactly this conversation.

For the underlying data on branded search behavior, Google Search Console remains free and is the only first party source for it.

When Should You Start Formally Measuring a New Channel?

When the channel produces enough volume that the measurement would change a decision. Below that, a spreadsheet and a conversation with the people who converted is the correct instrument, and building anything more elaborate is a way to feel busy.

The threshold I use is roughly thirty conversions a month. Under that, sample sizes are too small for any model to tell you something you could not learn by asking the customers directly. Over it, patterns become real and the investment in tracking pays back.

What to build first when you cross the threshold: consistent campaign tagging, a single source of truth for what counts as a lead, and the how-did-you-hear field preserved on every record. Those three cover most of what you need and none of them require a platform purchase.

What to avoid building: a multi-touch attribution model, at almost any small business scale. They are expensive, they encode assumptions nobody examines, and they produce confident numbers from data that cannot support them. I have never seen one change a decision at a company under a few hundred million in revenue.

The honest position with a client or a boss is that some channels are measured and some are inferred, and you will tell them which is which. That is more credible than a dashboard that assigns precise fractional credit to everything and quietly makes most of it up.

What is the channel you know is working but cannot prove? Mine was podcasts for about two years.