AI search has an attribution problem

Look, in my opinion, attribution doesn't really matter all that much in the first place, especially if you're doing anything B2B. The buyer's journey is so non-linear that it just doesn't really matter. Someone hears about you on a podcast, sees you in a newsletter weeks later, asks a peer in Slack, then finally lands on your site by Googling your brand. Good luck assigning credit to one of those.

But let's assume for a second that it did matter to you. Maybe you're in consumer. Maybe your CMO is on your ass about this stuff because your CEO is on their ass about increasing attributed MQLs. Fair enough I suppose. Let's play it out.

AI search is the newest frontier for organic lead gen, and a top 5 hot startup space right now. And as more and more of the buyer's evaluation process happens in chat itself, we give way to making revenue less and less attributable. A few reasons.

AI referred traffic

AI referred traffic is flawed. It's traffic where the utm_source parameter is something like chatgpt, anthropic, perplexity, and so on. But what if the LLM doesn't cite anything? What if your buyers don't click citations?

Think about your own LLM usage. How often are you actually clicking the links they provide? It's kind of a pain. You have to click sources, then click the link you care about, then wait for it to load. That's a lot of friction for something you could just look up yourself.

No, that's not what you do. Instead you see the name of the thing you want to research and do one of two things:

1. Go straight to the website (like maintouch.com) ← this is direct traffic.
2. Look the company up separately ← this is branded search.

So now we're in an interesting spot. For most brands, when you ramp SEO and AEO activity, you should expect an increase in branded and direct traffic. If you anchor on AI referred traffic, which is what will get used in last touch attribution, you screw yourself.

But what happens if you have other stuff going on? Maybe you got some good PR. Maybe you went viral on a Reddit thread. Maybe you did an out-of-home campaign. All of these things will also increase direct and branded traffic, which muddies the water even further. You now have five things all pushing the same two buckets, and no clean way to split them apart.

This means last touch attribution will be all wonky.

Self reported attribution

This is the classic "how did you hear about us?" But again, the buyer's journey isn't linear, and in this case the narrator is also super unreliable. If I spend 30 minutes chatting with a coding agent picking the best tool for a specific part of the codebase, but then go on LinkedIn and see an ad for you and decide to sign up, I might say social brought me there. That's a misattribution, and it happens constantly. People report the last thing they remember, not the thing that actually moved them. And you can't really survey your way out of this.

But doesn't MMM solve this?

Kind of. Marketing mix modeling is the actual grown-up answer to a lot of this. Instead of tracing one person's path, it works top down: it models how much your aggregate activity drives your aggregate revenue. It doesn't care whether anyone clicked a citation, it doesn't ask your buyers to remember anything, and it doesn't fall apart when cookies and UTM params do. For the two problems above, that's a real step up.

But it doesn't solve all of it, and the reasons rhyme with the earlier ones. MMM is only as good as the variation in its inputs. So if your AEO, PR, and brand campaign all move together, it hits the same wall last touch did: it can't pull them apart. Collinearity doesn't care how fancy your math is.

It's also data hungry, which punishes exactly the companies reading this. AEO is new, so there's barely any history to feed it, and "spend" on AEO isn't a clean number like a paid channel, so even the input is fuzzy. SEO and AEO are so delayed that it might tell you next quarter that organic worked, not whether last week's post is landing.

Okay so what do we do?

First things first, you need to remove yourself from hoping for perfect attribution. It's not coming. The channels that matter most are getting harder to see.

Practically, when evaluating SEO and AEO, you should be triangulating across many metrics:

• Clicks and impressions (branded and non-branded)
• AI referred traffic
• Self reported lead sources

Naturally some of these are upstream of others. Impressions move before clicks, clicks before branded search, and so on. But if some or all of them are trending up together, you're likely in a good spot. Stop trying to draw a straight line from one chat session to one closed deal. Watch the whole system move instead.