How AI Search Changes Marketing Attribution

Why last-click attribution breaks in AI search

Last-click attribution gives credit to the final link a customer touched before converting. It worked because the path to a business used to be a chain of clicks, and the last one was usually a search result or an ad. AI answers cut that chain. A person asks Google’s AI Mode for a dentist who takes their insurance and gets three names with reasons. They pick one and call. No click happened. Your analytics either records nothing, or records a “direct” visit when they type your name later, or a branded search that looks like it came from nowhere.

The result is a quiet accounting error. Direct and branded traffic go up, organic sessions go flat or down, and the honest conclusion, that AI answers are now sending you customers, is invisible because the model you use cannot see it. Plenty of businesses are cutting the work that feeds AI answers because the report says organic is declining. The report is not lying. It is just measuring the wrong thing.

The zero-click customer: what actually happens

Walk through it from the customer’s side. They ask a question in plain language. The engine assembles an answer from sources it trusts, names one to three businesses, and often includes a phone number, hours, and a short reason. If the answer is good enough, the search is over. If they want to confirm, they search your name directly, look at reviews, and maybe visit your site for a minute before calling.

Notice what your tools see. Search Console shows an impression with no click, if it shows anything at all. GA4 sees a branded search or a direct visit. Your call tracking hears “I found you online.” The AI answer, the thing that actually made the decision, leaves no fingerprint in any of them. In our category study of which sources AI answers cite, brand-owned websites earned only about five percent of citations. The rest went to reviews, directories, news, and forums. So the decision is often being made from pages you do not even own.

Measure mentions, not just sessions

The practical answer is to add a mention layer on top of the session layer you already have. None of this requires new software, though some of it benefits from a schedule.

Impressions without clicks. In Search Console, watch queries where impressions rise and clicks do not. That pattern is what AI Overviews look like from the inside. The page is being read by the engine and quoted, and the customer is satisfied before they click. Treat those as a visibility win, not a CTR failure.

AI Overview presence. For your ten most important queries, check once a month whether an AI Overview appears and whether you are in it. Keep a simple sheet: query, overview present, you named, competitors named.

Scheduled answer checks. Ask ChatGPT, Perplexity, and Google’s AI the questions your customers ask, on a fixed schedule, and record whether you are named and what the engine says about you. Doing it once tells you nothing because the answers change. Doing it monthly gives you a trend. This is what we built the AI Citation Checker for, and the AI Search Visibility API if you want it automated.

Branded search lift. Branded query impressions in Search Console are the cleanest proxy for AI-driven demand a small business has. When AI answers start naming you, people search your name to confirm. If branded impressions climb while nothing else changed, something upstream is working.

Ask at intake. Put “How did you hear about us?” on every form and in every first call, with “an AI assistant like ChatGPT” as a listed option. It is crude and it is the only place the customer can tell you directly. We have watched that option go from zero to a steady share within a year for clients who track it.

A small-business attribution model for 2026

You do not need a multi-touch model with fractional credit. You need three layers that you can explain to yourself in a sentence each.

Layer one: search visibility. Positions and impressions for the queries that matter, from Search Console and a rank check. This tells you whether you are findable at all.

Layer two: AI answer presence. Whether ChatGPT, Perplexity, and Google’s AI name you for those same queries, and what they say. This tells you whether the engines trust you enough to recommend you. It is the layer most businesses are missing and the one that explains the gap between layer one and layer three.

Layer three: direct and branded demand. Branded search impressions, direct visits, calls, and the intake answer. This is where AI-driven customers actually land in your numbers.

Read them together. If layer one is strong and layer two is weak, you rank but the engines do not recommend you, which usually points to an identity problem: inconsistent name, category, or description across the sources the engines read. If layer two improves and layer three follows a few weeks later, you have your attribution, not as a decimal, but as a sequence you can see. That lag is real and it is the subject of how long it takes for AI to mention a brand.

What to report every month

One page. Top of the page: customers this month and where they said they came from. Middle: the three layers, each as one number and one direction, with a sentence of interpretation. Bottom: what you changed and what you expect it to move. If you already have the one-page SEO report, this is the same page with the AI answer row added.

What to leave off: session totals, bounce rate, and anything that goes up when you buy ads. Those numbers were useful when the click was the unit of attention. The unit now is the answer, and the only honest question is whether you are in it.