TL;DR
Marketing attribution is having its quiet collapse moment. The dashboards still show numbers. The numbers still get reported up. The numbers are increasingly disconnected from what is actually happening upstream of the funnel. I stopped trusting my attribution reports about eighteen months ago, and I have been refining the replacement portfolio ever since. Below is what I watch now instead, and why each input does work the old attribution stack cannot.
I want to be careful with the headline. I am not arguing that attribution is useless. I am arguing that attribution is no longer doing the work most marketing leaders think it is doing.
The attribution model in most marketing stacks measures what happens once someone hits a measurable surface. A landing page. A form. A paid click. A trackable touch. The model assigns credit across the touches that contributed to the outcome.
That model worked when the funnel started somewhere measurable. In 2026, a meaningful share of funnels start inside an AI tool, in a conversation that does not produce a click and is not visible to your stack. The attribution report is faithfully crediting the touches it can see, while the first impression that actually shaped the decision is happening somewhere it cannot.
So the report keeps producing numbers. The numbers keep going up and to the right. The actual brand-shaping work is happening upstream of the dashboard, and the dashboard does not know.
Here is what I started watching instead.
One. AI citation rate, tracked quarterly
I covered this in detail in a previous post, so I will keep it short here. The AI citation rate is the percentage of relevant questions on which an AI tool names you in its response. It is the closest analog to an organic ranking that exists for AI surfaces.
If your attribution report is going up while your AI citation rate is going down, your business is taking a brand impairment you are not seeing yet. The attribution report will catch up to the citation rate eventually. The citation rate is the leading indicator. The attribution report is the trailing one.
Track it quarterly. Twenty questions, five tools, three categories. The methodology is fixed. The trend line is what matters.
Two. Branded search volume, tracked weekly
Branded search is the most boring metric on the list and the one I trust most. It is simple. It is hard to game. It is a clean signal of whether people are looking for you specifically.
The reason I weight it is that AI tools are increasingly funneling intent toward branded search. A prospect who reads about you in an AI response often types your name into a search bar to verify. That second step shows up in branded search volume.
A healthy marketing function shows branded search going up faster than category search. An unhealthy one shows the opposite, regardless of what the attribution dashboard says about lead volume.
Three. Direct traffic, tracked monthly, with a specific lens
Direct traffic used to be a junk drawer. It is now one of the most informative channels in the report.
Why? Because AI tools that do not pass referrer information are dumping their downstream traffic into the direct bucket. A growing direct traffic share in a market where AI usage is up is a positive signal. A shrinking direct traffic share is a warning.
The specific lens I apply is the direct traffic to product page ratio versus direct traffic to homepage ratio. Direct traffic to a product page is usually someone who got a specific recommendation. Direct traffic to a homepage is usually someone who knew your brand. Both matter. The product page direct traffic is the one I watch as a proxy for AI-driven discovery.
Four. Average time on a key page, tracked weekly
This is not a new metric. It is a metric I have started trusting more than the conversion rate on the same page.
The reason is that AI-referred traffic behaves differently than search-referred traffic. AI-referred traffic arrives with more intent and more specific questions, but it also arrives with less context. A higher time on page from this segment is a sign that the page is doing the work of giving them the context the AI did not provide. A lower time on page is a sign that the page is missing what they expected to find.
Conversion rate is the lagging metric. Time on page is the leading metric. I watch the leading metric weekly.
Five. Net new email subscribers from organic, tracked monthly
Email subscription is the most underrated growth metric in marketing. It is not because email is the strongest channel. It is because email subscription is the highest-intent micro-conversion that an organic visitor will make before they ever become a customer.
Net new email subscribers from organic, segmented by source, is a leading indicator of demand quality. It is a metric that paid media cannot fake, that AI traffic respects, and that the conversion-rate dashboard tends to ignore.
If the email list is not growing on organic strength, the brand is in slower trouble than the dashboard thinks.
What I stopped watching as a primary metric
I still look at multi-touch attribution reports. I do not act on them as a primary signal. I treat them the way I treat a weather forecast that is sometimes right.
I stopped acting on last-touch attribution years ago. I stopped acting on first-touch attribution about eighteen months ago. I stopped fully trusting multi-touch attribution about a year ago. None of these models can see the AI conversation that increasingly shapes the journey. None of them are getting better at it on the current vendor roadmaps.
This does not mean I throw the reports out. It means I demote them. They become one input among five, not the input.
How to talk about this with a CEO who reads the dashboard
The CEO is going to ask why the numbers in the dashboard are not the ones you are leading with. Have the answer ready.
The answer is that the dashboard measures the funnel you can see. The portfolio I just described measures the funnel you cannot. Both are real. The unseen funnel is growing as a share of the total. A marketing function that only reports the seen funnel is going to look fine on the dashboard and feel quietly worse in the business.
The CEO will respect this framing more than they will respect an argument with the attribution report. The conversation gets easier the second time you have it.
The bottom line
The attribution report is still a report. It is no longer the report.
Five inputs. Different cadences. One quarterly review. The marketing function that operates from this portfolio is going to see the shift coming six to nine months before the function that operates from the dashboard alone.
Pick two of the five to start watching this month. Add one per quarter. By the end of the year you will have a portfolio that does the work the old attribution stack was supposed to do and never quite did.
Kevin Farley writes about AI visibility, AI readiness, and strategic growth for financial services. Read more on the blog.