Attribution in a changing user journey.

Marketing budgets are under greater scrutiny, and the expectation to prove commercial return keeps rising. According to NIQ’s 2026 Outlook, 84% of CMOs cite return on marketing investment as their primary metric for budget allocation. Senior marketers are being asked to show a clearer line from spend to revenue, often with less budget available and less tolerance for investment that can’t be justified.

The problem is that proving that return is getting harder. Attribution worked better when more of the user journey was visible – when people clicked, landed, browsed and converted in ways that left a traceable trail. That world is receding. Journeys are fragmenting, lengthening, and moving into places attribution models were never built to reach.

So the pressure is moving in opposite directions: marketers are being asked for more revenue accountability at the same time as their ability to prove contribution is becoming less complete. The answer isn’t to abandon attribution. It’s to understand what it can and can’t tell you, combine it with other signals, and keep budget decisions anchored to the commercial outcomes that matter.

Attribution has always been messy because buying decisions aren’t linear. Last-click dominates because it’s easy, not because it’s accurate, and the reasons for imperfect data are well documented – declining third-party cookies, cross-device behaviour, untrackable touchpoints, etc. Compounding that, platforms self-report performance in ways that favour their own contribution rather than necessarily reflecting what actually drove the sale.

None of this makes attribution useless. Imperfect signals still have value – the mistake is treating them as fact and allocating budget accordingly.

On top of an already difficult problem, something more structural is changing: AI tools are fundamentally reshaping how people discover, evaluate, and decide.

Our SEO in an AI World guide examines this in depth, but the data is clear. Organic search clicks have fallen 42% since Google began expanding AI Overviews, and when a search result includes an AI Overview, only 8% of users click through to a source. People are getting answers without leaving the platform, and without generating any signal your attribution model can read.

At the awareness stage, AI-generated summaries and recommendations are replacing traditional discovery channels. A user asking ChatGPT or Claude which company sells the best coffee beans will receive a synthesised answer, possibly including your name, without ever visiting your website. That exposure, which may directly influence their next move, is often invisible to your analytics. Discovery is also fragmenting across channels. Instagram and TikTok now function as genuine search tools, particularly for younger demographics. Travel decisions are shaped by what people find on YouTube or Reels before a Google search is ever triggered. The starting point has shifted, and in many cases it’s somewhere no UTM can follow.

In the consideration stage, the compression is more significant. Users are synthesising information through AI rather than visiting multiple brand touchpoints. The research and comparison phase that used to leave a trail of clicks and sessions is increasingly happening inside an LLM conversation. According to a recent report, buyers are already 70% of the way through their journey by the time they engage with a brand directly. That 70% doesn’t happen on your website. It happens in the parts of the journey that are often invisible: the dark social, peer conversations, and AI-assisted research, out-of-home campaigns. LLMs are now a significant part of that invisible middle, with 94% of buyers already using them as part of their buying journey (6Sense).

At the decision stage, AI tools are influencing final choices with zero visibility to the marketer. No click. No cookie. No attribution. As Adobe has explored in its own analysis of LLM-driven customer experiences, the challenge isn’t just about tracking, it’s about recognising that the AI layer is now a genuine decision-making environment, not just a research aid.

What limited signals do exist – referral traffic from AI tools where it surfaces, branded search trends, share of voice in AI responses tracked through third-party tools – are real, but immature.

“In LLM’s, there’s no impression data, no query data, just the clicks you happen to catch and third-party tools that show you whether your brand appeared for prompts you chose to track. Until the tooling catches up, we’re inferring more than we’re measuring. A meaningful and growing share of the journey now happens somewhere no attribution model can reach. Accepting that is the starting point.”

Francesca Hume, Search Lead, Invanity

The answer isn’t paralysis. It isn’t false precision either. It’s a smarter approach. One that holds attribution data loosely, triangulates across multiple signals, and keeps attention on the outcomes that matter most.

Forbes has explored this shift and the consensus is consistent: the marketers navigating this well are the ones who’ve stopped treating attribution as a truth machine and started using it as one lens among several.

Don’t chase perfect. No attribution model captures the full picture, and a perfect attribution model will never arrive – channels, tactics and preferences change too quickly. But don’t let that stop you from making progress. Better decisions will still be made with inferred attribution than with no attribution at all.

Use multiple lenses. Blend last-click or data-driven attribution with incrementality testing – controlled experiments that isolate the actual impact of a channel rather than relying on self-reported platform data. Use share of search as a proxy for brand impact. Track branded search volume over time as an indicator of AI-driven awareness that hasn’t yet translated into a visible touchpoint.

Think in contribution, not causation. Reframe attribution as understanding which channels are contributing to a journey, not which one caused a conversion. That reframe changes the questions you ask and makes the data more useful, rather than forcing it to answer something it can’t.

Track what you can, infer the rest. Maintain rigorous UTM discipline. Invest in first-party data collection. But build in headroom for untracked influence: dark social, AI referrals, word of mouth, out-of-home, etc. The headroom isn’t a failure of the model; it’s honesty about how decisions are actually made.

Anchor to business outcomes. When attribution gets noisy, ground budget decisions in the commercial outcomes that matter – revenue, pipeline, sales, customer acquisition and profitability – rather than platform-level metrics. Track spend, track revenue return, and agree on a framework for understanding what’s happening in between those two points. Agree on that framework at the start of the engagement and stick with it. Even if it’s not perfect, comparing apples with apples consistently is better than chasing ever-changing precision.

Attribution is difficult. It’s getting harder. The AI layer has added a significant new dimension to an already complex problem, one the industry’s tools are not yet equipped to solve.

But the answer isn’t to stop measuring. It’s to measure more honestly: with clearer eyes about what the data can and can’t tell you, with multiple signals rather than one, and with business outcomes as the north star when the signals disagree.

The marketers who navigate this best won’t be the ones with the cleanest attribution model. They’ll be the ones who understand its limits, build accordingly, and stay focused on what’s actually driving growth, even when they can’t prove it with a click.