Search Didn’t Break. Your KPIs Did

Clicks are falling. That part is obvious. AI Overviews, AI Mode, ChatGPT Search and every other answer-first interface are making it harder for marketers to rely on the old flow of impression, click, session, conversion.

But that does not mean search has stopped creating value. It means the way many teams measure search has become badly outdated.

For years, Search has been one of the cleanest channels in digital marketing. You could point to ranking, CTR, CPC, CPA and ROAS and build a neat little story around performance. In an AI-native world, that story starts to break down. Users get answers without visiting your site. They compare products inside AI interfaces. They validate brands across multiple search environments before they ever click a result.

Behind all of that, the quality of your first-party data starts to matter a lot more, because AI search experiences and more agentic search experiences rely on clean, consented signals to learn which journeys actually lead to outcomes.

So the challenge is not simply that zero-click search is growing. The challenge is that too many brands are still using a 2016 measurement model to judge a 2026 search experience.

That is the real shift. Search value did not disappear. Our scorecards just stopped reflecting how modern search actually works.

Clicks are the exhaust, not the product.

There is now enough evidence to say this with confidence: when AI-generated experiences appear in search, clicks to traditional listings tend to fall materially.

One recent analysis found organic CTR for queries with AI Overviews dropped from 1.76% to 0.61%, while paid CTR fell from 19.7% to 6.34%.​ Other studies have estimated click reductions in the region of 47% to 58% when AI-generated summaries are present.

That sounds catastrophic if clicks are the only thing you care about.

It sounds different if you step back and look at user behaviour.

The point of search has never really been the click. The click was just the most visible signal we had. What search has always done is shape decisions. It helps people discover a problem, frame a shortlist, validate a brand, compare options and then, sometimes, take action. AI-native search does not remove those jobs. It compresses them.

In other words, the click is increasingly the exhaust of the journey, not the product itself.

That matters because if your dashboard still starts and ends with sessions, CTR and last-click conversion, search will look weaker at exactly the moment it is becoming more influential higher up the decision-making process.

Zero-click is not failure.

This is where a lot of commentary gets stuck. Zero-click gets framed as a loss by default. Sometimes it is. But not always.

If someone asks an AI surface a category question, sees your brand cited, later searches your brand name, comes back through paid search and converts a week later, which part of that journey mattered most? The click-based model will usually over-credit the last interaction and under-credit the discovery and validation stages that happened earlier.

That is why zero-click should not automatically be treated as failure. In many cases it is simply an unobserved or under-observed interaction. The user still learned something. The user still formed a preference. The brand may still have moved closer to conversion. The platform just did a worse job of handing us the evidence in the neat format marketers got used to.

This is exactly why Search teams need to stop asking only, “Did it generate a click?” and start asking, “Did it create visibility, influence or demand?”

The KPI stack needs a reset.

If clicks are no longer the cleanest proxy for value, then Search needs a broader KPI stack.

The first layer is visibility. Not simply rank, but presence across the modern search experience. Are you surfacing in AI Overviews? Are you being mentioned in AI assistants? Do you appear in the comparison moments that matter in your category? AI share of voice and share of SERP presence are becoming much more useful signals than a standalone average position report.

The second layer is influence. This is where brand search volume, share of search, assisted conversions and brand mention quality start to matter more. If your category visibility rises and your brand demand rises after it, that is a strong signal that search is still doing strategic work even when fewer users click the first interaction.

The third layer is engagement where it still happens. Clicks still matter. So do visits, leads and sales. But they need better segmentation. Teams should be comparing performance for queries with AI features versus those without, and separating informational, comparison and transactional journeys instead of rolling everything into one headline CTR trend.

The fourth layer is business outcome modelling. This is where many teams still have a gap. Dashboards can tell you what is happening. Modelling helps explain what it is worth.

Why MMM matters even more now.

This is where marketing mix modelling starts to move from “nice to have” to genuinely important.

AI search introduces blind spots everywhere;

• More answers happen without a site visit.
• More comparison happens inside platforms that do not hand over clean user-level data.
• More exposure takes place across fragmented environments, from Google AI Overviews to AI Mode to ChatGPT and other assistants.

That makes it harder to rely on deterministic click-path measurement alone. Not impossible, but incomplete.

MMM helps because it does not depend on observing every single click. It looks at the relationship between marketing activity and outcomes over time, then estimates contribution at a channel or tactic level.

Modern MMM works best when it is fuelled by robust first-party data such as conversions, revenue, CRM outcomes and product signals that the brand controls and can join back to media and search exposure in a privacy-safe way. As AI-assisted MMM matures, those first-party signals are increasingly blended with media inputs to capture how AI surfaces and more agentic search journeys influence real customers over weeks and months, not just in one click.

This becomes even more important as search moves into more agentic experiences. These systems do not just retrieve results; they are increasingly designed to optimise toward outcomes and learn from feedback. Those feedback loops depend on high-integrity signals such as who bought, what they bought and whether they returned, which almost always live in the first-party data stack.

In practical terms, that means MMM can help answer the questions that zero-click search makes harder to solve with platform reporting alone:

• If organic clicks are falling, is Search actually contributing less revenue, or is the value simply being distributed differently across the journey?
• When AI visibility improves, does branded demand rise later on?
• Is paid search becoming more efficient at closing demand that was created by AI surfaces, PR, social or TV?
• Where is the point of diminishing return once AI-powered interfaces start absorbing more of the early research stage?

This is the bit that matters for senior stakeholders. In an AI-native world, dashboards tell you what changed. MMM helps tell you whether it changed the business.

Modelling does not replace attribution. It fixes its blind spots.

There is a temptation to turn this into a false choice between attribution and MMM. It should not be.

Attribution is still useful for tactical optimisation. If you are running paid search, shopping or branded campaigns, you still need to know which queries, audiences, assets and landing pages are converting. That level of granularity remains valuable.

But attribution should now be treated as the zoomed-in layer, not the full truth. It shows the observable path. MMM gives you the macro view across the whole market and helps capture contribution even where user-level observation has broken down.

Both attribution and MMM become far more credible when they are anchored in the same first-party truth set: a clean, governed view of customers, transactions and lifecycle outcomes that AI systems can learn from and marketers can audit.

The best setup is not one or the other. It is a measurement system where:

• Attribution is used for in-platform and on-site optimisation.
• AI visibility metrics are used to monitor emerging presence across answer engines and AI-assisted search.
• MMM is used to validate the commercial impact of those changes over time, using first-party outcomes as the anchor so AI surfaces and Search do not get over- or under-credited based on partial click data alone.
• Experiments and incrementality tests use first-party conversion and revenue data as the ground truth, giving both human marketers and AI systems a reliable signal to optimise against.

That combination is far more resilient than trying to force every modern search interaction through a last-click lens.

What brands should measure now.

If Search leaders want a cleaner scorecard for this new environment, the shortlist should probably look something like this:

Measurement layer	What to track	Why it matters
Visibility	AI share of voice, share of SERP presence, presence in AI Overviews and assistants	Shows whether the brand is visible in the moments where discovery and evaluation now happen
Influence	Brand search growth, share of search, assisted conversions, mention quality in AI answers	Captures whether visibility is turning into preference and demand
Engagement	CTR, visits, conversion rate and CPA by query type and by AI vs non-AI environments	Keeps performance accountability, but with better context
Modelling	MMM contribution, incrementality tests, scenario planning, diminishing return curves, and the first-party outcome data those models run on	Connects Search activity to business impact even when direct click data is incomplete
Data spine	Quality and coverage of first-party data, including consented IDs, CRM match rates, server-side tracking and customer outcome depth	Determines how well AI Search, agentic experiences and models can learn from real outcomes rather than proxy metrics

This is also a useful way to reset conversations with CMOs. Search should no longer be judged purely as a traffic channel. It should be judged as a demand-shaping system that spans visibility, influence and conversion.

The reporting question has changed.

For years, the reporting question was fairly simple: how many clicks did search drive, and what did those clicks convert into?

Now the question is more demanding and, frankly, more interesting.

Where is the brand visible? What kind of demand is it creating? How does that demand move across channels? Which parts of the journey are observable, and which now need to be inferred through better models?

And underneath all of that sits a simpler question: does the brand actually own enough first-party data for its models, AI systems and measurement stack to learn anything useful in the first place?

That requires a more mature Search function. One that is comfortable with ambiguity, but not hand-wavy. One that can connect SERP presence, AI visibility, brand demand and econometric evidence into a more credible performance story.

Because the real risk for brands is not that AI search will destroy value.

The real risk is that marketers will keep using outdated metrics, misread what is happening, and cut investment in the very moments that are still shaping consumer choice.

Search did not break.

Your KPIs did.