Search historically offered a reasonable deal. You had a person searching, a results page popped up, a link was clicked and a site visited. Then the conversion, which may or may not have happened. The advertisers received the signals and while this measurement was never perfect, with look-back window differences or recency tossing a curveball, the actual journey had enough actionable data to support a decision. A business decision. Now this deal is breaking down.
People are no longer only entering “keyword” but asking full questions of LLMs. They’re adding context and following up on the answer, and the AI system does not retrieve a list of possible destinations but reads across them, synthesizes information and returns a response.
The search experience has changed from finding a source to getting an answer but the tech stack for measurement still relies on the same set of signals and assumptions, chief among them that value begins when a site is visited.
What happens when a simple query is no longer a reliable unit of intent
We can all agree the traditional query was imperfect but it was useful. “Best running shoes” could be classified, priced, compared and reported. Search teams could see the term, the ad, the click and the landing, and SEO teams could see an impression or a position and a visit. The query was a solid proxy for intent.
Conversational search is a different beast where a person may begin by asking for running shoes for heel pain, then refine by surface and brand, then reject a few recommendations based on price and finally ask which remaining options have the best return policy. This is not five isolated searches but one decision developing through a series of questions or a conversation.
But reporting does not see it this way. Google’s own Search Console documentation says that when a user asks a follow-up in AI Mode, it’s counted as a new query. This is consequential. The platform records another impression, click, position, etc. The person experiences continuity but the current measurement records separate events and the meaning sits in between the prompts.
Intent is being refined throughout the conversation while reporting continues to count its parts individually.
AI answers are creating influence without clicks or visits
The click has always been an incomplete measure of influence. We in PPC have argued for years that optimization of clicks or high CTRs is diametrically opposed to actual conversions. Now AI search is making this fact impossible to ignore.
Pew Research Center analyzed 68,879 Google searches conducted in March 2025 and the results when an AI summary appeared are astounding.
People clicked a result link in 8% of visits compared with 15% when there was no summary.
A link inside the AI summary was clicked in just 1% of visits.
People ended their browsing session on 26% of pages with an AI summary, versus 16% on pages without one.
Those numbers are usually presented as a traffic problem for the publisher. And, make no mistake, they are. But they also describe an advertiser measurement problem. When an AI answer uses publisher reporting to explain a category or compare features of a product, the publisher may have influenced the decision without receiving the visit, and an advertiser may benefit from that recommendation without recording a referral, an assisted conversion or even an identifiable exposure.
The influence still happened; we just can’t confirm it. So publishers lose the visit, advertisers lose the attribution and the platform has interaction data that connects the two of them.
The available metrics describe visibility but not contribution
Here is where it gets a bit twisted. GSC can report clicks and impressions associated with links in AI Overviews or AI Mode, which is useful, but it preserves the old measurement framework:
Was a link displayed?
Did somebody click it?
It doesn’t show if a source helped shape an answer when its link was not visible. It can’t show which answer within the conversation changed the person’s criteria. And it doesn’t connect a cited comparison with a later branded search or direct visit or marketplace purchase … or, I’d argue, just as important, an offline conversion.
Even “impression” demands more scrutiny. Can a link record an impression just because it appeared within an AI response? This still doesn’t tell a publisher whether it influenced the response. On the other hand, content may contribute to an answer without producing an impression at all. So at the end of the day, visibility and influence are no longer interchangeable, if they ever were.
This creates dashboards that are “technically” correct but, for a business, utterly useless for strategy and decision making.
The reported clicks happened; the recorded impressions met the “definition” and conversions followed the rules but the decisions don’t conform to these rules and definitions.
More activity can conceal less understanding
The problem is not merely undercounting.
It is making the wrong business decision with data and, worse, confidence.
In one small paid-search account I reviewed, 255 clicks produced a 5.76% click-through rate at an average cost of $3.06. The account spent $780.79 and generated zero conversions. Okay, some of the engagement metrics suggest the ads were doing their jobs, but under the hood, you can see traffic drifting into searches for AI tools rather than the product or service being sold.
This example comes from paid search rather than an answer engine but it exposes the same lack of transparency. A click can prove that something was selected and at the same time, can’t prove that the platform understood the intent. The interaction had value, yes, but the data should absolutely not guide the optimizations, strategy or any budget decisions. It’s a house of cards.
As platforms automate query matching and AI takes over discovery, this gap between recorded event and intent grows. More queries, impressions and engagements are creating more data while giving advertisers and publishers little visibility into why a decision was made.
Publishers and advertisers now share the same blind spot
The old search gave publishers traffic in exchange for making their content available and that traffic aimed to support subscriptions or advertising. Now the AI systems weaken this value proposition by taking far more information than they give. Cloudflare estimated that in June 2025, OpenAI crawled publisher pages about 1,700 times for every referral it sent. Anthropic’s ratio was 73,000 to one. Ouch.
For publishers, this is audience and inventory that is no longer for sale; for advertisers, the loss is action data. If fewer people land on publisher pages, there are naturally fewer referral paths, assists/lift and audience signals used to evaluate the media. The same zero-click interaction removes value from both sides, even when the answers do their job and move people closer to buying.
This is why the issue can’t be left to the SEO, AEO (answer engine optimization), analytics and PPC teams independently because it’s not a channel-specific reporting problem.
It is a market-level loss of observable actions and engagements.
The market will continue to optimize towards what it can still see
Measurement systems do more than report performance. They also determine who or what receives the money.
If AI influence remains opaque, the budgets will continue to favor the last observable action: the branded search, the direct visit or the conversion event. The source that created the demand may look weaker and the channel may look more efficient even though that’s not the case. Publishers may inadvertently cut the content that influenced decisions because it can no longer produce enough traffic and advertisers may shift spending towards the touchpoints that remain measurable instead of the ones doing the most work.
This has become a strategic risk.
The industry is changing how people behave and discover and make decisions faster than it is changing how it values or even measures influence. We’re adding conversational interfaces, LLMs, generated answers and agents within a measurement architecture that doesn’t quite fit because the buying journey or “path to conversion” is no longer linear.
The result will not be the absence of measurement. It may be worse.
Precise measurement of a shrinking portion of the decision.
This is a guest post written by Shelley Stone, follow her Substack at AdTech Shelley
Related Articles
The follow-up-counted-as-new-query detail is the quiet scandal here. The practical patch we landed on: stop waiting for signals to arrive and go sample the answers directly. A fixed set of buyer conversations, run on a schedule across engines, timestamped, scored for who got named. It measures the decision surface instead of the click that never happens. Crude next to analytics, but it sees exactly the layer this piece says is invisible.