The ad unit is leaving the search results page.
Google is expanding AI Max so advertisers can test budgets, bidding, and controls around campaigns that match broader intent. OpenAI is testing ads in ChatGPT's free and Go tiers, with a model built around conversation instead of a list of blue links. The details will keep changing. The direction won't.
AI assistant ads are turning marketing measurement into a much harder problem. The customer may ask one long question, compare three products inside an answer, click nothing, return two days later, and buy after a follow-up conversation. The old trail of impression, click, session, conversion is about to get very thin.
Search is becoming a conversation
The search page gave marketers a neat fiction: intent could be compressed into a query, and the query could be tied to a result. It was never perfectly true, but it was useful enough to build entire departments around.
A conversational assistant breaks that compression. A buyer can say, “I need a quiet air purifier for a 600-square-foot bedroom, I have a cat, and I don't want to replace filters every month.” That prompt contains product category, constraints, context, objections, and a budget signal. It also creates a much richer surface for an assistant to recommend something.
Google's own description of its AI Search work points toward answers that help people complete more involved tasks. Its ad products are following the same path. The new AI Max testing and planning tools are a clear sign that campaign systems are being asked to interpret intent, not just match keywords.
That sounds like better relevance. It may be. But relevance is not the same as visibility, and visibility is not the same as a measurable click.
The click was doing more work than we admitted
For years, the click served as a tiny receipt. It told the marketer that a person saw a message, found it interesting enough to continue, and arrived somewhere the brand controlled. That receipt was incomplete, noisy, and often overvalued. It was still a shared unit of measurement.
In an assistant interface, the recommendation itself can do more of the selling. The product may be named, summarized, compared, and narrowed down without a visit to the advertiser's site. A customer could remember the brand from the conversation, search it directly later, and convert through a channel that reports no relationship to the original assistant exchange.
OpenAI's published explanation of its ChatGPT ads test says the company is designing around consumer trust and usefulness. That distinction matters. An ad that appears inside a useful answer will be judged less like a banner and more like a recommendation. The standard for honesty rises with the intimacy of the interface.
It also means the old attribution dashboard starts lying by omission. Direct traffic looks stronger. Branded search looks stronger. “Organic” assisted conversions pick up activity that may have begun inside an answer nobody can inspect.
That is the same measurement drift I wrote about in the attribution drift problem, except the missing middle is now a conversation rather than a hidden social impression.
The new scarce asset is answer eligibility
Ranking used to be the main fight. Then it became ranking plus click-through rate. In assistant-led discovery, the fight shifts again: can the system confidently include your brand in the answer?
That depends on more than a page's position. Assistants need clear product facts, consistent claims, accessible policies, useful reviews, strong third-party references, and content that survives extraction without losing its meaning. A brand may be visible in search and still be absent from the answer because its proof is thin or contradictory.
This is why AI search measurement can't be reduced to a new rank tracker. The question isn't simply whether a page appears. The question is whether the brand becomes a trusted piece of the assistant's response, under which prompts, beside which competitors, and with what wording.
That wording is not a minor detail. If the assistant describes your product as “cheap” when you position it as durable, or calls a regulated service “safe” without the right qualification, the answer can damage the brand while technically mentioning it.
Small brands face the sharpest tradeoff
Large companies will buy access, test creative, and hire teams to monitor how assistants describe them. Smaller companies will be asked to compete on the quality of their evidence while having fewer resources to produce and maintain it.
That creates a nasty asymmetry. A local restaurant may have the best menu in town, but if its hours, allergy information, reservation rules, and recent reviews conflict across the web, an assistant has reasons to hesitate. A small ecommerce brand may make an excellent product, but if every retailer describes it differently, the safest machine answer may be a competitor with cleaner facts.
The answer isn't to publish more content for the sake of volume. It is to make the important facts boringly consistent. Product names, prices, availability, service boundaries, credentials, policies, and proof should agree wherever a customer might encounter them.
This is less glamorous than generating a hundred AI-written articles. It is also more likely to survive the next interface change.
Creative has to earn its place in the answer
The first generation of AI advertising will tempt marketers into a familiar mistake: treat a new surface like a new placement. Take the old headline, resize it, and call the work finished.
That won't hold. A conversational ad has to fit the moment around it. It may need to answer a constraint, acknowledge uncertainty, disclose its commercial relationship, or give the user a reason to ask a better follow-up. The creative unit is no longer only the message. It is the message plus the context that makes the message acceptable.
This is where brands with a real point of view have an advantage. Not because an assistant automatically rewards originality, but because clear positioning gives the system something coherent to retrieve and repeat. Vague brands become mush inside a summary. Specific brands have a better chance of being described accurately.
The practical test is simple: take your five most important customer questions and read your answer aloud. Does it sound like a useful human response, or a brochure trying to sneak into a conversation? The difference will matter more as assistants become a discovery layer.
Measurement needs a wider frame
Marketers don't need to abandon clicks. They need to stop treating clicks as the entire story.
A better measurement model will combine direct response with answer-level signals. Track whether the brand is included in relevant assistant responses, how often key claims are repeated correctly, which competitors appear beside it, and whether the assistant sends a qualified customer or simply creates familiarity.
Then connect those signals to harder business outcomes: branded demand, direct traffic quality, conversion rate, repeat purchase, customer support questions, and sales-cycle length. None of these proves causation on its own. Together, they show whether the brand is becoming easier to choose.
The measurement crisis is not that data disappears. It is that data arrives in fragments owned by different platforms, with no shared definition of influence. The deeper problem with AI advertising is that persuasion can become invisible while optimization becomes more automated.
A serious team should set up a prompt panel before spending heavily. Use a fixed set of customer questions, run them across the assistants that matter to your audience, archive the answers, and repeat the exercise monthly. Record inclusion, position, claims, citations, disclaimers, competitors, and next-step behavior. It won't produce perfect attribution. It will reveal changes your click report can't see.
The interface is part of the brand now
The strategic mistake would be to ask only where ads will appear next. The better question is what kind of relationship the interface is creating between a person and a commercial message.
Search ads trained people to scan. Social ads trained people to interrupt themselves. Assistant ads will ask people to accept guidance inside a conversation. That is a more powerful position, but it carries a larger penalty for manipulation, sloppy claims, and hidden incentives.
OpenAI's advertising principles describe boundaries around how ads should work in ChatGPT. Google's 2026 marketing announcements show a different route into the same future. Neither platform has finished the design. Marketers shouldn't wait for a finished design to decide what they won't compromise.
The next phase of marketing won't be measured by how often a brand interrupts a person. It will be measured by whether the brand earns a place in the answer, keeps its facts straight, and remains useful after the recommendation ends.
That is a much higher bar than getting the click. It is also a better test of whether the marketing was doing anything worth preserving.