AI assistant advertising is no longer a hypothetical channel for a conference panel. OpenAI says it's testing ads in the United States for ChatGPT's Free and Go tiers, while Microsoft says AI-driven sessions are growing quickly across the web. The next marketing fight won't be over who gets the most impressions. It'll be over who gets to shape the recommendation a buyer trusts.
That changes the funnel. Search ads interrupt a query. Assistant ads can enter a conversation, a comparison, or a decision that has already been narrowed by the model. The useful question isn't whether ads belong in AI assistants. It's whether the old rules for paid media survive the move.
The Placement Is the Conversation
A search ad appears next to a list of options. The user still has to inspect pages, compare claims, and decide what matters. The ad can influence the shortlist, but it doesn't usually pretend to be the shortlist.
An assistant works differently. A person can ask, “Which software should I use for a five-person agency?” and receive a compressed answer shaped around constraints, context, and a recommendation. If a sponsored result appears inside that exchange, the commercial message sits closer to advice than to an ad unit.
That proximity is the opportunity and the problem. The platform can make an ad more useful by matching it to the actual need. It can also make the commercial boundary harder to see. A label that technically says sponsored won't repair a recommendation that feels like a trusted friend made it.
OpenAI's own approach to advertising and expanding access says ads will be separated from ChatGPT's answers and won't influence the answers themselves. That's the right promise. The hard part will be proving it at scale, especially when the system knows enough about the conversation to make the ad feel uncannily relevant.
Intent Gets Compressed
Marketers have spent years building stages around intent. Awareness has broad language. Consideration has comparison terms. Purchase has a product, price, or location attached. Assistant interfaces squeeze those stages together because the user can state the problem in plain language and ask for a next move.
That means the useful targeting signal may not be a keyword. It may be the combination of a constraint, a preference, and a moment of uncertainty. “I need a payroll tool that handles contractors and doesn't make my team learn another complicated system” carries more commercial meaning than a short list of search terms.
The old funnel also assumed that marketers could see the path. Someone clicked a result, visited a landing page, filled out a form, and became a lead. Assistant journeys may produce fewer visible steps. The model can summarize options, pass a user toward a product page, or trigger an action without giving the brand a clean view of everything that happened before the visit.
Microsoft's three eras of the web playbook argues that AI agents are becoming a fast-growing audience and says automated traffic is expanding faster than human traffic. Whether every number holds across every category, the strategic point is sound: brands will need content and product data that machines can interpret before a human ever sees the brand site.
That is not traditional SEO with a new label. It is a product information problem, a trust problem, and a measurement problem at the same time. The same shift is already changing how AI advertising moves media buying, only now the recommendation itself is part of the inventory.
The New Brand Surface Is the Data
If an assistant recommends a product, it needs something to work with. That something may come from a product feed, a review, a retailer page, a public document, a support article, or a prior conversation about the category. The brand's polished homepage is only one input, and it may not be the most important one.
This is where many marketing teams are exposed. They have strong campaign language but weak machine-readable proof. Prices are inconsistent. Policies are buried. Product differences are vague. Reviews describe one experience while the landing page promises another. The assistant doesn't see a brand story. It sees a set of signals and tries to reconcile them.
The fix isn't to flood the open web with more copy. It is to make the facts agree. Product names, use cases, limitations, service areas, policies, and evidence should line up across the places an assistant is likely to read. The work is less glamorous than generating another campaign concept, but it affects whether the system can describe the brand accurately.
I made a similar argument in why AI search visibility needs a measurement model, not a rank. Being mentioned is not the same as being selected. Being selected is not the same as being trusted. Those are separate signals, and assistant advertising makes the gap between them more expensive.
Measurement Loses the Middle
Paid media teams are used to a chain of events. Impression, click, session, conversion. The chain is imperfect, but it gives the operator something to optimize.
Assistant advertising adds a middle that may be difficult to observe. The user asks a question. The assistant interprets it. A set of options is assembled. A sponsored recommendation may appear. The user may click, ignore it, ask a follow-up, or remember the brand later and visit directly. The dashboard may record only the last step.
That creates a familiar temptation: credit the measurable click and ignore the unmeasured influence. It happened with view-through attribution, dark social, and branded search. AI assistants will produce a new version of the same argument, with even less visibility into the decision path.
The answer is not one perfect attribution model. There won't be one. Start with a measurement stack that admits what it cannot see. Track direct response, assisted demand, branded search movement, qualified conversion, repeat behavior, and the quality of the users who arrive. Then compare assistant-exposed cohorts with sensible controls where the platform allows it.
A click can tell you that a recommendation created motion. It cannot tell you whether the recommendation was accurate, whether the user was a fit, or whether the brand's future reputation paid for the conversion.
Disclosure Is Part of Performance
The industry will treat disclosure as a compliance detail until a user feels deceived. Then it will become a trust crisis.
Assistant ads need a clear visual and verbal distinction from organic answers. They also need enough context for a person to understand why a product appeared. “Sponsored” may identify the transaction, but it doesn't explain the recommendation. If the placement is based on the user's conversation, the platform should make the commercial logic legible without exposing private information.
This is where marketers need to resist the usual instinct to optimize the label away. A hidden ad may earn a click. A clear ad has a better chance of earning a customer who knows what they chose. Short-term performance can punish honesty, but brands that sell expensive, regulated, or high-consideration products cannot afford to build demand on confusion.
The IAB's work on AI ad disclosure is part of a broader industry attempt to define how commercial content should be identified as interfaces change. Standards will help, but the practical test is simpler: could a reasonable person tell where the assistant's advice ends and the advertiser's message begins?
Build for the Decision Layer
The teams that benefit from assistant ads won't be the teams that buy the cleverest placement first. They'll be the teams that make the decision layer dependable.
That means cleaning the product facts that assistants read. It means writing comparison pages that answer real constraints instead of repeating slogans. It means keeping policies current, publishing evidence for important claims, and giving sales and support teams one version of the truth to work from.
It also means deciding where automation stops. Let a platform match an eligible need to a relevant offer. Don't let it invent a promise, blur a limitation, or quietly change the audience definition because the dashboard likes the result. Define the claims that require approval and the outcomes that matter after the click.
Microsoft's agentic advertising materials point toward a web where automated systems are not just crawling pages, but acting on behalf of people. That makes machine-readable information part of the customer experience. The page is no longer only for the human who reads it. It is also for the system that decides whether the page deserves to be recommended.
The old funnel won't disappear in a single launch. It will just become harder to see. Buyers will still compare, hesitate, ask friends, and look for proof. The difference is that an assistant may perform more of that work before the brand gets a visit.
The winners won't simply be the brands with the biggest budgets. They'll be the brands whose facts, promises, and customer experience remain coherent when a machine compresses the journey into a sentence.
That is a much tougher brief than buying another placement. It is also the one marketing teams should be preparing for now.