Portfolio Approach to Emerging AI Ad Channels for Enterprise Brands

Enterprise brands entering AI advertising face a choice that looks simple and isn't: where to put the first dollar. The answer this piece makes is that no single surface deserves that dollar. A portfolio, split deliberately across conversational AI placements, AI search-adjacent inventory, and the earliest pure-LLM channels, in proportion to how mature and reachable and intent-rich each one actually is, beats a bet on any one of them. US AI ad spending is set to reach $68.25 billion by 2030, more than double what it is today, so the migration of attention toward these surfaces is not speculative eMarketer. But the money hasn't caught up with where that attention is heading: more than 80% of 2026's AI advertising spend still sits next to AI-generated content, things like Google's AI Overviews, rather than inside an actual chatbot conversation beet.tv.

That gap between where dollars flow now and where usage is clearly heading is exactly the setup that rewards a portfolio approach. Andrea Tortella, CEO, drew the comparison at Beet Retreat Berkshires in September 2026 to the early days of Google, Facebook, and TikTok, when the brands willing to buy in before the auction matured captured returns nobody could replicate later. That's not a nostalgia trip. It's a structural argument: inventory that's underpriced relative to its eventual reach doesn't stay underpriced once everyone notices.

Tortella also framed the shift in two acts. Act 1, which ran through most of 2025, was about using AI to sharpen tools that already existed, generative video, smarter bidding, tighter measurement. Act 2, now underway, is different in kind: ads are moving directly inside the LLM environment itself. Enterprise teams that treat Act 2 as one more surface to pilot, the way they piloted a new social platform or a new demand-side vendor, are misreading what's in front of them. It is already three distinct legs, each maturing at its own pace, each reaching a different kind of attention, each demanding a different measurement approach. The real operating question for 2026 is how to allocate across a channel family whose parts are aging at completely different speeds beet.tv. It's how to allocate across a channel family whose parts are aging at completely different speeds.

The three legs of an AI advertising portfolio

The first leg, AI search-adjacent inventory, is the one enterprise buyers already half-understand because it looks like search. Ads sit beside or below an AI-generated answer, in Google's AI Overviews and similar formats, rather than inside the back-and-forth of a conversation, and this leg is by far the most mature: it's the largest current spend category and it's forecast to grow 152% to $26.42 billion in 2026 alone eMarketer.

The third leg sits at the opposite end of the maturity curve. Enterprise AI assistants, third-party LLM surfaces, agentic workflows, these are all still forming, with no standardized auction and supply that's fragmented and, in places, not even confirmed to exist yet. Between those two sits the conversational leg, the placements that live inside an actual chat session with a model like ChatGPT eMarketer.

What separates the three legs is their maturity, reach, and intent quality. It's what kind of intent each one is built to catch. Leg 1 reads intent at the level of a single query. Leg 3, when it matures, will read intent across a multi-step agent workflow, a task the agent is executing on the user's behalf rather than a question the user is asking.

Availability across surfaces is uneven too, and enterprise planning has to reflect that rather than assume parity. ChatGPT ads are confirmed and already buyable at scale. Gemini inventory remains unconfirmed, and Google has publicly denied a widely circulated report that ads were coming to Gemini in 2026. Anthropic has gone the other direction entirely and stated, as a matter of policy, that Claude will stay ad-free. None of that is a reason to wait on the whole channel. It's a reason to build the portfolio around what's actually buyable today while tracking what opens next. Treated correctly, the three legs aren't competing for the same budget line, they're complementary, each one catching the buyer at a different point in the decision journey.

How conversational intent differs from keyword intent

Search intent has always worked as a kind of shorthand. Someone typing "project management software" is declaring a category outright, and the keyword auction was built entirely around that compression. Conversational intent doesn't compress the same way. A user might never type the category term at all, and instead spend several turns describing a problem, coordinating a remote team, tracking deadlines across time zones, before an assistant or an ad system has enough signal to know what that person needs. A user can spend several turns describing a problem, coordinating a remote team, or tracking deadlines across time zones before an assistant or an ad system has enough signal to know what that person needs. It appears as accumulation rather than declaration.

That changes what the ad system is actually bidding on. ChatGPT's targeting draws on the topic of the conversation, a user's past chats, and past interactions with ads, but the data itself never leaves OpenAI's hands, advertisers bid against the signal without ever seeing the raw conversation behind it. That's a meaningfully different arrangement than the keyword auctions enterprise teams have run for two decades, where the query itself was often visible in some form downstream.

Verve Group's March 2026 launch is worth understanding in detail because it shows what infrastructure for this actually looks like in practice. The platform became the first open-market advertiser to operationalize high-fidelity intent data pulled from AI chat interfaces for programmatic buying, combining zero-party data, search intent, and pseudonymized chat activity into a single layer that processes more than 1 billion signals a day, all without cookies or device identifiers Verve Group. That last detail matters more than it might first appear. LLM contextual targeting pulls relevance from what's happening in the conversation right now, not from a browsing history stitched together over months, so there's no persistent identifier required at all. That's a structurally different privacy model than cookie-based behavioral targeting, and it's one built to survive the regulatory tightening that's already underway rather than get caught by it later.

The ad formats available inside LLM interfaces and their trade-offs

Three formats define the current conversational landscape, and each carries a distinct set of trade-offs enterprise buyers should weigh before committing creative production to any one of them.

The sidebar panel is the safest and least remarkable option. It runs at a lower CPM, closer to conventional display economics, and it's desktop-only for now, functioning as a persistent impression unit inside workspace-style apps rather than a real monetization engine. The sponsored follow-up suggestion question is more interesting and more contested: Perplexity built and shipped this format, then walked it back in February 2026 over concerns that it eroded user trust. Mechanically, it works fine. The blur between an assistant's genuine suggestion and a placement someone paid for is what got flagged.

Response-grounded brand mention is the earliest-stage of the formats, with the highest potential revenue ceiling and the highest UX risk, and it is not at scale in 2026.

Format choice should track the context a user is already in. Research-oriented or professional LLM apps favor the subtler sidebar and chip formats, while consumer shopping contexts can support a more direct inline card, since the user showed up already carrying commercial intent. And disclosure isn't a nice-to-have anywhere in this mix. The IAB's AI Transparency and Disclosure Framework shipped its first version in 2026, FTC guidance on the subject is explicit, and state-level disclosure laws are stacking up one after another, so any publisher partner in this space has to pass through a clear "Sponsored" label without exception. Done right, that label is what keeps native, contextual formats from feeling disruptive the way a banner or a pop-up does: it draws a clean line between the answer and the ad, and that line is what protects the trust the whole format depends on.

How to weight the three legs today (a starting allocation framework)

Weighting should follow a leg's combination of reach, intent quality, and how mature its measurement actually is, not how much buzz surrounds it or how fast a forecast says it will grow.

By that logic, Leg 1 deserves the largest share of budget today eMarketer. It has proven measurement, buying infrastructure that already exists, and $26.42 billion in projected 2026 spend that signals this is core media now, not an experiment enterprise teams can afford to sit out while they "wait and see" eMarketer. ChatGPT ads are already buyable across more than 40 countries, and a 1,641% year-over-year jump in chatbot ad spending reflects actual advertiser commitment, not just a platform's own announcement eMarketer. Leg 3 should be treated the way a trading desk treats an option position: small, deliberate, entered with success criteria written down before any money moves, because the measurement infrastructure supporting it isn't standardized yet.

That baseline shifts by industry. E-commerce and retail, the largest vertical in global digital ad spend at $182.7 billion, carries the highest conversational intent of any category, since a shopping assistant query is commercial by definition, and that argues for over-weighting Leg 2 beet.tv. Health & Wellness was the fastest-growing AI ad category in Q1 2026, up 165% year-over-year, reaching $48.1 billion, but the compliance risk in AI responses is significant, so weight should go toward Leg 1 until brand-safety infrastructure on Leg 2 matures beet.tv Verve Group. Technology and telecom, at $87 billion, skews toward professional and research-heavy LLM usage beet.tv. Sidebar and chip formats on Leg 2 may simply outperform an inline card beet.tv. Automotive, at $52.1 billion, runs on long consideration cycles, a buying journey where conversational intent gathered across multiple turns pays off, making Leg 2 a natural home for mid-funnel content beet.tv Verve Group.

None of that allocation matters much without the infrastructure to execute it. A cross-surface demand-side platform that can actually read conversational context is what makes a multi-leg portfolio workable, as opposed to a generalist platform buying blind to what's happening in the conversation, or a single-surface network that caps a brand's reach at one app's audience.

What signals should move budget between legs as the channel matures

None of this allocation is static. It's a rebalancing framework, and it should move on specific, observable signals rather than on instinct or on whichever platform issued the last press release.

New surface availability is the first signal to watch closely. When a confirmed LLM surface opens up ad inventory, the buyers who show up early capture scarcity-priced placements and get first-mover data on what actually performs, the same way publishers who integrate early learn which placement works before the broader market bids CPMs up to parity. Measurement infrastructure reaching real parity is the second: Leg 3 should only get meaningful budget once attribution can actually tie spend to an outcome, and enterprise teams are better off admitting that the attribution problem in assistant-mediated discovery isn't fully solved than pretending a last-click model just carries over.

Conversion data from live campaigns is the clearest quantitative signal available right now. ChatGPT ad conversion rates are running 4-7% in commercial verticals, with top performers clearing 8%, which compares favorably to the 2-4% range typical of Google Search in matched verticals, and a leg that sustains conversion quality above a defined threshold has earned a bigger share of the budget beet.tv. Platform policy is the signal that's hardest to plan around but impossible to ignore. Perplexity pulled back on ads entirely in February 2026 over trust concerns, and Anthropic has committed to keeping Claude ad-free as a matter of policy, both reminders that a surface-level decision can eliminate an entire leg with no warning, which is itself the argument for staying diversified rather than concentrated.

Agentic AI maturity is the last major signal, and it's the one that will eventually reshape the whole framework. Gartner projects that 60% of brands will use agentic AI for one-to-one interactions by 2028, and the IAB's Outlook found that 96% of buyers are already aware of agentic AI for buying and executing campaigns. Once that infrastructure reaches production-grade reliability, Leg 3 budget should scale up to match it, not before. The practical cadence for all of this is a quarterly review tied to CPM movement, conversion data, and platform news, paired with an annual structural check on whether any leg has earned reclassification.

What enterprise teams need in place before running this portfolio

None of the above works without first-party and zero-party data in usable shape. Conversational AI targeting doesn't lean on third-party cookies or device IDs the way display and search retargeting always have, so the enterprise's own consented data becomes the main signal it can actually bring to the table, and without it, campaigns fall back to generic contextual matching that wastes the precision this channel is supposed to offer.

Before committing serious budget to Leg 2 or Leg 3, an AI visibility audit belongs on the checklist beet.tv. If a model can't accurately place a brand's products or services based on its existing entity data and its footprint elsewhere online, a sponsored placement just lands the brand in a context where it's poorly understood to begin with, undermining the ad before it ever runs. That audit needs to happen before the scaling, not after.

Measurement has to be built for the channel it's measuring, not borrowed wholesale from search. Assistant-mediated discovery doesn't map cleanly onto last-click attribution, so success metrics need to be defined per leg: Leg 1 can lean on metrics comparable to search, Leg 2 needs a conversion-quality lens that weighs rate over raw volume, and Leg 3 needs leading indicators until its attribution infrastructure actually catches up to the other two. Gartner found that 80% of enterprise applications shipped or updated in the first quarter of 2026 embed at least one AI agent, up from just 33% in 2024, so the internal AI build-out and the external AI advertising opportunity are unfolding on the same timeline, and brands that wire their first-party data infrastructure into both will simply move faster than the ones treating them as separate projects beet.tv.

The organizational lift required to start isn't large. A defined test budget for each leg, success criteria written down before the first dollar goes out, and one named person accountable for rebalancing decisions is enough to begin. Where this framework tends to fail is the absence of anyone actually owning the decision to move the money. It's the absence of anyone actually owning the decision to move the money. What enterprise teams need in place before running this portfolio is outlined by SOURCE PAGES (what the pages behind the outline's links say).