Internal Agency Case for Adding Conversational AI Capabilities

Conversational AI advertising is not a channel agencies can plan for later. ChatGPT opened a self-serve Ads Manager on May 5, 2026; Google now auto-upgrades campaigns into AI-served placements, and you cannot opt out; Microsoft Copilot enrolls eligible accounts on its own. If you run Search, Shopping, or Performance Max campaigns, you are already placing ads inside AI-generated answers, whether you decided to or not. The internal argument agencies need to win is whether to run this channel with skill or keep running it by accident.

Why agencies cannot treat conversational AI advertising as a future-state planning item

ChatGPT's ad program began as a closed pilot in February 2026, and only advertisers who could commit a substantial upfront spend and work directly with OpenAI could join. That era ended on May 5, 2026, when OpenAI opened a self-serve Ads Manager to any approved U.S. advertiser, and no minimum commitment was required. Agencies could once treat the channel as something to watch, but that single change forced them to start running it instead.

Google has not given advertisers a choice. AI Mode and AI Overviews already serve ads through the same Search, Shopping, and Performance Max campaigns agencies manage today, with no opt-in and no opt-out for advertisers (publishers, at least, can now exclude their own sites from AI features through Search Console, but advertisers have no equivalent lever). Campaign-level broad match settings and Automatically Created Assets auto-upgrade to AI Max starting in September 2026, and Dynamic Search Ads get a mandatory upgrade in February 2027. None of this requires an agency to sign up for anything. It happens inside accounts agencies are already managing.

Microsoft Copilot runs on the same logic. Eligible campaigns are automatically opted in, with no display guaranteed and no way for an agency to suppress placement. An agency managing Google or Microsoft Advertising accounts today is already an AI advertiser.

Where the paid opportunity is concentrated today

Four platforms could plausibly carry paid placements inside AI-generated responses, but each has gone its own way, and if your agency's pitch treats them as interchangeable, it loses credibility fast. As of October 2026, the buyable surfaces sit in two places. Google AI Mode and AI Overviews serve ads through existing Search, Shopping, and Performance Max campaigns, using broad match or AI Max for Search campaigns, while Shopping and Performance Max campaigns qualify automatically without added setup. No separate media buy is required beyond what agencies already run. Microsoft Copilot works the same way: it serves ads from existing Microsoft Advertising accounts, and eligible campaign and ad types are opted in by default.

Perplexity sits outside this map. The company wound its ad program down and told the Financial Times that sponsored placement risks making users suspicious of everything they see, then shut the program down permanently and confirmed it has no plans to bring it back. Perplexity does not offer paid placement inside its responses, so if your agency pitches that, it is pitching something that does not exist; push organic visibility there instead.

Knowing which surfaces are live and buyable now, and which require organic strategy only, is what keeps an internal case from reading as speculation. Platforms that specialize in connecting agencies to the inventory actually monetizing conversational AI serve exactly this function: they separate operational reality from a vendor's roadmap slide.

You need to address one constraint directly in any pitch touching B2B clients. ChatGPT's paid subscription tiers run ad-free, so the ad surface reaches consumers on free and entry-level accounts, not the subscribers who pay for premium access. For a B2B advertiser trying to reach a senior buyer who pays for a premium subscription tier, that's a structural ceiling on the channel today, not a targeting problem to solve with better copy.

Why conversation intent differs from a search query

A search query tells an advertiser what a user typed. A conversational prompt tells an advertiser what that user is actually trying to do, often including the feelings, constraints, and tradeoffs that a keyword never captures. That gap between expressed intent and motivational intent is the real reason conversational AI advertising deserves its own capability, built separately from the search team.

A user typing "project management software" into Google has already decided to shop. A user telling an AI assistant that their team keeps missing deadlines, that the whole group works remotely, and that coordination has become unmanageable is describing the same need, in language that contains no product category. The intent is there. It just arrives wrapped in context that a keyword-matching system was never built to read.

That shift changes what "targeting" means as a discipline. The unit of targeting in conversational AI is not a keyword but a context hint: a short natural-language description, usually one or two sentences, of the conversations, topics, or situations where a given ad belongs. That hint gets matched not against a single query but against the live conversational trajectory: everything the user asked earlier, how the AI responded, and which follow-up questions came next. Relevance is not fixed at the moment of the query. It moves with every exchange in the conversation, which makes writing a good context hint closer to behavioral prediction than to keyword list-building.

Microsoft has built Copilot's relevance model to weigh the broader arc of a session as it evolves. The privacy architecture that results from this approach also differs structurally from search advertising: relevance is derived from the live conversation rather than from historical browsing data, which reduces (though does not eliminate, since signed-in platforms may still use persistent identifiers) dependence on cross-site tracking at a moment when privacy regulation is tightening around exactly that kind of tracking.

The practical consequence is that writing a context hint is a genuinely new craft, and it shows in the data: teams that write hints as comma-separated topic lists, the way they'd build a keyword list, get weak matching results, consistently. The real discipline is to treat motivational intent, not expressed intent, as the raw material for targeting. Platforms built for real-time prompt analysis and contextual deployment across AI surfaces, the kind that specialize in reading a conversation-level context hint rather than matching a single query, exist precisely because this is a different skill than running a Google Ads account well.

New skills requirement and search team capabilities

The intent signal argument has a direct staffing consequence: conversational AI advertising needs new competencies at the targeting, creative, and measurement layers, not a more advanced version of what a search team already does.

At the targeting layer, the operative question is no longer "what string did this person type" but "what is this person trying to accomplish, and is this offer actually useful to them right now. That's a copywriting and inference skill, closer to strategic planning than to keyword architecture, and it is not a skill most paid search specialists have had reason to build.

At the creative layer, ad copy has to function as a continuation of a conversation the user is already in the middle of, not as the opening line of a new search session. Copy written for a results page interrupts. Copy written for a chat interface needs to read as the next useful thing someone would say, so these are different writing problems with different failure modes.

The landing page layer carries more weight than agencies may expect. OpenAI's ad system reads the landing page itself as a relevance signal and crawls it during the review process, so the page is part of the targeting system, not a destination bolted on after the ad is approved. That changes how performance teams brief a landing page and how they QA it before launch.

Measurement is the layer where agencies need to be most honest with themselves and with clients. The channel currently lacks the reporting granularity a search team is used to: there is no search- or prompt-level reporting, so agencies cannot see what a user was asking when an ad got served, and buyers are actively pushing platforms for negative targeting controls and prompt-matching visibility that simply do not exist yet.

A useful way to make this tangible for a CFO or managing director is to describe the five layers a working practice needs: a demand-and-auction layer for buying across ChatGPT, Google AI, and Copilot; a context-and-targeting layer for building hints and custom audiences; a creative layer producing distinct messages per angle and per surface; a landing page matched to each ad; and a unified attribution layer that counts each conversion once across AI and non-AI channels. None of that is a reassignment of existing headcount. It's a build, and it should be budgeted as one, with the advantage that the skills are learnable and the field attempting them is still thin.

The Economics of Adding This Capability

Building this capability now costs less than building it under a client's deadline, and the difference compounds because the channel is still thin on competition. An agency that develops context-hint expertise and a working attribution model before the field gets crowded establishes a margin structure that a late entrant will have to buy, not build.

The auction itself rewards early precision in an unusual way. A small-budget advertiser with a sharply written context hint can outperform a large-budget advertiser with a vague one, which is not how most paid media auctions behave. That property rewards an agency that invests in the craft of writing hints well before the channel matures and the average advertiser catches up.

The pressure on the client side is not hypothetical. Retail search advertising alone is set to reach a very large absolute spend figure in 2026, and travel, finance, and e-commerce clients already show measurable consumer AI adoption inside their purchase journeys. Those clients will move budget toward AI surfaces because their customers are already there, regardless of whether their agency led that conversation or got dragged into it.

At the same time, the same platforms now offering AI inventory are pressuring the channel agencies built their margin on. When an AI Overview appears on a Google results page, paid click-through rate drops substantially. The search campaigns an agency has run for years are earning less on the same spend, on the same platform, because that platform has started answering the query itself before a paid link gets a chance to.

That combination, pressure on the legacy channel and an unclaimed advantage in the new one, is the financial argument. Innovation budget spent now building context-hint skill, landing page alignment, and attribution tooling converts directly into a billable managed service later. The alternative, building the same capability after a client demands it, costs more and arrives at lower margin, because the agency pays a premium to catch up when it could have compounded a head start.

Structuring the internal pitch to earn a yes from finance

The pitch that gets approved is the one framed as margin defense and early capture of demand that's already forming, not as a bet on an unproven channel. Finance does not need to be convinced that conversational AI is interesting. Finance needs to be shown that the agency is already exposed to it and is choosing whether to manage that exposure well.

Start the pitch with the fact that's hardest to argue with: if the agency manages Google or Microsoft campaigns, it is already serving ads inside AI answers today. The decision in front of leadership is whether to run what's already happening deliberately and well, or to keep letting it run on default settings built for search.

Lead with the measurement gaps. Acknowledging openly that prompt-level reporting doesn't exist yet, that negative targeting isn't available, and that attribution is genuinely harder in this channel is a credibility asset in front of a skeptical audience, not a weakness to paper over. A 2026 survey found 84% of advertisers report neutral or negative results with Google's AI Max, against what the platform itself claims, and clients caught between those two accounts need an agency willing to say what is and isn't measurable yet, and to design tests that produce internal data the client can actually trust.

Structure the ask in three phases so finance has milestones to approve. An audit phase answers a narrow question: what is the agency already serving inside AI surfaces today, and through what settings. A skills phase builds context-hint training, landing page alignment, and attribution setup. A managed service phase turns that capability into a billable offering you sell to clients. Each phase is a smaller ask than the one before it was large, and each produces something finance can see before approving the next.

Prioritize the clients where the pitch will land fastest. Retail, travel, and e-commerce are the verticals with measurable consumer AI adoption inside purchase journeys. Those clients will feel the pressure to shift budget soonest, so pitching the capability to them first concentrates the investment where the demand is closest to arriving.

You can state the competitive parallel directly to a managing director weighing the ask. Agencies that positioned themselves narrowly as search specialists when social advertising first emerged had to retrofit that capability later, under competitive pressure, and it cost more than building it ahead of demand would have. The same choice is in front of agencies now, with conversational AI in place of social. Self-serve advertising inside conversational AI platforms is already live and already scaling. If your agency has no structured approach to it, you are already placing ads inside these surfaces, either by accident or through generic settings built for a different channel. The audit phase tells an agency what it's already doing inside AI surfaces, before it decides what to do about it on purpose, as a short, low-risk first step.