Conversational Video Ads Turn the Ad Unit Into the Destination

Three Platforms Made the Ad Unit Conversational in Fifteen Days

Conversational video ads arrived this month. Google's Business Agent now lets viewers ask product questions beside a YouTube video ad, OpenAI is testing Sponsored Agents inside ChatGPT, and Amazon rebuilt its ad platform around an agent. The pattern is blunt: the ad unit is becoming the destination, and the product feed — not the edit timeline — now answers the questions your creative provokes.

The moves came in a single burst. On September 16, OpenAI announced it was testing Sponsored Agents, which let people start a conversation with a business-sponsored agent after clicking an ad in ChatGPT. On September 24, Google's Demand Gen Drop introduced Business Agent for YouTube Ads, where viewers can engage with conversational AI alongside video ads with product feeds and get immediate answers without leaving the viewing context. On September 29 at unBoxed, Amazon renamed its unified buying platform Amazon Ads Agent and rebuilt it around agentic AI. Three companies, three surfaces, one shared bet.

The three bets are not identical, and the differences matter for planning. Google's experiment puts the conversation next to a video ad while the viewer is still watching. OpenAI's puts it after the click, inside a chat product where the user is already typing. Amazon's is aimed at the advertiser's side of the table — it automates planning, targeting and analytics rather than talking to shoppers. Together they mark one shift: the industry's largest ad platforms now treat a conversation as a first-class ad format, not a support widget buried behind a help link.

Inside the YouTube Beta: What Business Agent Actually Does

Google's version is a limited US beta, and the eligibility rules say a lot about who Google thinks this is for. Advertisers need an active Demand Gen campaign linked to a Merchant Center product feed, spending at least ten dollars a day, serving English-language mobile in-stream inventory. Shorts placements are excluded. Google's team enables the experience on existing eligible campaigns once a brand signs up, so there is no new campaign type to build and no re-editing of existing video assets to pass the gate.

The same September Drop collapsed steps elsewhere in the funnel. Consumers can now reach a landing page with a single tap on image ads on YouTube Shorts and Gmail; on Shorts, the one-click path applies when advertisers upload 9:16 full-bleed image assets. Google reports that advertisers adding Gmail to Demand Gen campaigns see image creatives achieve a 40 percent increase in conversions at the same ROI, a figure drawn from Google's internal Gmail Ads data from February 2026 and worth treating as directional rather than as a planning number.

Rollout is gradual in the United States first, with Australia and Canada expected to follow in early 2027. For everyone else this is a preparation window, not a wait-and-see window — because the hard part is not turning the feature on.

The Creative Brief Now Includes the Answer

A video ad that opens a conversation has to earn two things: attention and a question. The creative's new job is to provoke the specific product question a viewer actually has — does it fit, how does it ship, what does maintenance cost — and trust the agent to answer it well. That only works if the answer exists in the data. A sparse catalog fails the moment a viewer asks something the AI cannot extract from the feed, which means the feed is now part of the creative, not adjacent to it.

That preparation is the same discipline behind {{link}}: machine-readable metadata that a system other than the edit suite can read, quote and act on. Google's merchant documentation already supports conversational attributes — product questions and answers, document links, related products and variant information — supplied through a supplemental feed. Teams that did the structured-data work for search will recognize the shape of the task immediately, because it is the same task pointed at a conversation instead of a results page.

None of it works without {{link}}, the unglamorous data spine that lets an agent answer a product question accurately on the brand's behalf. The practical checklist is short: audit the facts the ad will be asked about, record who owns each fact and how fast it changes, collect real pre-purchase questions from support tickets and reviews, and set response boundaries — approved claims, exclusions, and the moment a human or a product page takes over.

That preparation is the same discipline behind structured data for video SEO: machine-readable metadata that a system other than the edit suite can read, quote and act on.

None of it works without agentic AI data readiness, the unglamorous data spine that lets an agent answer a product question accurately on the brand's behalf.

Sponsored Agents Move the Conversation After the Click

OpenAI's test sketches the same shift from the other side. Its launch example is a shopper who sees a dining table in an ad and wonders whether it fits the room, how many people it seats, how to care for the finish. Instead of clicking through to a product page, the user can open a clearly labeled conversation with a business-sponsored agent, ask follow-ups, and follow a link to the brand's site when ready. OpenAI is explicit that the sponsored thread is separate from ChatGPT's independent answers and from the user's original conversation, which keeps the sponsored experience from quietly rewriting what the model says.

What OpenAI has not published is just as important: how an advertiser builds the agent, what data it can access, how it is billed, and how its claims are reviewed. Until those answers arrive, the safe reading is that a Sponsored Agent is a knowledge base wearing an ad unit — and the preparation work is closer to writing a support documentation set than writing ad copy. A wrong answer from a sponsored agent is a wrong statement made in the brand's voice, a different risk category than a low-CTR headline, and it will land on the same desk that approves legal supers today.

Amazon Makes the Agent the Platform Itself

Amazon's announcement is less about the ad unit and more about who runs the campaign. Amazon Ads Agent is the new name for the unified platform that merged the Ads Console and the DSP, now powered by agentic AI across three campaign types: sponsored ads, Display Video and Audio, and Full-Funnel Campaigns. In a Full-Funnel campaign the advertiser provides budget, products and creative, and Amazon's AI handles planning, execution and continuous optimization, with shared signals flowing across formats so upper-funnel investment improves lower-funnel performance.

The detail that lands hardest for video teams is buried in the creative tooling: with built-in creative tools, advertisers can make TV-quality video and display images from existing product detail pages. That is the commerce-side version of the same agent-first logic — the platform generates the video from the feed, collapsing a production brief into a catalog operation. Amazon also reported that advertisers using natural-language targeting recommendations saw more than 25 percent additional unique customers while lowering cost-per-impression by more than 10 percent on average, and new media-planning, AI-targeting and natural-language analytics capabilities will roll out over the coming months.

What Conversational Video Ads Change This Quarter

How fast to move depends on where you sit — the split mirrors {{link}}, where smaller spenders lean on agentic AI for creative testing and pre-planning while large buyers point it at inventory discovery and evaluation. Teams with robust product feeds and performance-oriented catalogs should treat the YouTube beta as a first experiment: pick one product with a well-documented buying decision, write down the questions the agent must answer, and define the failure cases you will review before launch.

Two preparation steps cost nothing and pay off regardless of platform. First, reconcile the video and the feed: every claim a cut makes on screen must be backed by feed data the agent can retrieve, or the conversation collapses into a support ticket. Second, write the video to open a conversation — an implicit invitation to ask, instead of a hard final-frame call to action — because a viewer who asks a question has already done what the ad used to need a click for.

For teams that already treat {{link}} as the distribution layer for finished work, the conversational unit adds a new acceptance test: the asset has to open a conversation, not just hold attention. Measurement is the open question — neither Google nor OpenAI has said how in-ad conversations will be reported or whether they count toward conversions. Log the qualified questions your beta generates, compare them against search and support query logs, and you will know before the dashboards do whether conversational video ads deserve a line in next year's budget.

How fast to move depends on where you sit — the split mirrors buyer-size divergence, where smaller spenders lean on agentic AI for creative testing and pre-planning while large buyers point it at inventory discovery and evaluation.

For teams that already treat AI video delivery platforms as the distribution layer for finished work, the conversational unit adds a new acceptance test: the asset has to open a conversation, not just hold attention.