The average display banner earns a 0.46% click-through rate — not because the media plan is wrong, but because the format is mute. A static rectangle cannot do the one thing a salesperson does: respond.Conversational ads change that premise. Instead of an image, the creative is an AI agent — the ad answers questions, handles objections, and converts interest into a qualified lead, inside the ad slot, without the reader leaving the page.
What exactly is a conversational ad?
A conversational ad (also called a talking display ad or, our term, a Legate Ads™ agent) is a standard display creative with two states. Before engagement it behaves like any banner: lightweight, brand-styled, clearly badged as an AI. On one deliberate tap it becomes a live conversation with an agent that represents exactly one advertiser — its products, its pricing, its voice. The visitor asks real questions ("do you service my zip code?", "what does installation cost?") and the agent answers from the brand's approved catalog, then captures contact details when interest is real. Every conversation is transcribed.
Haven't we tried this before? (Yes — AdLingo)
Google's Area 120 shipped conversational display ads in 2018 under the AdLingo brand: chat inside the banner, bought programmatically. It wound down by 2022, and the post-mortem is instructive. The conversations were scripted decision trees that collapsed the moment a real customer asked a real question, and production was so laborious Google built a tool just to make bot-building ten times faster. The format was right; the intelligence wasn't there. Modern large language models close exactly that gap — an agent can now hold an unscripted conversation grounded in a brand's actual catalog, at a cost measured in cents.
How talking display ads run in Google Ad Manager
This is the part ad-ops teams care about: a well-built conversational ad is astandard third-party creative tag. In Google Ad Manager it traffics like any other creative — paste the script tag into a creative, attach it to a line item, target and pace as usual. The same tag fans out across a media plan through CM360, with per-site placement reporting. There is no SDK, no publisher integration project, no new auction mechanics. Before any engagement the unit stays within IAB weight guidance; the conversation only loads on a deliberate user action.
What "governed" means — and why it's the whole product
The first question every advertiser asks is some form of: what stops it from saying something stupid?A tribunal has already held an airline liable for promises its chatbot invented, so the answer cannot be "trust the model." A governed brand agent is engineered from that ruling backward:
- The catalog is the leash. The agent may only claim facts from a document the brand approved.
- A streaming output filter checks every sentence as it's generated — off-catalog prices, invented discounts, guarantees, competitor mentions, and off-domain links are blocked mid-sentence, replaced with a safe hand-off, and logged for review.
- Disclosure is built in: a visible AI badge on every unit and a first-message disclosure, matching the strictest of the emerging US state rules and the EU AI Act.
- No surveillance: no cookies, no profiling — the unit knows only what the visitor chooses to type.
- An instant kill switch and full transcripts make the whole channel auditable.
The governance layer is not a feature bolted onto the ad — it is the reason the format is deployable at all. Our white paper, "Ads That Talk", covers the architecture in depth, including the five objections agency leaders raised when we asked them to attack it.
The economics finally work
The market already prices conversation at a premium: sponsored one-to-one messages trade at $0.50–$3.00 per delivered message, while AI inference for a whole engaged conversation now costs pennies. That spread is the opportunity. For advertisers, a conversational unit is priced like premium rich media with a per-conversation or per-qualified-lead layer; for publishers, it is a new top-tier format their sales teams can sell above video CPMs. And the by-product may be the most underrated asset: transcripts — thousands of verbatim consumer questions per campaign, the consumer research a strategy team would otherwise commission separately.
Where conversational ads fit — and where they don't
Honest positioning matters more in a new format than anywhere else. Conversational creatives are theresponse layer of a media plan, not a replacement for brand creative. They fit considered purchases where buyers have real questions: home services, healthcare, legal, B2B, financial products, education. They are the wrong tool for impulse CPG, pure awareness plays, and luxury storytelling — if a campaign's job is to be seen rather than answered, buy reach, not conversation.
Trying one is the fastest way to judge one
The live demo on our Legate Ads™ page is the actual product — a real ad unit served by our engine, representing us. Talk to it, try to trip it up, watch it decline to invent a discount. If it holds up, the four-week pilot runs with exit criteria agreed before launch: engagement rate, cost per qualified lead, transcript quality. Clear the bar, scale; miss it, both sides learned something for the price of a test.
Renier P. du Plessis is the founder of Agentic Tech LLC (Atlanta) — 15 years of engineering inside an ad-supported national news publisher, now building governed AI agents for advertising. Member of the Claude Partner Network.
