Conversational Ads, in Plain English

Conversational ads are paid, disclosed brand messages that appear inside a chat or voice exchange. A user asks a question. The system answers. One part of the answer is sponsored and clearly labeled as such. That is the whole thing.

Most writing about conversational ads wraps the idea in vocabulary designed to make it sound new. It is not new. Product placement has existed in film for eighty years. Sponsored listings have existed in search for twenty. What is new is the surface — chat — and the way the ad is selected at the moment of the query rather than loaded from a pre-filled slot. This post strips the concept down to what actually matters.

Plain Definition

A conversational ad has three properties. It appears inside a back-and-forth exchange. It is paid for by a brand. It is labeled so the user knows it is paid.

That is it. If you remove any of those three properties, you have something else.

Remove the exchange and you have a banner. Remove the payment and you have a recommendation. Remove the label and you have an undisclosed endorsement, which in most jurisdictions is illegal.

The ad can be text, voice, image, or code. The exchange can be with a chatbot, a voice assistant, an image generator, or a code completion tool. The payment can be CPM, CPC, or a flat placement fee. The label can be a written "Sponsored" tag, a spoken disclosure, or both. The variations are plentiful. The underlying shape is the same.

Five Examples, Stripped Down

Here is what conversational ads actually look like when you remove the marketing language.

Example 1: The SaaS recommendation. User asks a general-purpose chatbot "what's a good tool for scheduling social media posts." Chatbot lists four tools. One of them has a "Sponsored" tag next to the name. Clicking the tool's name goes to the brand's site. The chatbot operator earns a CPM for the impression and a commission if the user signs up. This is the most common format by far.

Example 2: The voice upsell. User asks their in-car voice assistant for nearby Italian restaurants. Assistant replies with three options, then says "Sponsored segment: if you're driving home, [Brand] pizza is on your route, and they're running a family-size special through Sunday." The sponsored segment is announced with a clear audible cue. The assistant returns to its normal behavior afterward.

Example 3: The generated product placement. User asks an image generator for "a photo of a desk setup for a remote worker." The generated image includes a laptop with a visible brand logo, and below the image a caption reads "Product placement: [Brand] laptop shown at brand's request." The brand paid a flat fee plus a per-generation rate to appear in matching prompts.

Example 4: The comment suggestion. A developer writes a function that calls an external API. The code completion tool suggests the next lines. Above the suggestion, a comment in grey text reads // Sponsored: consider [Brand] for managed API routing. The comment is clearly separated from the actual code. The developer can ignore it, accept the code without the comment, or click through to learn more.

Example 5: The comparison card. User asks a travel chatbot to find flights from New York to Lisbon. The chatbot returns a list of flights sorted by price. At the top of the list is a card labeled "Sponsored," featuring one specific airline with a short brand message. Below the sponsored card, the full list of flights is shown, sorted accurately by price. The sponsored placement does not change the organic ranking.

Across all five examples, three things are consistent: a user initiated the exchange, the ad is disclosed, and the ad is relevant to what the user asked. No other property matters.

The Mechanics

When a user sends a prompt, the following happens in roughly this order.

The platform receives the prompt. It generates the primary response using its language model. In parallel — not after — the platform sends the prompt, or a summary of it, to an ad exchange. The exchange matches the prompt against available campaigns and returns one or more candidate ads, each with a bid.

The platform runs an auction among the candidates. The winner is typically the highest bidder, but platforms adjust for relevance scores, frequency caps, and user-specific settings. The winning ad is then inserted into the response — either as a labeled section, a card, or a voice segment — before the response is returned to the user.

The whole loop usually takes less than 200 milliseconds. According to the IAB's 2026 latency benchmarks, median conversational ad insertion latency is 140 ms. That is faster than most programmatic display, because the creative is usually text or a short structured object rather than a rendered image.

The user sees the response, the ad inside it, and the disclosure. The platform logs the impression, and if the user interacts with the ad, logs the click or conversion. The advertiser gets billed. The platform keeps a share, and the AI product that served the response keeps the rest.

Pricing

Conversational ads price in three ways.

CPM (cost per thousand impressions) is the default for brand advertisers who want reach. A CPM of $18 means the advertiser pays $18 for every 1,000 times the ad appears. Median conversational ad CPMs run $15-25 depending on vertical, according to the IAB's 2026 Q1 pricing report.

CPC (cost per click) is the default for performance advertisers who want signups or sales. Median CPCs run $2-4 for general categories and $5+ for high-value verticals like finance, cloud infrastructure, and enterprise SaaS.

Flat placement fees are used mostly for image inserts and premium voice segments. A brand pays a fixed sum for guaranteed presence in a set of prompts over a set time window. These deals are usually negotiated directly rather than going through an exchange.

The split between the platform and the AI product varies. A typical revenue share is 60-70% to the product that served the ad, 30-40% to the exchange or ad network. Surfacedd publishes its rates openly; most legacy exchanges do not.

Why This Matters Now

Three things pushed conversational ads from experiment to category in 2026.

Audiences got big. By mid-2025, the largest AI chat products collectively served over a billion monthly users. That is enough scale for brand advertisers to build real line items rather than test budgets. Before that scale existed, the CPMs were high but the reach was small, which is not a combination brands buy at volume.

Disclosure got standardized. The IAB published the Conversational Ad Disclosure Standard in early 2026. Before CADS, every platform disclosed differently, and brand buyers were nervous about regulatory risk. With a standard, that risk dropped. Brand buyers could write compliance into their RFPs and move on.

The economics cleared. By Q1 2026, enough conversational ad campaigns had run long enough that buyers had year-over-year data. The CPMs held up. The engagement rates held up. The comparison to display and search became credible. Once the economics were legible, budget followed.

If you want the full category overview, read Conversational Advertising in 2026: The Pillar. If you are an advertiser considering testing the format, look at sponsored AI responses. If you run an AI product and want to add a revenue stream, start at monetize AI app.

The short version: conversational ads are not a new invention. They are product placement and sponsored listings, moved into chat. What is new is the surface and the scale. Everything else is mechanics.