AI SurfacesAdvertisersSparkBlogAboutContact Us
← All Posts Technical Overview

How Conversational Ads Work: From Ad Request to Measurement

Conversational ads work by asking an ad server, turn by turn, whether a conversation should carry an ad, rendering a brand-controlled interactive unit only when the intent fits, and then measuring both the interaction and the conversion that follows. This page walks through that sequence for Spark on the Adgentek Ad Server.

New to the category? Start with what conversational AI advertising is, then come back here for the mechanics.

Step 1: The AI app requests an ad

A conversational ad starts with a request from the AI app, not a page load. After the assistant responds, the app passes the conversation turn to the Adgentek Ad Server through its integration, usually the JavaScript SDK, along with the placement it wants to fill. The app decides where an ad is allowed to render. The ad server decides whether this turn should carry one at all.

Step 2: The ad server classifies intent

The ad server reads the conversational context and classifies the turn into one of nine intent buckets: purchase, transactional, consideration, comparison, research, informational, navigational, support or entertainment. It also extracts entities such as product categories, brands and price ranges. Matching runs on what the person is asking about, not on cookies or a page URL.

Step 3: The ad server decides whether to serve, and what

If the turn carries commercial intent and the publisher's rules allow an ad, the ad server matches it against live advertiser demand and picks a format that fits. A comparison or consideration turn can carry an interactive Spark unit. An informational turn may carry nothing. Publishers control which formats appear, which categories are blocked, frequency caps and floors.

Step 4: The unit renders with clear disclosure

The conversational ad appears inside the app's own interface as a branded unit with a sponsorship label, a header, suggested questions and a free-text field. It is a self-contained card next to the assistant's answer, not text mixed into it, so the person can tell the brand's content from the assistant's. See AI advertising vs display for why display units do not transfer to this setting.

Step 5: The person interacts with the ad

The person taps a suggested question or types their own, and the unit answers from the brand's approved content and guidelines, either fixed copy the brand wrote or dynamic creative generated within those guidelines where the brand opts in. Either way, the brand sets the claims, prices and comparisons the unit can make. Each interaction is also a qualification step: the person tells the brand what they need while getting answers. Spark caps interactions per session so the unit stays an ad rather than a second chatbot. Read more on brand safety in AI advertising.

Step 6: Intent signals are captured without raw prompts

Each interaction becomes a structured signal such as product type, budget range, feature priorities and timeline. Adgentek passes derived signals, including intent classification, categories and entities, to demand partners, and raw user prompts are not shared. Intent data from AI ads shows what that signal looks like for a brand.

Step 7: Conversion and measurement

When the person is ready, the unit offers a call to action that fits the category, such as Build and Price, Shop Now, Book a Consultation or Find a Dealer. Impressions and clicks are tracked from the render in the app. Conversions are measured on the advertiser's side: advertisers send conversions server to server through the Conversions API, with last-click attribution when the click identifier is passed back. For outcome-based buys, write the attribution window and source of truth before launch, as covered in how to write an outcome brief.

Where conversational ads run

Spark runs inside independent AI apps, chatbots and publisher sites that integrate the Adgentek Ad Server, so brands are not tied to a single assistant's ad program. For the buying options, see AI chatbot advertising strategy. For how the format compares with a sponsored link, see conversational ads vs sponsored links.

See Spark in action across industries or contact hello@adgentek.ai.

Publishing an AI app or chatbot?

Integrate the Adgentek Ad Server and start earning from your AI conversations, live demand, a few lines of JavaScript to integrate, documented at docs.adgentek.ai.

Get Started Free →

Common questions

How do conversational ads work?

The AI app asks the Adgentek Ad Server, after each response, whether the turn should carry an ad. The ad server classifies the turn's intent into one of nine buckets, matches it against live advertiser demand within the publisher's rules, and returns a format that fits, such as an interactive Spark unit. The person interacts with brand-approved answers, and the interaction and any conversion are measured.

What intent signals does a conversational ad capture?

A conversational ad captures intent classification, IAB content categories, named entities such as brands, products and locations, and the signals people give inside a Spark unit, such as product type, budget range, feature preferences and timeline. Only derived signals are transmitted to demand partners, and raw user prompts are not shared.

How is a conversational ad measured?

Impressions and clicks are tracked from the render in the AI app, and engagement inside the unit shows which questions people chose. Conversions are measured on the advertiser's side: advertisers send conversions server to server through the Adgentek Conversions API, with last-click attribution when the click identifier is passed back.

How does an AI app add conversational ads?

An AI app adds conversational ads through the Adgentek JavaScript Web SDK: a script tag, a container element in the app's own interface, an init call, and a call after each AI response. The docs at docs.adgentek.ai walk through it, and a coding agent can follow them.

Ready to get started?

Contact Us
hello@adgentek.ai
NVIDIA Inception Program