OpenRTB for AI: A Technical Deep Dive

OpenRTB (Open Real-Time Bidding) is the protocol that powers the programmatic advertising industry. For years, it has enabled real-time auctions for display and search advertising. PromptBid adapted OpenRTB 2.6 for conversational AI, extending the standard to support a new advertising channel. This post dives into the technical architecture and explains how OpenRTB enables scale and interoperability in conversational advertising.

What is OpenRTB?

OpenRTB is a protocol specification maintained by the Interactive Advertising Bureau (IAB). It defines a standard language for real-time bidding in programmatic advertising. When a user loads a webpage with ad inventory, the publisher sends a bid request to an ad exchange. The exchange forwards the request to multiple demand partners (advertisers), who evaluate whether they want to bid on the impression. The highest bidder wins and their ad is displayed.

This all happens in milliseconds. The speed and standardization is what makes it work at scale. Without a common protocol, every publisher would need custom integrations with every advertiser. With OpenRTB, anyone can plug into the ecosystem using the same interface.

The standard has been refined since 2010 and powers hundreds of billions of dollars in annual advertising spend. The current version, 2.6, is mature and widely supported by advertising technology platforms.

Core Components of OpenRTB

An OpenRTB transaction involves several key components:

Bid Request

The publisher sends a structured request to the ad exchange containing information about the impression. This includes details about the user (cookies, IP address, device type), the content context (page URL, content category), and inventory details (ad format, size, placement).

Bid Response

Demand partners respond with a bid price and creative (the actual ad content). The ad exchange evaluates all responses and selects the highest bidder.

Impression Event

Once an ad is served, an impression is recorded. If the user clicks the ad, a click event is recorded. If the user completes a desired action on the advertiser's website (purchase, signup), a conversion event is recorded.

Tracking URLs

The bid response includes tracking URLs that fire when events occur. This allows the advertiser and publisher to track performance and measure ROI.

Adapting OpenRTB for Conversational AI

The core auction flow works the same in conversational advertising, but the content of the bid request changes fundamentally. In display advertising, the request includes page URL and cookies. In conversational advertising, the request includes conversation context.

The Standard Bid Request (Display)

{
  "id": "1234567890",
  "imp": [{
    "id": "1",
    "banner": {
      "w": 728,
      "h": 90
    }
  }],
  "site": {
    "id": "123",
    "name": "Example Site",
    "domain": "example.com",
    "cat": ["IAB1"],
    "page": "http://example.com/article"
  },
  "user": {
    "id": "user123",
    "buyerid": "buyer123"
  }
}

The PromptBid Bid Request (Conversational AI)

{
  "id": "conversational-bid-12345",
  "imp": [{
    "id": "1",
    "native": {
      "request": "..."
    }
  }],
  "app": {
    "id": "app-123",
    "name": "My AI Assistant",
    "bundle": "com.example.aiapp",
    "cat": ["IAB1-1"]
  },
  "ext": {
    "conversation_context": {
      "messages": [
        {
          "role": "user",
          "content": "I need a new laptop for video editing"
        },
        {
          "role": "assistant",
          "content": "For video editing, I'd recommend..."
        }
      ],
      "intent": "product_research",
      "topics": ["laptops", "video_editing", "professional_equipment"],
      "urgency": "medium"
    }
  },
  "user": {
    "ext": {
      "no_cookies": true,
      "context_only": true
    }
  }
}

The key differences:

Conversation Context: Instead of page URL, we include the actual conversation. Advertisers can see exactly what the user is asking about.
Intent Recognition: We extract and include the user's intent ("product_research", "comparison", "purchase_intent") rather than relying on behavioral signals.
Topic Extraction: We identify relevant topics from the conversation to enable category-based targeting.
No Cookies: The user object explicitly indicates that no cookie-based tracking is available. Targeting must be contextual.
Native Format: We use the native ad format rather than banner, because conversational ads are text-based and match the conversation interface.

The Auction Flow

The real-time auction flow in PromptBid proceeds as follows:

1. User Input

The user sends a message to the AI application. The application processes the message to extract intent, topics, and conversation context.

2. Bid Request Generation

The application constructs an OpenRTB bid request containing the conversation context and sends it to the PromptBid exchange.

3. Demand Routing

PromptBid routes the bid request to demand partners whose targeting criteria match the conversation context. If a demand partner targets "laptops" in "product research" contexts, they receive the bid request.

4. Bidder Evaluation

Each demand partner evaluates the bid request. They run algorithms to determine: Is this a valuable impression? What price is it worth? The demand partner submits a bid response with a price and creative.

5. Auction & Winner Selection

PromptBid evaluates all responses. The bid with the highest price wins (with certain quality and frequency constraints applied).

6. Ad Delivery

The winning ad creative is returned to the AI application and displayed to the user. All this happens in under 100 milliseconds.

7. Event Tracking

The bid response contains impression, click, and conversion tracking URLs. These fire when events occur, enabling performance measurement.

Latency Requirements & Architecture

Sub-100ms latency is critical for conversational interfaces. To achieve this, PromptBid implements several optimizations:

Parallel Processing

Bid requests are sent to multiple demand partners simultaneously. We don't wait for each response sequentially—we collect responses as they arrive and make a decision once we have enough data.

Timeout Handling

If a demand partner's response doesn't arrive within 80ms, we don't wait. We proceed with bids from partners who responded faster.

Edge Deployment

PromptBid runs on distributed edge infrastructure closer to users and applications. This reduces network latency compared to centralized auction houses.

Caching & Precomputation

Advertiser bid strategies and creative assets are cached. We don't need to compute everything from scratch for every request.

Interoperability & Ecosystem Value

The most important aspect of using OpenRTB is interoperability. By adopting the standard, PromptBid can connect with the existing ecosystem. Demand partners already using OpenRTB can plug into conversational advertising without building a new integration. Advertisers can use their existing bidding algorithms and platform integrations.

This is why standards matter. OpenRTB represents decades of standardization work in the display advertising industry. By extending it rather than creating a proprietary protocol, PromptBid makes it possible for conversational advertising to scale with the existing advertising infrastructure.

Future Extensions

OpenRTB 3.0 is under development and will introduce new capabilities: better support for first-party data, improved privacy controls, and expanded context types. PromptBid will continue evolving with the standard, ensuring compatibility while optimizing for conversational AI use cases.