How to make a ChatGPT Ads AI agent

ChatGPT Ads is a new advertising surface. OpenAI is rolling out sponsored results and ad placements within ChatGPT conversations, giving advertisers access to users at the moment they are asking questions, researching products, and evaluating options. The intent signal is different from Search (where the query is explicit) and different from social (where the algorithm infers interest). In ChatGPT, the user is having a conversation, and the context of that conversation is what determines whether your ad is relevant.

An AI agent for ChatGPT Ads needs to solve the same core problems as any other channel agent: measure actual business impact through cross-channel attribution, configure the right goal and funnel metrics, and operate within guardrails. But the channel is early enough that there are specific challenges around measurement maturity, creative format, and optimization levers that change how you should build it.

‍

The attribution foundation

‍

ChatGPT Ads is a new platform with new tracking infrastructure. Like every other ad platform, it will build its own attribution model designed to maximize credit for itself. Your agent should not rely on it.

Your agent should use your own cross-channel attribution model. Every incoming visitor is parsed and identified by traffic source, channel, and campaign. Your attribution model assigns credit to each touchpoint, deduplicated across both paid and organic channels. The key is having campaign-level attribution on all revenue movement metrics: new customers, expansion, churn, contraction, reactivation. A tool like Roadway handles this out of the box, or you build it internally.

For ChatGPT Ads specifically, capture the click identifier or UTM parameters that OpenAI passes on ad clicks, store them alongside the user record, and join to revenue data in your warehouse. This is the same pipeline you use for every other channel. The earlier you set this up, the sooner your agent has real data to work with instead of relying on OpenAI's reporting alone.

‍

What makes ChatGPT Ads different

‍

It will be the highest intent platorm we've ever seen. ChatGPT users are in a conversational context. They are not scanning a list of search results or scrolling a feed. They are mid-conversation, asking follow-up questions, comparing options, and refining their thinking in real time. This changes the ad experience in a few ways.

Intent is contextual, not keyword-based. On Google Search, you bid on keywords. On ChatGPT, placement is determined by the conversation context. A user asking "what is the best project management tool for a remote team of 15" is expressing rich intent, but it is embedded in a conversation, not a search query. The targeting model is fundamentally different, and the optimization levers available to advertisers will evolve as OpenAI builds out the platform.

The creative format is native to conversation. Ads in ChatGPT need to feel like a natural part of the conversation rather than a banner interrupting it. Early formats include sponsored recommendations within responses. The creative challenge is making the ad feel like a helpful answer rather than an interruption. This is closer to native advertising than to traditional display or search ads.

The platform is early. The targeting options, bidding mechanisms, reporting, and API access are all less mature than Google or Meta. This means more manual work initially, fewer optimization levers, and less granular data. The agent needs to work with what is available and adapt as the platform matures.

‍

Goal, funnel, guardrail, memory

‍

Goal metric. Revenue or paid customers attributed to ChatGPT Ads through your cross-channel model. Be realistic about volume early on. This is a new channel and scale will be limited initially.

Funnel metrics. Ad impressions within conversations, click-through rate, landing page CVR, sign-up or trial, activation, paid conversion. Track by conversation topic or targeting category if the platform provides that breakdown. The funnel between impression and click matters more here than on most channels because the conversational context heavily influences whether a user engages with an ad.

Guardrails. Maximum CPL before flagging, minimum data volume before making structural changes (this channel will have less volume than Google or Meta, so the threshold for statistical reliability is important), budget ceiling as a percentage of total paid spend (appropriate for a channel still being evaluated), minimum evaluation window before judging performance.

Memory. Every campaign run on ChatGPT Ads: targeting configuration, creative format, performance data, and how it compared to the same period on other channels. Platform changes (new targeting options, new ad formats, API updates) and how they affected performance. This history matters more on a new channel because the platform itself is changing rapidly and the agent needs to track what worked under which platform conditions.

‍

Tools and skills

‍

Your agent needs two types of inputs: tools (API integrations that let it read and write data) and skills (markdown files that give it context and decision-making frameworks).

Tools (APIs):

• OpenAI Ads API (as available) - campaign performance data, impression and click metrics, targeting configuration. Write operations for campaign management as the API matures. Access requirements and scopes will evolve as the platform develops
• Attribution / data warehouse - cross-channel attribution data joined to revenue, with ChatGPT Ads defined as a distinct channel
• Analytics API (GA4 or equivalent) - referral traffic from ChatGPT ad clicks, landing page behavior, conversion tracking
• CRM - customer records with acquisition source, revenue and LTV for customers acquired through ChatGPT Ads

Skills (markdown files):

• Channel evaluation framework - how to evaluate a new channel during early testing: minimum test duration, minimum budget, what success looks like at low volume, when to scale vs. when to pause
• Conversational ad creative guide - what works in a conversational ad context vs. search or social, tone and format guidelines, how to make sponsored content feel like a helpful recommendation rather than an interruption
• Targeting strategy - which conversation topics and user contexts are most relevant to your product, how to map your ICP to ChatGPT's available targeting options
• Cross-channel comparison framework - how to compare ChatGPT Ads performance to established channels when the volume is lower, adjusting for statistical significance at smaller sample sizes
• Platform change tracking - how to log and respond to platform updates (new ad formats, new targeting, API changes), what to re-test when the platform changes
• Landing page optimization for conversational traffic - how users arriving from a ChatGPT conversation differ from search or social traffic, landing page expectations and messaging alignment

‍

The three levels

‍

Monitoring. Campaign performance against your attributed revenue data, CTR and conversion rates by targeting category, landing page performance from ChatGPT traffic specifically, cost efficiency compared to other channels, and platform changes that may affect campaign setup. Because volume will be lower than established channels, the agent should flag when data is insufficient for reliable conclusions rather than making recommendations on thin signal.

Planning. Whether to increase or decrease ChatGPT Ads investment based on attributed performance relative to other channels, which targeting categories or conversation contexts are producing the best results, creative adjustments based on CTR and conversion data, and the broader portfolio question: does this channel earn a larger budget share or is spend better allocated elsewhere? The agent uses its cross-channel attribution data to answer this honestly.

Action. Campaign adjustments via the OpenAI Ads API as write operations become available: budget changes, targeting modifications, campaign pause/enable, bid adjustments. As with any new platform, the API capabilities will expand over time. Start with read-only monitoring and add write operations as the platform matures. Manifest-and-approval before any changes execute.

‍

How to set it up in Roadway

‍

1. Create a new Coworker
2. Filter for the channel
3. Choose your goal metric (this is what your agent will optimize for)
4. Choose the funnel metrics that lead to your goal metric
5. Choose your guardrail metrics and define their limits
6. Choose your refresh schedule
7. Publish

Work with AI Coworker to plan and execute campaigns. Reach out to us if you need any help - happy building: contact@roadwayai.com

‍

FAQ

‍

How are ChatGPT Ads different from Google Search Ads?

Google Search Ads target explicit keyword queries. ChatGPT Ads are placed within conversational contexts where the user is asking questions, comparing options, or exploring a topic through dialogue. The intent is often richer (a full conversational thread vs. a few keywords) but the targeting and optimization infrastructure is less mature. The ad format is also different: ChatGPT Ads need to feel like a natural part of the conversation, not a search result listing.

Is ChatGPT Ads worth testing right now?

If you have budget allocated for channel testing and your attribution infrastructure is in place, yes. Early movers on new ad platforms often benefit from lower competition and lower CPMs before the market matures. The key is setting appropriate expectations: volume will be lower than established channels, the platform will change frequently, and you need your own attribution data to evaluate real performance rather than relying on the platform's reporting.

How do you measure ChatGPT Ads performance accurately?

The same way you measure every other channel: through your own cross-channel attribution model. Capture the click identifier or UTM parameters from ChatGPT ad clicks, store them with the user record, and join to revenue data. Compare attributed performance to what OpenAI's reporting shows. Trust your model, not the platform's self-reported numbers. This is especially important on a new platform where the attribution methodology is still being developed.

What kind of creative works best for ChatGPT Ads?

Early signals suggest that conversational, helpful creative outperforms promotional creative. The user is in the middle of a conversation and is looking for answers, not offers. Ads that read like a knowledgeable recommendation (here is a tool that solves the specific problem you are discussing) perform better than ads that read like traditional ad copy (sign up now, 50% off). Think of it as native content within a conversation rather than a display ad.

How much budget should you allocate to ChatGPT Ads?

Treat it as a channel test. Allocate enough budget to generate statistically meaningful data over a defined test period (typically four to eight weeks), but do not reallocate significant budget from proven channels. A common approach is 5-10% of your experimental or testing budget. Let your attribution data tell you whether to scale, hold, or exit. The agent's cross-channel comparison is what makes this decision data-driven rather than speculative.

‍