Home AI One Chatbot’s Journey To Introducing Ads That Don’t Suck

One Chatbot’s Journey To Introducing Ads That Don’t Suck

SHARE:
Our robot friends
Our robot friends

Ads in chatbots are becoming standard practice, and advertisers and AI companies are scrambling to determine how to make the most out of this new format.

They need to deliver relevant, contextual ads that don’t feel invasive or overly repetitive and fit into the chatbot’s interface while still being clearly distinct from the organic chat responses.

Since LLMs are a mostly untested channel, though, the ones choosing to introduce ads don’t have much data to influence their strategy.

Luzia, an AI assistant mainly used in Latin America, struggled to monetize through traditional networks like Google Ads. Google had the largest inventory and was easy to integrate, but the ads were better suited for the platforms they were originally designed for, such as search or gaming, said Natalia Solano Gutierrez, Luzia’s head of product.

Luzia wanted ads that were specifically designed for an AI chat interface, said Gutierrez, rather than a format “optimized for another experience.”

It found a solution in generative AI ad network Koah, which launched in 2025 to help AI publishers (i.e., chatbots) monetize by deploying relevant ads designed for chatbot interfaces.

Strategic thinking

Luzia launched in 2023 as a chatbot within WhatsApp, linked to a phone number like a human contact would be. It incorporated ads in the app after about a year, and while the team determined that there was “an appetite” for ads, it needed a better targeting strategy, said Gutierrez.

The strategy at the time was almost nonexistent and involved scraping product listings from Google and showing them to Luzia’s users at random, she said.

In 2024, Luzia launched as its own app and began experimenting with in-app ads in 2025, starting with Google Ads.

Luzia wasn’t satisfied with the “fixed format” of Google Display Ads, Gutierrez said. It ultimately settled on Koah as its ads partner because Luzia’s team preferred how Koah’s ads allowed them to play with elements like coloring, size and position, and included interactive formats like polls.

Subscribe

AdExchanger Daily

Get our editors’ roundup delivered to your inbox every weekday.

Koah benefits from the array of ad formats, too; its monetization model blends cost-per-click (CPC) with cost-per-thousand views (CPM) and a newer tactic, cost-per-engagement (CPE), said Choi. ChatGPT’s model, for comparison, is solely CPM.

Koah has found that the best-performing ads are native, text-based ads that are “clearly demarcated” and not embedded directly in the chat, according to Co-Founder Nic Baird. Most LLMs seem to agree. ChatGPT and Copilot’s ads, for instance, look relatively similar to search ads, presumably to establish user trust by keeping them distinctly separate from the chat.

Often, the ad loads before the chatbot’s query response, said Baird, and Koah’s goal is to “deliver such a relevant result to the user that they just click on the ad before the result loads.”

But these ads aren’t necessarily going to drive immediate purchases; chatbots aren’t just “the next Google,” said Baird. They’re usually a mid-funnel, rather than bottom-funnel, opportunity for advertisers, he added.

Getting (not too) personal

If someone wants to buy groceries online, they’ll likely go directly to Instacart rather than doing so through ChatGPT, Koah Co-Founder Mike Choi said. Instead of just advertising contextually based on a specific query, he said, Koah targets users with ads based on personal data, like previous chat conversations within the platform and the user’s location.

For instance, said Choi, if a student is using Luzia for help with math homework, it might seem obvious to serve them an ad for Quizlet, or a math tutoring program. But a student is going to be solving math problems “all year long,” said Choi. After encountering the same ad dozens of times, he said, the user will say, “I’ve seen enough of that.”

Instead, Koah might serve a DoorDash ad with a “fun” spin, Choi said, like a message saying, “You’ve been studying math for way too long. You should order a burrito.”

Advertisers tell Koah what sorts of topics and keywords they want to appear next to, like “back to school” or “spring fashion,” and deploy the right ads accordingly.

Of course, Koah and Luzia are constantly working to ensure that the ads they deliver are relevant but don’t “feel creepy,” said Choi. After all, he added, no one wants their AI to look like the Claude Super Bowl ad.

Must Read

Walmart’s Ad Revenue Totaled $6.4 Billion In 2025 As The Ecommerce Flywheel Started To Spin

“Fully a third of our profit in the most recent quarter was related to advertising and membership income,” Walmart CFO John David Rainey told investors on Thursday.

Comic: AI-TA?

Q4: Omnicom’s IPG Merger Is An AI Test Case

Omnicom just reported its first earnings since closing the IPG deal and, shocker, it’s saying AI is main growth driver for combined holdco.

Digital-native brands need to figure out how to win in retail shelves. They're finding it difficult, to say the least.

Big CPG Brands Are Quick To Cut Ad Spend Amid A Tough US Market

Companies like P&G, PepsiCo and Colgate-Palmolive are cutting marketing spend as the easiest and quickest way to protect profitability.

Privacy! Commerce! Connected TV! Read all about it. Subscribe to AdExchanger Newsletters

How The Minnesota Star Tribune Protects Advertisers While Covering ICE Crackdowns

Amid a federal crackdown and local unrest, Minnesota’s biggest newsroom is proving brand safety and hard news can coexist.

Hasbro And Animaj Form A New YouTube Ad Sales House For Kids And Family Content

The kids companies Hasbro and Animaj have formed a co-venture for selling their ads on YouTube and streaming media.

I Asked ChatGPT Where My Ads Were – But It Was Wrong, OpenAI Said

It’s official: ChatGPT has launched ads and the test will expand in the coming weeks. But don’t ask the LLM for details, unless you’re looking for misinformation.