Ozone is developing a platform that lets publishers simulate how their content would surface in AI answer engines like ChatGPT, showing how it’s processed and cited so they can optimize structure, improve visibility — and help them set guardrails for future agreements.
The experimentation platform is part of the launch of R&D Labs by Ozone, a digital advertising publisher alliance group. R&D Labs is a tech sandbox for all publishers that want more data insights and understanding into how their content is being surfaced and used within LLMs, to help inform editorial, product and monetization strategy. Ozone plans to publish findings from the initiative for other publishers to learn from in the next month.
It’s one of the first big initiatives by Ozone since it began signing on a wave of U.S. publishers – including the Wall Street Journal, New York Post, the BBC (U.S.) and CNN – and pitching the buy sider harder last year. The simulation platform has been running for three months, and eight publishers have opted into the research, according to Bryan Scott, CMO at Ozone.
While AI-generated answers can appear authoritative and well-sourced, closer inspection often reveals inaccuracies, from misquotes to misattributed or opinion-based claims presented as fact. The company says its simulation platform helps explain how those outputs are generated.
Because this is an R&D program, Ozone isn’t charging publishers for using the simulation tool. “We have a small group of opted-in, active publisher participants for the program, surrounded by broader engagement with our wider community,” Scott said, without naming names.
The platform will simulate AI answer engine responses, and show how content was structured, organized, retrieved and cited in the outputs. The idea is that publishers can take these insights and tweak article content structure to see how it impacts those AI-generated summaries, according to Ozone CEO Damon Reeve.
“There is complete information asymmetry between what the publisher knows, how their content is being consumed, and what the LLMs know,” Reeve said. “And so the publishers are held very much at arm’s length by the LLMs.”
Reeve added that he believes that’s the case even if a publisher has a licensing deal. “In terms of who’s asking what questions: ‘how is my content being used? Is it working or not?’ — anything like that, the publishers we speak to are completely in the dark,” he said.
Which is why Ozone is simulating AI responses, and not analyzing the real thing. A group of publishers working with the group made their data available for this research program, which is then used to make predictions and assess LLM performance in the experimentation platform, Scott noted.
“We don’t really know what happens under the hood of LLMs,” Reeve said. But insights from the experimentation platform could help publishers get a better understanding of how AI answer engines generate their outputs, and where and how publisher content appears, he added.
Two publishing execs — who spoke under the condition of anonymity — told Digiday that they were still weighing the value of improving their AI visibility and optimizing content for AI answer engines when they are still seeing very few click-throughs and referral traffic from those platforms. So while a tool like this may help their audience development and SEO teams gain a better understanding of how LLMs source and cite content, they weren’t yet putting many resources into this area because the juice isn’t worth the squeeze.
One publishing exec said the click-through rate from AI search tools was around 0.3 percent. “[This] makes it tough to expedite investment on. We’re still in a watch-and-learn phase,” they said.
Ozone Labs spans research into how publisher content is used in LLMs, collaborative hackathons to build and test new prototypes with industry partners, and a public portfolio of early-stage projects and open-source output, including experiments with emerging ad-tech standards such as AdCP and arTF.
If publishers or marketer partners want help building a buy or sell-side agent, that could be developed with Ozone, and then open-sourced for others to use, noted Reeve. Ozone will work with publishers and marketers on a case-by-case basis and will be run by existing team members, with experimentation led by CTO Scott Switzer.
Comments (0)