When Marketing AI Institute founder and CEO Paul Roetzer posted his take on New York City's decision to ban generative AI in public school classrooms, the response was immediate and voluminous. Hundreds of comments flooded in, ranging from thoughtful to pointed to aggressive. Sorting through all of them manually would have taken hours.
So rather than copy and pasting comments into a document or skipping the analysis altogether, we handed the job to AI.
We used GPT-5.6 Sol's browser use capability, accessed through the ChatGPT/Codex desktop app. This capability enables AI to essentially take over the browser and walk through the steps of computer use as a human would, navigating websites, clicking on links and more.
This is step-by-step how the AI did it:
- Opened the LinkedIn post in a signed-in browser session. No manual navigation required
- Switched from the default “most relevant” view to the view that showed the full comment thread
- Started at the first comment and worked systematically through the page in order
- Loaded additional comments as they appeared, scrolling and triggering LinkedIn's "load more" behavior automatically
- Expanded replies within comment threads without being prompted to do so
- Opened comments hidden behind LinkedIn's "See More" button — meaning longer comments that are truncated in the default view were read in full
- Separated Paul’s own replies from the audience's comments to avoid skewing the analysis
- Distinguished individual commenters from the number of comments they posted to avoid duplicating one person’s opinion.
- Grouped responses into "for" and "against" positions on the NYC school AI ban
- Surfaced recurring arguments that appeared across multiple comments
- Pulled specific examples that illustrated the range of perspectives
Throughout the entire process, a human watched as the AI worked. It’s not advised to walk away while an AI agent is operating inside a live, signed-in account.
What the Results Showed
The final analysis landed at roughly 50/50 or nearly identical positive and negative sentiment toward Roetzer's post and the broader position it represented on AI in schools.
That split was not the most interesting finding. The more revealing insight was that the loudest, most aggressive comments did not represent the majority. When every commenter was given equal weight — not every comment, but every person — a far more nuanced picture emerged. Many people who would have been drowned out by the louder voices on either side held more measured, thoughtful views on the topic.
Why This Matters for Marketers
Social listening is not a new concept. What is new is the ability to do it quickly, efficiently and accurately with AI.
This makes the barrier between "I want to know what my audience is saying" and "here is what your audience is saying" significantly lower.
Consider what this same approach could do across other marketing workflows:
- Analyze audience sentiment on a product launch post without waiting for a formal survey
- Track recurring objections in comments on a competitor's content to inform messaging strategy
- Monitor community responses to a campaign before deciding whether to amplify or adjust
- Understand what arguments keep surfacing around an industry debate your brand has a stake in
- Separate the signal from the noise on any high-volume comment thread where the loudest voices don't necessarily represent the majority
The underlying principle is simple: anywhere a marketer would otherwise copy-paste content from a browser to analyze it, AI browser use can take over that process.
An Important Caveat
This kind of capability comes with responsibility attached.
Browser use — AI that operates inside a live, signed-in session — requires care. It should be run in a controlled environment, actively monitored throughout, and scoped narrowly to the specific task at hand. The appropriate posture right now is supervised and deliberate, not automated and unattended.
We ran our experiment as a very limited test. It’s important to note that certain sites, including LinkedIn, have terms of service that could ban your account if you apply this automation widely or at scale.
We share this example, in its limited use, to help people understand the significant capabilities of AI browser use and its ability to change what you can do in marketing.
Raising the Ceiling of What's Possible
What's possible with AI is rising fast. GPT-5.6 Sol handled this task with a level of accuracy and autonomy that would have been surprising even a few months ago. GPT-6 Astra, OpenAI's newest release, which was not yet available at the time of this test, is specifically positioned as a major leap forward in computer use speed, accuracy, and reliability.
Marketers who start experimenting carefully with this now in controlled conditions will be ahead of the curve when these capabilities become the standard.
This post draws on the AI Use Case Spotlight segment of Episode 237 of The Artificial Intelligence Show, hosted by Paul Roetzer and Mike Kaput. To listen to the full episode, visit: https://podcast.smarterx.ai/shownotes/237
For more on building AI-ready marketing teams, explore AI Academy at academy.smarterx.ai.
Mike Kaput
Mike Kaput is the Chief Content Officer at SmarterX and a leading voice on the application of AI in business. He is the co-author of Marketing Artificial Intelligence and co-host of The Artificial Intelligence Show podcast.