Ask an AI assistant to recommend a hotel, a product in a category or something to do this weekend. It assembles results from what has been published, linked and discussed about those options online. At most organizations, no one decided what it would say.
Reputation has always been assembled this way. An organization's own channels state the story, earned media reinforces it and communities test it. What is new is that AI systems now synthesize all three into one answer most people never trace back to a source. At Meow Wolf, where I served as VP of Communications for three years, the team treated that as an extension of existing practice rather than a new specialty.
Five operational AI adjustments did most of the work:
- Audit AI representation the way you audit coverage. Ask the major AI systems the questions the brand's audiences actually ask, using their language and bring the unedited answers to leadership. The goal is understanding how audiences will perceive your organization. When the European Broadcasting Union and the BBC evaluated more than 3,000 AI-generated news answers, 45% contained at least one significant issue. Providing executives with the gap between the story intended and the machine's version changes conversations quickly.
- Value coverage for where it lives, not just who reads it. Earned media does not stop working after publication. It becomes part of the information environment AI systems draw from. Muck Rack's analysis of cited links found earned media accounts for 84% of AI citations, with journalism making up 27%. Track which publications consistently appear in answers about your category and how those stories connect to the broader conversation.
- Participate where reputation is already taking shape. The rest of that earned majority is community discussion, which is why Reddit ranks among the most cited domains in Semrush's analysis of AI answers. Joining those conversations as individuals, not anonymous brand accounts, shows which narratives are forming and builds credibility that cannot be manufactured.
- Monitor citations, not just mentions. Visibility and influence are no longer the same thing. Semrush found that many ChatGPT citations came from pages ranking outside Google's top 20 results. Instead of measuring only the volume of coverage, identify which outlets and stories repeatedly appear in AI-generated answers and adjust media strategy toward the reporting that is actually shaping how organizations are described.
- Keep editorial judgment human. The same systems that reward human conversation are quick to strip your voice from a draft. The AI tools introduced recognizable habits—the staccato triplets, the reflexive "it's not x, it's y," the cadence that reads fine once and says nothing. Every draft needs a human edit for voice before it goes out, both for search ability and quality. I have gone back to my own early ChatGPT-assisted edits and cringed at how obvious it was that a robot had swapped out my voice.
The measurement trap is worth naming. Dashboards built for this count citations, mentions and share of voice in AI answers. Those are useful signals that say nothing about whether the description is right. In March, Axios reported that Google search referrals to the news sites that communicators pitch fell 34% YOY by Chartbeat's count, while chatbot referrals remained under 1% of their traffic.
Machines now read and summarize coverage more often than consumers click through. The check that matters takes ten minutes: read the answer and decide whether you would have written it.
Communications teams already have the skills to manage how the machines answer the questions. Use them now, or spend years explaining an answer you had no hand in shaping.
Kati Murphy is Principal of KM Communications. She previously served as Vice President of Communications at Meow Wolf and Executive Director of Public Affairs at the Art Institute of Chicago.