10th November 2026
On 4th November, AMEC brought the measurement and insight community together in central London for its AI Day, with eMedia Monitor (eMM) proud to be Platinum Sponsor. The aim was to look past the hype at what is actually working, what the industry is learning, and where AI is adding real value across communications, research and measurement. Through presentations, panels and case studies, speakers from agencies, in-house teams, data providers and academia tackled one shared question: how do we turn AI’s potential into meaningful, measurable impact?
Across 18 sessions, a common message emerged: AI is rapidly expanding what communications teams can measure and analyze, but its real value depends on the quality of the underlying data, the context surrounding it and the human judgment applied to the results. From AI visibility and audiovisual analysis to governance and industry standards, five key themes stood out.
Tracking where organizations appear in AI-generated answers was one of the day’s most discussed topics, yet speakers repeatedly warned against treating visibility as success.
Golin’s Jonny Bentwood set the scene: every audience now has a machine in front of it. People ask AI longer, more layered questions and rarely challenge a fluent answer. Most prompt tracking, he argued, remains guesswork, while raw citation counts can be deeply misleading. His summary: “We are no longer competing for headlines. We are competing to be chosen.”
The stakes are rising. Andrew Bruce Smith of Escherman cited OpenAI’s claim that ChatGPT’s search-like use now equals around 20% of Google’s volume, while UNICEPTA’s Martin Schulze highlighted zero-click rates approaching 70%. Yet being cited is not the same as being represented accurately. Smith pointed to research from the ONS showing that while 97% of Google AI Overviews cited its work, only 21% represented its figures fully and accurately. Schulze illustrated a similar disconnect: a bank ranked first for overall AI visibility but was absent from answers relevant to its most important client persona. Visible, but not chosen.
Stuart Bruce of Purposeful Relations showed why understanding where AI draws its information matters. For a maritime trade body, owned and member websites influenced answers more than earned media, while a leading trade publication was effectively invisible behind its paywall. In another example, a company that shared its response to an industrial dispute only with journalists, without publishing it on its own website, found that AI failed to pick up its statement and instead presented only the union’s side of the story.
The message was clear: AI visibility is a starting point, not an outcome. Meaningful measurement requires understanding whether organizations are being represented accurately, on the topics that matter, to the audiences they need to reach.
AI can count mentions, logos and keywords at enormous scale, but a count only means something once placed in context. Cut a mention out of the story around it, or separate a picture from its words, and the same data can tell you the opposite of what happened. Choosing the right unit of analysis, and keeping context attached to it, came up again and again.
For eMM, this was particularly relevant. Maya Koleva, eMM’s Head of Product, explained that broadcast monitoring has traditionally relied on short clips cut around individual mentions. While efficient, this approach can sacrifice valuable context. When eMM compared clips with full story segments for Uniqlo’s broadcast coverage, the clips made wider industry stories appear brand-focused, shifted sentiment assessments and left nearly 20% of speakers unidentifiable. Analyzing the complete story also makes broadcast more directly comparable with print and online articles. As Maya put it: “Same data, different unit, different answers.”
While Maya demonstrated why preserving the full broadcast story matters, Brian Merron of HEKA Insights showed that even highly accurate visual detection still requires interpretation. Using eMM broadcast data for a Formula One project, his team more than doubled logo detection precision from 0.379 to 0.815, reducing 15–20 hours of weekly manual monitoring to near-real-time analysis. Yet detection is only the starting point. When Cristiano Ronaldo moved two Coca-Cola bottles aside at a press conference and held up a bottle of water instead, logo detection alone would have recorded valuable visibility for Coke, despite the wider story suggesting otherwise.
Accurate detection is not the same as meaningful analysis. Whether identifying brand mentions or visual exposure, preserving the surrounding context is essential to understanding what the coverage actually means.
AI does not fix weak data; it scales it. Where poor inputs once produced one flawed report, they can now generate thousands of confident, convincing answers. This presents two challenges for communications teams: ensuring the data underpinning their own analysis is reliable, and making sure the information they publish is accessible to AI systems.
Joyce Higgins of FleishmanHillard won’t open a new AI environment until the team can answer one question: “Who is your data partner?” You wouldn’t bring someone to a meeting who adds no unique perspective, she argued, so why bring data that adds none? Maya Koleva’s AI-ready checklist starts with choosing the right unit of analysis, then asks whether the data is complete, consistent, traceable and permitted for use, warning that any mistake in a transcript or clip carries straight through into everything AI does with it. Amar Vyas of M&C Saatchi Fluency reinforced the importance of traceability: if you cannot connect a recommendation back to its supporting evidence, don’t use it.
The second challenge is ensuring that organizations’ own information can be found and understood by AI. UNICEPTA’s Martin Schulze described how a bank’s most relevant content was buried in PDFs that AI struggled to access, leaving competitors and regulators to shape the answers instead. James Crawford similarly urged organizations to ensure that information about their brands, products and people is consistent across the internet before focusing on prompt optimization.
The underlying message was straightforward: AI needs reliable information, whether it’s analyzing media coverage or generating answers about an organization. Without accurate, accessible and traceable data, greater speed and scale simply amplify the risk of getting things wrong.
While the need for human oversight was widely recognized, speakers differed on how much checking is necessary, by whom, and what happens when the checker becomes complacent. The emerging consensus was one of proportion: match the level of human review to the potential consequences of an error, while recognizing verification as part of the true cost of AI.
Courtney Glymph of YourStory PR advocated keeping humans involved in reviewing around nine out of ten outputs, while Tabita Andersson, VP Marketing and Communication at Rimini Street, warned that the real risk is complacency once an AI agent starts performing well. Ines Schumacher of FleishmanHillard TRUE Global Intelligence offered a simple reminder: “Don’t ever get comfortable.”
Vuelio’s Amy Chappell and Kyle Lindsay brought the debate to life, with Amy advocating for the continued importance of human analysts and Kyle championing the potential of AI. Kyle noted that humans are not a perfect gold standard either: human errors may be more frequent but relatively minor, while AI can make fewer mistakes with potentially serious consequences if left unchecked. Both ultimately converged on the importance of harm-weighted accuracy. As Amy put it: “It’s not just more checking, but it’s smarter checking.” Amar Vyas of M&C Saatchi Fluency added a practical consideration: reducing a 10-hour task to two hours achieves little if human verification takes another eight.
The lesson wasn’t that humans must check everything, but that oversight needs to be proportionate to the risks. AI can accelerate analysis, but human judgment remains essential in deciding whether the results are reliable, what they mean and whether they can safely inform decisions.
As AI moves from tool to colleague, the question shifts from what it can do to who takes responsibility for its decisions and outputs. The message was clear: accountability remains with people, from those using AI in their daily work to the organizations deploying it and the industry setting the standards.
For individuals, Amar Vyas of M&C Saatchi Fluency put it bluntly: “You can’t blame Claude. You can’t blame OpenAI.” Ines Schumacher of FleishmanHillard TRUE Global Intelligence reinforced that responsibility extends from the person running the query to the leader acting on its findings.
At an organizational level, Tabita Andersson of Rimini Street saw an opportunity for communications teams to help shape AI ethics and policy across the business, while acknowledging that the industry is still finding its way: “I don’t think we’ve got it all figured out yet.”
Beyond individual organizations, the challenge is to establish shared principles for measuring AI’s impact. Golin’s Jonny Bentwood warned that “if we don’t set the standards for measuring in AI, someone else will.” James Crawford outlined how AMEC’s GEO Principles, which encourage organizations to evaluate AI visibility against business objectives rather than appearances alone, will be updated ahead of next May’s Summit.
Accountability cannot be delegated to a machine. As AI becomes embedded in communications and measurement, organizations need clear responsibilities, appropriate safeguards and shared standards to ensure its outputs can be trusted.
In the day’s closing address, Andrew Bruce Smith of Escherman offered six predictions for the year ahead, several of which reinforced the themes running throughout the conference.
He warned that AI citation counts risk becoming the new advertising value equivalent (AVE): another convenient but misleading measure of communications success. Instead, he anticipated a shift toward stakeholder-focused analysis that moves beyond simplistic sentiment scores to examine how different audiences understand an organization.
For media intelligence providers, he predicted that the quality of underlying data and the ability to connect it with other systems would increasingly matter more than the dashboards used to display it. He also expected calibration statements, explaining how AI-generated findings have been tested and validated, to become a standard feature of evaluation reports and RFPs.
Perhaps most importantly, he predicted that the skills gap, rather than the technology gap, would become the industry’s biggest challenge. It was a fitting reminder of why events such as AMEC AI Day matter: bringing practitioners together to share experiences, challenge assumptions and learn from one another as the industry adapts.
Contact: eMedia Monitor GmbH · media@eMediaMonitor.net · +43 1 581 2630 100
eMediaMonitor