AI is an active force reshaping how teams plan, test and adapt in real time. At the 2025 AMEC Global Summit, Penelope Penelope Mantzaris, SVP at Edelman DXI took the stage with leaders from Shell and Ketchum to explore one of the most impactful and practical AI applications yet: synthetic audiences.
Moderated by Kyle Mason (Shell), the panel moved beyond the hype to highlight real-world use cases, ethical considerations and the future of synthetic models in comms. From stress-testing messaging in a crisis to simulating trust dynamics, the conversation made one thing clear: this isn’t experimental anymore. It’s operational.
In this recap by Penelope Mantzaris, SVP at Edelman DXI, you’ll find key insights, quotes and practical examples that show how synthetic data is becoming essential to modern communications strategy.
What you need to know: Key insights from the panel
- Synthetic audiences simulate real people using AI-trained models, built from real-world data, to help communicators plan, test, and refine strategies faster and more precisely.
- They're not a research replacement, but a complementary tool, especially valuable when time, cost, or privacy limit traditional research.
- Agencies are already using them for a variety of use cases: to simulate trust dynamics in real time; to test messaging for highly specific patient groups and more.
- They shine in crisis and fast-paced environments, enabling teams to scenario plan, test tone and voice, and avoid reputational missteps.
- Ethical concerns and data validation are real: bias, hallucinations, and data quality underpin the need for continuous human oversight and updating.
This isn’t sci-fi. It’s already happening.
Moderator Kyle Mason (Shell) opened with a clear take: AI is everywhere in comms right now, but too often the conversations are hypothetical. What excited him was the chance to hear how Ketchum and Edelman DXI are embedding synthetic audiences into their work, from campaign planning to trust measurement.
What is a synthetic audience?
“What a synthetic audience refers to is a simulated or artificial group of people that is created using AI and algorithms. They’re custom GPTs built on real-world data - from CRMs, syndicated research, first-party sources - to simulate how specific groups think, feel, and behave.”
— Mary Elizabeth Germaine, Ketchum
Synthetic audiences act as AI proxies for real human segments. Built using layered data, they allow teams to:
- Pressure-test campaigns quickly
- Craft personalized messages for micro-audiences
- Experiment with sensitive concepts safely (e.g. under NDAs)
Why They Matter Now
Mary Elizabeth and Penelope Mantzaris (SVP, Edelman DXI) both pointed to the pressure of time in PR and comms. As Mary Elizabeth put it:
“Ad agencies get 6 months to plan. PR gets five days.”
Synthetic audiences fill that gap. In Penelope’s words, they’re a “front-loaded accelerator" especially for:
- Testing hard-to-reach or high-risk audiences (like regulators, patients, activist citizens)
- Iterating messaging instantly, without redoing the research each time
- Navigating polarization by simulating how different media ecosystems or opinion segments might respond
Crisis Mode: AI at the Speed of Outrage
When comms teams have hours - or less - to respond to a brewing crisis, synthetic audiences offer a crucial edge.
“The pre-planning is a key. In a crisis, you have the iteration, you have the speed - if you've got your audience set up. It might be your regulator audience, your journalists, etc, and you embed them in your scenario planning. Do we want to talk with an authoritative voice? Do we need to strip out the jargon? What's going to land with each of those audiences in the event of a crisis?”
— Penelope Mantzaris, Edelman DXI
Mary Elizabeth added that you can also layer live social or media sentiment data into the models for added realism, helping tailor the response in near-real time.
However, both were clear: synthetic tools are not appropriate for all crises. In highly sensitive situations involving loss of life or trauma, real human insight remains paramount.
Measuring the Immeasurable: Trust, Outcomes & Impact
Penelope offered a glimpse into Edelman’s next-gen toolkit:
“We’ve built a synthetic audience model called Trust Stream, powered by 20+ years of Edelman Trust Barometer data. It simulates how the general public, employee, or investor audiences would respond to content - and scores the trust impact in real time.”
This kind of layered, ongoing measurement makes synthetic models more than just a planning tool, they're becoming dynamic barometers of audience sentiment.
Mary Elizabeth added a note of caution:
“You can’t guarantee outcomes with synthetic models, but you can get directionally smarter and sharper about what will likely resonate.”
Challenges & Guardrails
With power comes responsibility. The panel was unanimous: data quality, model validation, and regular updates are essential.
This isn’t a matter of asking ChatGPT to “think like a Gen Z woman.” As Penelope put it:
“This is building in layers of data, demographic, psychographic, media buying, preferences, lifestyle, hobbies, all of that. You have a whole person, much as you would in a piece of poll research, where you have a persona that can give you not just plausible insights, but real nuanced, usable insights.”
Avoiding bias and hallucinations requires both art and science: human checks, smart inputs, and constant evolution.
What’s next?
Both panellists agreed: synthetic audiences will soon be standard practice. Not just a niche tool for cutting-edge agencies, but a core capability for comms teams globally.
"As we become a little bit more sophisticated at that impact point, we are moving beyond just your vanity metrics, and really embedding things like trust, reputation, and so on to your evaluation as well.”
— Penelope Mantzaris, Edelman DXI
“It's going to be the core of what we do as communications professionals. Really think about how you're using them, how you can bring it in, because when we talk AI, it is disruption.”
— Mary Elizabeth Germaine, Ketchum
“From a brand perspective, everything that we do is highly audience centric. If we have any measures that aren't audience centric, it's tough for us to assess the overall impact that relates to the organizational objectives. So, I think this is critical.”
— Kyle Mason, Shell