Five Key Questions to Ask Before Trusting AI-Generated Insights

How to Evaluate AI-Generated Insights: Five Questions to Consider

By Amber Daugherty, Director Value & Impact, Big Valley Marketing

We’re getting very good at using AI to produce answers. But are we getting any better at knowing which answers we can trust? That’s a question I kept coming back to during a panel with Joyce Higgins, Ines Schumacher and Oresti Patricios at AMEC’s AI Day, From AI Adoption to Decision-Ready Intelligence: Building Systems You Can Trust.

We discussed everything from synthetic audiences and data quality to training, accountability and AI’s limitations across different markets. And while the technology is moving at an extraordinary pace, much of what makes AI-generated intelligence trustworthy comes down to familiar principles: understanding your data, asking better questions, applying critical thinking and recognizing when you don’t have enough evidence to make a decision. Here are five questions every communications and measurement professional should be asking.

  1. Where is your data coming from?

We’ve all heard garbage in, garbage out. It’s simplistic, but it’s a useful starting point for AI. Joyce Higgins, Head of AI Strategy at FleishmanHillard UK, made an important point: research and data specialists need to be involved from the beginning of AI projects, not brought in after workflows have been built. Connecting an AI system to a data source isn’t the same thing as understanding that data. An API might provide access to information, but do you know what’s available? Are you retrieving it correctly? How was the audience defined? Is the information appropriate for the question you’re trying to answer?

Joyce described involving research specialists in selecting data sources, constructing audience profiles and determining how information should be incorporated into AI workflows. We shouldn’t start by asking what we can automate. We should ask what we’re trying to understand, what evidence we need and whether the available information is fit for purpose. Before building an AI workflow, ask: do we understand our data well enough to trust what the system will do with it?

  1. Who or what is missing from the information?

Oresti Patricios, CEO of Ornico, described the challenges of using AI for media intelligence across African markets, where languages, accents and cultural contexts aren’t always adequately represented. In South Africa alone, organizations must navigate multiple official languages and significant cultural diversity. AI systems don’t necessarily perform equally well across those contexts. Oresti explained that his organization initially hoped to automate more of its media intelligence workflows but quickly discovered that human intervention was essential to address errors and ensure accuracy.

That raises a bigger question: whose perspectives are represented in the data, and whose might be missing? If you’re monitoring media coverage or trying to understand audience perceptions, incomplete representation isn’t just a technical limitation. It can fundamentally change your conclusions. For global organizations especially, it’s worth questioning whether an AI system’s apparent confidence is masking gaps in language, geography or cultural context. We can’t make informed decisions about audiences we don’t adequately understand.

  1. Are you generating information or intelligence that can actually inform a decision?

One of the most compelling examples came from Ines Schumacher, Senior Partner, Platforms Lead North America at FleishmanHillard TRUE Global Intelligence. Her team used a synthetic focus group ahead of a product launch and identified a potentially important customer segment that hadn’t previously been considered. But here’s the interesting part. They didn’t simply accept the AI-generated finding. They investigated the underlying information and validated the insight through primary research with real people before incorporating it into the launch strategy.

Joyce shared another example in which synthetic audience research helped a client anticipate eight of the ten questions subsequently asked by financial journalists following an announcement. These applications are exciting because they can help organizations identify overlooked perspectives, anticipate reactions and make better decisions. But an interesting AI-generated answer isn’t necessarily defensible intelligence.

Ines described five principles of decision-ready intelligence: high-integrity data, business context, appropriate guardrails, curiosity and critical thinking, and accountability. The output isn’t the end of the process. It’s something to interrogate. What’s the evidence? What assumptions are being made? What needs verification? And, importantly, what decision are we trying to support? An AI-generated insight isn’t valuable simply because it’s interesting. Its value comes from helping us make a better decision.

  1. Are you training people to use AI, or just giving them access to it?

There’s a big difference. Joyce described an ongoing training program covering everything from prompting and model limitations to APIs and advanced AI workflows. One challenge is that AI adoption doesn’t follow a linear path. Someone might be experimenting with sophisticated agentic workflows while lacking a basic understanding of how to provide appropriate context or verify information. And because the technology changes so quickly, training can’t be a one-time exercise. Oresti described similar challenges with employees and clients. His organization has invested significantly in AI education, but providing training doesn’t automatically change behaviours.

If we want communications professionals to use AI responsibly, we need to do more than teach them which buttons to press. They need to understand where the technology can fail, how to recognize questionable outputs and when to investigate further. A polished summary, convincing chart or confident recommendation can look authoritative even when the evidence is weak. Which brings me to the next question.

  1. Who is accountable when AI gets it wrong?

During our conversation, Ines described reviewing a presentation containing a completely fabricated graph. The underlying data didn’t exist. Someone had apparently incorporated AI-generated material without adequately checking it. But what interested me even more was the context. The person responsible was a junior team member working under considerable time pressure. Can we reasonably expect people to verify AI-generated work if we don’t give them the time, support or confidence to do so? Joyce was clear that, within her organization, humans remain responsible for AI-generated outputs. I agree. But accountability needs to be supported by how organizations actually operate. That means clear expectations for verification, processes for tracing conclusions to their sources and an environment where employees feel comfortable challenging outputs or asking for more time. As Ines pointed out, just because a system produced an accurate result once doesn’t mean the next result will be accurate. And when the evidence isn’t sufficient to support a conclusion, we need to be willing to say so. Sometimes the most responsible recommendation is that we don’t yet have enough evidence to act.

The goal isn’t more AI. It’s better decisions.

AI has enormous potential to improve how communications professionals gather information, understand audiences and advise organizations. But adoption alone isn’t an outcome. Neither is producing more reports, generating insights faster or automating workflows that previously required hours of manual effort. Those things may improve efficiency. They don’t necessarily improve the quality of our decisions. Are we solving the right problem? Do we understand the evidence? Have we considered who might be missing? Can we defend our conclusions? Ultimately, the value of AI-powered intelligence isn’t in how quickly it gives us an answer. It’s in whether that answer helps us make a better decision, and whether we have good reason to trust it.

Source Link

Share your love