From Rules to Queries: How AI Is Rewiring Media Intelligence

From Rules to Queries: How AI Is Rewiring Media Intelligence

By: Manuel Wecker (Global Account Manager, UNICEPTA by PRophet) | The Current State of AI in Media Monitoring and Analysis

How exactly is AI changing the work of those who monitor and analyze media? At the May session of the “Data & Insights Experts” interest group, Manuel Wecker (Global Account Manager, UNICEPTA by PRophet) presented a workshop report from a service provider’s perspective, offering a look under the hood of production systems rather than at marketing slides.

The desire for knowledge remains; the path changes

The central question of media intelligence is an old and enduring one. To paraphrase Harold D. Lasswell, it is: “Who says what about whom, how?” What has changed radically is the path to the answers. Wecker’s timeline brings this into focus: For over 20 years, “Data Rules” reigned—the era of algorithms. This was followed by about five years of “Data Answers” with rule-based machine learning. For about two and a half years now, “Data Questions” has been the order of the day—the era of large language models that understand text on a whole new level.

Three Levels at Which AI Operates

Wecker categorized AI applications into three areas. First, AI “under the hood,” in the backend production systems. Second, AI in core services such as media monitoring and media analysis—where customers experience it directly. Third, AI in social listening tools, which takes big data analysis to a new level. The Lasswell formula serves as a technical roadmap here: “Who” and “Whom” are derived from named entity recognition and linking; “What” from key message and topic detection; and “How” from sentiment analysis.

Reading comprehension is the bottleneck

As powerful as these tools may seem, their core remains machine text understanding. This is precisely where specialization pays off: The NER solutions from specialized media intelligence providers achieve an F1 score of 0.93, outperforming standard models such as BERT (0.77), RoBERTa (0.74), and spaCy (0.66)—especially in sentiment analysis, entity recognition, and disambiguation. Precision doesn’t come from the largest model, but from the one best tailored to the domain.

LLM as an autopilot, with a human hand at the wheel

The switch to LLM-based coding offers tangible benefits: high flexibility through targeted prompt management, minimized error variance, a broader range of analysis, and customization without the need for in-house pre-training. However, it is crucial that humans remain involved in the process. In the LLM-supported media review, the AI classifies and ranks articles, correctly grouping about 65% of them with 95% accuracy, and only about 0.34% of its suggestions are rejected by subject matter experts. Their feedback is continuously fed back into the model. The workflow—from setup and data input through enrichment and validation to analysis and visualization—combines customized client prompts with relevance and completeness checks and human approval.

What Remains

The panel identified three trends. Monitoring and analysis are converging because immense amounts of data can be analyzed with high accuracy in near real time using LLMs. Project-specific prompts without training data make it possible to flexibly tailor analyses to the specific communication situation. And natural language control is becoming the standard in social listening tools; lengthy Boolean search commands are a thing of the past. The fact that Microsoft introduced new Copilot connectors in mid-May—including UNICEPTA byPRophet—shows just how quickly AI agents are becoming part of everyday work.

Conclusion

Wecker’s assessment is both sobering and encouraging. AI has long been productive at Media Intelligence, but not as a “magic button”—rather, as a precise tool used within clearly defined processes. The competitive advantage lies less in the model itself and more in the combination of domain knowledge, effective prompt management, and human quality assurance.
Want to know where AI can make a real difference in your monitoring and analysis? In the “Data & Insights Experts” interest group, we share workshop reports and insights like these. Join us!