Prompting Myths vs. Measured Methods: What the research says Actually works

This session qualifies for 1 CPD/CPLD/CPE Hour credit.

Most prompting advice is built on habit, not evidence. Telling an AI to “act as an expert,” asking it to “think step by step,” and correcting it over a long back-and-forth conversation are among the most widely recommended techniques. Recent controlled research from institutions including the Wharton School, Microsoft Research, and Meta AI shows that these habits often deliver little benefit, and in some cases make results worse. One recent study found that leading AI models performed an average of 39 percent worse when instructions were delivered across multiple conversation turns rather than in a single, complete prompt.
In this session, Sandra Rogoza separates prompting myths from methods that measurably improve accuracy. Through live demonstrations using popular AI tools such as Claude, ChatGPT, Perplexity and others, she will put popular techniques to the test in real time and show what the evidence supports instead. The session draws on Sandra’s context engineering approach and connects it to the latest published research on how AI models process, prioritize, and sometimes misuse the information they are given.
This session is designed for professionals ranging from upper novice to expert. Attendees new to structured prompting will leave with practical techniques they can apply immediately, while experienced practitioners will gain a research-grounded view of where prompting is heading and resources to replicate the findings themselves.

What Attendees Will Learn
• Why expert personas do not reliably improve AI accuracy, and what to use them for instead
• Why AI models lose accuracy in long conversations, and how to structure prompts to prevent it
• Why giving AI more information can reduce accuracy, and how to curate context for better results
• How to use AI to verify its own outputs and reduce fabricated information
• How to recognize and counter AI’s tendency to abandon correct answers when challenged
Who Should Attend
• Leaders responsible for AI adoption, governance, or quality control within their organizations
• Data scientists, analysts, engineers, and energy sector professionals who use AI tools in their work
• Anyone who wants AI outputs they can trust, not just outputs that sound convincing
Thanks to Platform Innovation Centre, the Official Venue Sponsor.