What happens if you query the same prompt repeatedly over time and see how the AI changes the response?

A recent experiment subjecting identical queries to two leading AI platforms, Microsoft Copilot and Google Gemini, over a three-week period underscored this inherent variability. The task was designed to elicit a succinct, engaging, and informative definition of generative AI's beneficial impacts on business.

⌨️

Prompt used in the experiment:
"Write a succinct statement, 100 words or less, for business users offering a definition of generative AI that focuses on discussing the beneficial impacts of AI on business. Make the content engaging, compelling, and informative way to gain understanding. "

The Findings

Both platforms consistently met the word count, offering plausible and relevant responses. However, beneath this initial similarity each system approached the task differently and with a difference in variability in the responses. However, their approaches varied:

Directing AI: The Key to Unlocking Outcomes

The unpredictability of generative AI is evident in the wide range of responses it can produce for a single prompt. Unlike search engines that deliver static results, AI models craft answers anew with each query, drawing from a vast and constantly evolving knowledge base.

AI models are in a perpetual state of learning, refining their understanding with each interaction. Consequently, identical prompts can elicit different responses over time. This dynamic nature necessitates a flexible approach from users.

To fully harness generative AI, users must adopt a directorial mindset. View the AI as a versatile tool capable of producing countless variations. Just as a director guides a film from script to screen through multiple takes and edits, users must iterate and refine their prompts to achieve desired outcomes. Generative AI excels at responding to direction, embracing nuance and adaptation. By experimenting with different prompts and requesting multiple drafts, users can uncover the strengths and weaknesses in an AI's responses. Understanding that an AI is unlikely to repeat itself unlocks the potential to generate a wide range of creative outputs through subtle prompt variations.

Table of Full Results

Below are the full-text responses from each AI.

Additional Responses

We used the same prompt and input it into several generative platforms. Here are the responses we received.