Can AI predict the future?

Special guest blog by Peter Eigenschink.

I recently raised this question to ChatGPT. And, at first, the answer sounded like a sensible no. But at a second look, it read more like a fortune-teller, confidently inviting me to ask about my future career or how many kids I'm going to have.

My recent conversation with ChatGPT.

While ChatGPT assures it would estimate likely outcomes based on patterns, data, psychology, and trends ... like probability forecasting, knowing a bit about how LLMs work internally, I have my doubts that this is what's actually happening. So, risking to uncover that ChatGPT lied to me, again, let's explore whether LLMs can actually predict the future and what might hold them back.

First, let's acknowledge that LLMs are indeed powerful prediction machines. From vast amounts of text, they learn to predict the most likely words to follow a prompt. Combined with human feedback, this yields impressively realistic texts, wrapped in human-like chat experiences that power many of the AI applications we see today.

So in a sense, LLMs are predicting the future, but in the dimension of subsequent words, not in the dimension of actual time. This distinction fundamentally limits what kinds of predictions we can reasonably expect from LLMs or other generative AI models.

Now, suppose you want to forecast revenue from new business for the quarter or prioritize your opportunities and identify actions to maximize profit. An LLM will confidently provide an answer. But can you trust it? If not, do more context, more data, better prompting, or a more capable model lead to a reliable answer or is there an inherent limitation?

The four categories of business analytics methods.

Looking at it from a business analytics perspective, we have four kinds of methods in our toolkit to help us make better decisions in the future [1].

• Descriptive analytics tells us what happened in the past.

• Diagnostic analytics explains why the past is the way it was.

• Predictive analytics estimates how the future might look like based on the past.

• Prescriptive analytics tells us what to do to achieve future outcomes.

Methods from each category leave more or less of the predictive work to us. Descriptives only make past data more digestible for us, while prescriptive methods already tell us what to do to achieve our goals.

When applied to the real world, LLMs mostly operate in the realm of descriptive analytics. We can ask them to summarize sales call transcripts or do research on prospects for us. But when we ask them about future outcomes, they are still just going to predict words by remixing the vast amounts of past information they have access to. And, being good with words, they'll eloquently try to convince us of their prediction. However, for most business decisions, this approach won't provide the desired results.

So let's go back to the initial question: Can AI predict the future?

If we care about future outcomes in the real world, not in the word-sequence future the LLM lives in, we should not rely on LLMs' predictions. To forecast revenue, simulate price changes, or prioritize opportunities, we need to move beyond describing the past. This is where predictive and prescriptive statistical methods still excel. They are purpose-built to estimate results for specific situations, quantify how likely predicted outcomes are, and require far less data and resources than an LLM does. We understand how they work, when to apply them, and where they break down. While generative AI is currently often top-of-mind when people talk about AI, these analytical AI methods have a much longer track-record of delivering actual business outcomes [2].

So, AI can indeed help us predict the future, not through LLMs and generative AI, but, case-by-case, with purpose-built analytical AI models.

[1] Cote, C. (2021, October 19). 4 Types of Data Analytics to Improve Decision-Making. Harvard Business School, Business Insights. https://online.hbs.edu/blog/post/types-of-data-analysis

[2] Wladawsky-Berger, I. (2025, June 5). The Key Differences Between Analytical and Generative AI. https://blog.irvingwb.com/blog/2025/06/how-analytical-and-gen-ai-differ-and-when-to-use-each.html

(c) 2026 Peter Eigenschink