Same data. Same prompt. Three completely different reports. Here’s what that means for every insight you’ve ever presented — and how to fix it.
GenAI has made analysis easy. It hasn’t made it reliable. There’s a critical difference — and it’s one that could be quietly undermining your credibility.
Anyone can upload a dataset to an AI model and ask it to generate insights. That part is solved. The harder question is whether those insights hold up when a stakeholder pushes back, asks you to reproduce them, or wonders why last week’s analysis looks different from this week’s.
In this webinar, Marcus Pemberton, VP of Customer Success at Relative Insight, runs a live demonstration that exposes exactly what happens when you run the same dataset through a GenAI model more than once — and why the results should give every analyst pause.
Three runs. Same data. Same prompt. Three different reports.
Duration: 6 mins | Available Now | Ideal for: CX, insights and operations leaders
What you’ll see in this Webinar
Marcus runs a live, three-way demonstration — the same dataset and prompt uploaded into ChatGPT three separate times — then runs the identical exercise in Relative Insight. The difference is immediate and hard to ignore.
1. Three ChatGPT runs on identical data The dataset: social media mentions of AI platforms. The prompt: identical each time. The outputs? Run one surfaces AI model competition and industry rivalry, leadership and industry figures, and AI coding and developer tools. Run two returns OpenAI and Sam Altman, model platform comparisons, and developer and coding use cases. Run three produces general discussion and community commentary, industry news, and product feedback. The same model. The same data. Three different reports.
2. What happens when a stakeholder asks you to show your working If your underlying analysis changes every time you run it, it becomes difficult to validate, difficult to defend, and impossible to reliably act on. In an era where AI tools are not just generating insights but informing decisions, the real question isn’t what AI can produce for you — it’s what you can actually trust.
3. The same exercise in Relative Insight Marcus loads the same dataset twice — run at different times — through Relative Insight’s Accelerator AI report. The output: identical themes, identical metrics, identical conclusions across both runs. AI model comparison. Data center infrastructure. Key industry figures and stock market performance. The analysis is deterministic: the platform tokenizes all text, analyzes it at word, phrase, and topic level, and applies co-occurrence logic to produce an output that can be reproduced, defended, and built upon.
Why reproducibility is the foundation of trusted analysis
Non-determinism in AI isn’t a flaw to be patched in the next model update. It’s a fundamental characteristic of how large language models work. For teams whose job is to turn data into decisions, that has real consequences:
- You can’t track trends reliably if your signals shift between runs
- You can’t take a finding to leadership if you can’t recreate exactly how you arrived at it
- You can’t build a consistent reporting cadence on outputs that vary by the hour
The answer isn’t to stop using AI. It’s to understand which parts of the analytical process require determinism — and to use tools that are built to deliver it.
What you’ll learn
- Why the same AI prompt returns different results each time — and why that’s not a quirk but a structural characteristic of generative models
- What “non-deterministic” means in practice — demonstrated live with real data and a real prompt, three times over
- Why reproducibility matters more than speed — and how the inability to recreate findings erodes stakeholder trust over time
- How a deterministic analytical engine works differently — tokenization, word and phrase-level analysis, co-occurrence mapping, and consistent output generation explained plainly
- What it looks like when the same data always returns the same answer — and what that consistency unlocks for your reporting, your credibility, and your team
Who should watch this webinar?
This is for anyone who has ever presented an AI-generated insight — and felt a flicker of doubt about whether they could stand behind it:
- Heads and VPs of CX or Customer Insights who need their analysis to hold up under scrutiny from operations, finance, or the executive team
- Insights and analytics managers whose teams are using GenAI tools to process customer or market data and sharing those outputs with senior stakeholders
- VoC program leads responsible for consistent, repeatable signals that track performance across time
- Research leaders who’ve been asked to embed AI into their workflows — and are now asking the right question: which outputs can we actually trust?
If you’ve ever been challenged on how you reached a finding — or worried you might be — this webinar is for you.
