Skip to main content

How we prevent AI hallucination

Understand what AI hallucinations are, and the steps we take to keep Ko's answers accurate and checkable.

Written by Nima Tabatabai

What are hallucinations?

In trying to be helpful, AI assistants can occasionally produce answers that are incorrect or misleading. This is known as "hallucinating."

It's a known limitation of today's generative AI models. A model can state something that looks authoritative and sounds convincing, but isn't grounded in fact. It might confuse details, mix up sources, or fill a gap with a plausible-sounding guess. In other words, AI can write things that look correct but are mistaken.

This matters most for high-stakes, market-specific details — settlement periods, price caps, auction rules, regulatory thresholds, and the like — where a small inaccuracy can change a decision.

Ko is built specifically to avoid this.

How we keep Ko's answers accurate

Ko is designed to give answers you can trust and check for yourself.

Grounded in Modo's own data and research. Ko builds every answer from Modo Energy's datasets, articles, methodology, and help docs — not from the open internet.

Sources are cited. Every article, Energy Academy lesson, podcast, or help doc Ko refers to appears as a clickable citation. You can open the source and check it yourself.

Ko tells you when it doesn't know. Ko is designed not to guess on market-specific details — settlement periods, price caps, auction rules, regulatory thresholds, and similar. If something isn't in Modo's coverage, Ko says so rather than inventing an answer. Customers tell us this is one of the main reasons they trust it.

No open-web browsing. Ko has no access to the wider internet. It only draws on Modo's own content, which removes the most common source of inaccurate AI answers.

How we test and monitor Ko

Accuracy isn't a one-time check. We verify it continuously as Ko changes.

An automated eval suite. We maintain a suite of evaluations covering core capabilities, regression checks, and deliberately hard reasoning tasks. We run these to confirm answer quality and catch any regressions before they reach you.

Ongoing monitoring. We automatically review how Ko performs on real questions and act quickly on anything that looks off, so issues are caught and corrected fast.

Backed by deep domain expertise. Modo's analysts continually give Ko the context and corrections that keep its understanding of the energy market sharp.

What to do if you're unsure

As with any AI, responses may occasionally be inaccurate. A few habits help:

  • Check Ko's cited sources. The original article or dataset may carry context worth reading in full.

  • For high-stakes decisions, don't rely on a single AI answer. Probe answers with follow up questions or ask Ko to verify its own answer using other data sources.

  • Use the thumbs up or thumbs down after each response. Your feedback helps us improve Ko.

  • Have a concern about a specific answer? Ask our team to verify it. Get in touch and we'll check it for you.

Did this answer your question?