A Polished AI Answer May Still Be Carrying a Contradiction

One of the easiest ways to be misled by AI is to hand it contradictory instructions and then admire the sentence it produces.

Imagine an event request that says the kickoff is Tuesday. A later note says Thursday. It asks for complete beginners but assumes people already know prompt engineering. It demands eight detailed agenda items in fewer than 120 words, and it forbids both assumptions and clarifying questions.

An AI system may respond with a smooth invitation. It may choose Thursday without saying why, use technical language, compress the agenda until it is unusable, and never mention the impossible instruction about assumptions. The result can look finished because fluent prose has a way of making choices feel settled.

They are not settled. The conflict is still there. The responsible question is not “Does this sound good?” It is “Which instruction did this answer silently choose, and who had authority to choose it?”

That question is useful far beyond event notices. Conflicting context appears in project notes, policy drafts, customer requests, long chat threads, and meetings where several people remember the decision differently. A tool can help expose the issue, but it cannot establish which source is current or whose priority should win.

The repair is humble. State the confirmed fact. Name the audience. Set a realistic format. Permit one clarification question when a necessary detail is missing. Then compare the revised answer to the original. The goal is not to make the chatbot more obedient. The goal is to make uncertainty visible before someone acts on it.

Try the exercise with an invented prompt. Predict what the tool will do, run it, label the contradiction, and repair the context. You will start noticing the warning sign in real work: an answer that sounds decisive but never explains how a conflict was resolved.

Draft for review. This is an educational practice, not a method for resolving consequential decisions without appropriate human oversight.

---

Source: labs/contradictions-in-context/companions/article-medium-draft.md