Ask AI Twice. Start the Second Time Fresh.

Context accumulates Verified facts ideas / assumptions / hopes The conversation kept growing. The evidence didn’t. An AI system once put a value near $250 million on a business idea I was exploring. The company had no paying customers, no revenue, no outside investors, and no transaction to support a number like that. I remember challenging the valuation. I no longer have the transcript, so this is my recollection, not a documented case study. Several models had examined the idea and challenged one another before the number appeared. By then, it didn’t feel random. It felt earned. That feeling was the warning. Possibilities had been discussed and refined, but discussion had not created evidence. The conversation had grown more complete. The case for the number had not. 01 / THE FRESH CONTEXT TEST The history can become part of the answer A long exchange can make an AI more useful. It can clarify what you want, add important context, and help you work through a hard question. But it can also carry forward tentative ideas, assumptions, and hopes until they start to sound like established facts. The answer may reflect not only the facts you supplied, but the path you took to get there. It can become hard to tell the difference from inside the same conversation. This does not mean the recommendation is wrong. It means you should inspect what it rests on before acting. Try the Fresh Context Test Pick a recommendation that could materially affect your work, money, career, health, relationships, or business. Then open a brand-new conversation. Leave out the first answer and the conclusion you hope to receive. Give the new conversation only facts you can verify, and ask the same decision question. Compare the reasoning. Did the fresh conversation ask about missing information? Did it make different assumptions? Did its confidence or recommendation change? Which facts or unknowns explain the difference? If the answers agree, that is some evidence the recommendation is less dependent on the original conversation’s history. It is not proof that either answer is right; both can share the same blind spot. If they disagree, that does not make the first answer wrong. The cause might be missing context, ambiguity, framing, or ordinary variation between responses. Either way, the useful next question is: What changed, and what does that reveal about what each answer depends on? 02 / WHAT THE TEST CAN TELL YOU Make the foundation visible A recommendation is easier to assess when you can separate four things: the facts you supplied, the assumptions the AI added, the information still missing, and the reasoning that connects them to its advice. Ask the model to lay those out. Then check its factual claims yourself. A fresh conversation helps test how much the original history may be steering the answer. It does not independently verify facts, expose a model’s hidden reasoning, or guarantee a more accurate result. Keep the comparison useful The same check can help with a job offer, a purchase you have talked yourself into, or a business plan that has gathered momentum. Keep the decision and verifiable facts as consistent as you can across the two conversations. If you change the question as well as the context, you will not know which difference mattered. Start fresh before you decide Choose one important recommendation you have received recently. Put the verifiable facts in a new conversation and ask again. If the answer shifts, resist the urge to choose a winner. Find the assumption or missing fact that could explain the shift, then verify what you can before you act. You may reach the same conclusion. You may not. Either result can help you understand what the recommendation depends on. What important recommendation would you want to test from a fresh start? Reply with the decision, not any private details, and the fact you would verify first. SOURCE AND SCOPE Adapted from Don Long’s supplied essay, “How to Get Better Answers From AI,” dated July 20, 2026. The $250 million valuation, business circumstances, multi-model process, and author’s challenges are reported from Long’s recollection; he says the original transcript is unavailable. This issue treats the story as an illustration, not a documented case study or evidence that multi-model review caused the valuation. The Fresh Context Test is a proposed practical check, not a validated accuracy test. Agreement between two conversations does not establish correctness, and a changed answer does not establish which one is right.

Ask AI Twice. Start the Second Time Fresh.

Don Long

Continue reading.

AI Wasn’t Wrong. It Answered Too Soon.

Three Checks Before You Trust an AI Recommendation

Don Long

AI is changing the work.Let’s be precise about how.

Continue reading.

AI Wasn’t Wrong. It Answered Too Soon.

Three Checks Before You Trust an AI Recommendation

AI is changing the work.
Let’s be precise about how.