An AI answer arrives finished. No working, no sources, no hesitation. That polish is the problem: it looks equally trustworthy whether it is right or invented. Here is a practical routine for checking one, and for deciding when checking is not worth your time.
At what temperature does water boil at 3,000 metres above sea level?
Start by finding the checkable claim
Most answers are a mix of safe generalities and one or two specific claims. The generalities are rarely wrong and rarely useful. The specifics are where errors live: a number, a name, a date, a legal threshold, a command, a citation.
Underline those. You do not need to verify an answer. You need to verify two sentences in it.
Use agreement as a filter, not a verdict
Ask several families and look at the specific claims side by side. This question has a real answer that varies with conditions, so it is a good test.
At what temperature does water boil at 3,000 metres above sea level?
DeepSeek
1.1 s ✓At 3,000 metres above sea level, water boils at approximately 90°C (194°F).
This is because atmospheric pressure decreases with altitude, so less heat is needed for water to vaporise. The exact temperature can vary slightly with weather conditions, but 90°C is the standard figure for that elevation.
Qwen
3.3 s ✓At 3,000 metres (approximately 9,843 feet) above sea level, water boils at approximately 90°C (194°F).
Explanation:
- Atmospheric pressure decreases with altitude.
- At sea level, standard atmospheric pressure is about 101.3 kPa, and water boils at 100°C.
- At 3,000 metres, the atmospheric pressure drops to roughly 70 kPa.
- Because there is less pressure pushing down on the surface of the water, it requires less energy (lower temperature) for water molecules to escape into the gas phase, resulting in a lower boiling point.
GLM
21.8 s ✓At about 90 °C (194 °F).
Atmospheric pressure drops as you climb, and water boils when its vapour pressure matches the surrounding air pressure — so the boiling point falls roughly 1 °C for every 300 m of elevation. At 3,000 m the air pressure is about 70% of sea-level pressure, putting the boiling point around 90 °C. Day-to-day weather can shift it a degree or so either way.
This is why cooking takes longer at altitude: food simmering in 90 °C water cooks more slowly than at 100 °C.
MiniMax
2.6 s ✓At 3,000 metres above sea level, water boils at approximately 90°C (194°F).
This is about 10°C lower than at sea level (100°C). The reduction happens because atmospheric pressure decreases with altitude — at 3,000 m, air pressure is roughly 700 hPa (compared to about 1,013 hPa at sea level). With less pressure pushing down on the water's surface, it needs less heat energy to form vapour bubbles and boil.
A handy rule of thumb: the boiling point drops by roughly 1°C for every 285 metres (about 1,000 feet) of elevation gained.
GPT-OSS
2.2 s ✓Water boils at roughly 93 °C at an altitude of 3,000 m (about 70 kPa or 0.7 atm). The exact value varies slightly with local temperature, humidity, and barometric pressure, but it’s around 92–94 °C.
Where the numbers match, your remaining risk is a shared mistake. Where they differ, you have found the thing to look up. You also know what a plausible range looks like, which makes looking it up much faster. Run your own checkable question on the compare page.
Six habits that catch most errors
1. Ask for the source, then check the source exists
Invented citations are common and convincing. Search for the exact title. If it does not exist, treat the whole answer as unreliable, not just that line.
2. Prefer primary sources
For rules and thresholds go to the body that sets them: the tax authority, the regulator, the standards organization, the manufacturer’s documentation. Secondary summaries are where old numbers survive.
3. Watch for the confident date
Models are frequently out of date and rarely say so. Anything involving a price, a rate, a law, a version or a deadline should be assumed old until confirmed.
4. Re-ask the same question differently
Change the wording and ask again. A fact that is genuinely in the model comes back the same. A guess wobbles.
5. Ask it to argue against itself
“What would make this answer wrong?” often surfaces the assumption that was doing all the work.
6. Do the arithmetic yourself
Models are much better at arithmetic than they were and still make mistakes in multi-step calculations. Any number you will act on deserves ten seconds with a calculator.
When not to trust any model
- Anything about a specific person, especially something negative.
- Anything where the answer depends on today, or on where you are.
- Medical, legal, financial and safety decisions. Use the answer as a list of questions for a professional, not as the answer.
- Anything you would have to publish. Names, quotes and statistics need a source you have seen.
- Anything where you cannot tell a good answer from a bad one. That is the most dangerous case, and the one where comparing several families helps most.
Our accuracy and limitations page goes into why this happens rather than what to do about it.
A two-minute routine
- Ask the question on the home page with three families answering.
- Read the shortest answer to get the shape.
- Scan for specific claims and note where the families disagree.
- Check the one claim that matters against a primary source.
- Ask a follow-up: “What is the main assumption in your answer?”
- Decide. If you still cannot verify it and it matters, ask a person.
Frequently asked questions
Can an AI model tell me when it is unsure?
It can produce hedging language, but that language is not a reliable measure of correctness. A model can be confidently wrong and cautiously right in the same paragraph.
Does asking the same model twice help?
Yes, a little. Because answers are sampled, a shaky claim often changes between attempts while a solid one does not. Asking different families is a stronger test.
What about the "Where they agree" summary?
It is written by one of the model families from the other answers, so it inherits their limits. Use it to find the interesting part of the page, not as confirmation.