Ask a real question, get a real answer with sources — not ten blue links you still have to read.
Most models won't cite unless you ask, and even then some 'sources' are made up. Ask for real, checkable links and then click them.
If a link is dead or doesn't say what the AI claims, throw out that claim.
The default AI answer is bloated with caveats. Tell it to lead with the answer and keep the rest short.
Word limits work. So does asking for the answer first, reasons second.
Paste the exact claim and ask the AI to trace it to a primary source, not a blog that repeated it.
Half the viral stats online are misquoted or ancient. Find the original and check the date.
Ask for a briefing sized to the time you have — 5 minutes reads differently than 30.
Then ask the three questions a smart person in that meeting would ask, so you're not blindsided.
Ask the AI to explain it to a specific person — a 12-year-old, your boss, a client with no background. The target changes everything.
Then ask what it left out, because plain explanations always cut corners.
These models predict likely-sounding text, not truth. When they don't know, they don't say so — they fill the gap with something plausible.
It sounds most confident exactly when you should be most careful, especially with names, dates, numbers, and quotes.
Google hands you links and makes you do the reading. AI reads for you and gives an answer — faster, but you're trusting its summary.
Best move: use AI for the answer, use search to check the answer. Neither replaces the other yet.
A trustworthy source is close to the event or data (primary, not third-hand), recent enough to matter, and has no reason to spin it. A random blog quoting a study is not the study.
Ask who made it, when, and what they gain. If those three check out, you're on solid ground.
Use it to understand the general shape of the issue and the words to search — not as the answer you rely on. Laws vary by place and change, and AI's training is frozen and sometimes wrong.
For anything that has real consequences, take what it gave you to a real professional. Treat the AI as a fast first read, never the final word.
That's a gift, not a bug. It just told you the answer is uncertain or context-dependent. Ask it directly why the two answers differ.
Then find one real source that settles it. If you can't, the honest answer is 'it depends on X' — and now you know what X is.
Paste the customer's exact question plus what your product or business actually does. Ask for a clear answer and the one follow-up question you should ask them.
Draft your reply, then read it as if you were the customer. If anything sounds like a guess, verify it before you hit send — a wrong answer costs more than a slow one.
They're close for everyday questions. The real differences show up in careful reasoning, live web access, and cost.
Try one, and if the answer feels thin, paste the same question into another. Cross-checking beats loyalty to one tool.
| Tool | Strongest at | Watch out for |
|---|---|---|
| Claude Opus 4.8 | Careful reasoning, long docs, honest about limits | Can over-explain; check live-web access |
| ChatGPT | General questions, live search, huge ecosystem | Confident on gaps; verify facts |
| DeepSeek | Free/cheap, solid reasoning, good for basics | Weaker on very recent topics |
AI is faster and good enough for low-stakes stuff. The original source is slower but it's the truth, not a summary of it.
The trick is knowing which the moment calls for. High stakes, going public, or anything you'll be quoted on — read the source yourself.
No, and don't. Trust the shape of the answer — the framing, the questions to ask, the general picture. Verify the hard facts, because it can invent a source or flip a number and never blink.
The people who get burned are the ones who paste the answer straight into an email or a paper. Spot-check the specifics and you'll be fine ninety-nine times out of a hundred.
Yes — because you can't check an answer you don't understand. The people who thrive with AI are the ones who know enough to catch it when it's wrong.
AI makes shallow knowledge useless and real understanding more valuable. It's a research assistant that's occasionally lying to you. You still have to be the one who knows.