AI is a fast research assistant and a patient data helper, but it is not a source. Use it to find, sort, and explain information, then check what matters yourself.
Start here: go toe to toe with it
You won't understand what AI can do by asking it trivia. The best test is a conversation about something you already know well, such as your job, a hobby, or a skill you have spent years on. There you can tell right away when it is right, when it is wrong, and when it shows you something new.
- Pick a topic you know and enjoy. Fishing, baking, car repair, gardening, your trade, a sport, a period of history.
- Bring a real pain point. Something that annoys you, confuses you, or that you have never solved. "My sourdough always comes out flat." "How do I price my side work?"
- Push back. When it says something you know is wrong, tell it, and explain why. Watch how it responds. This shows you its limits faster than anything else.
- Notice what is new. Write down at least one idea, method, or fact you didn't know. Then check it.
- Keep going for 20 minutes. The first answer is the least interesting one. The good stuff comes in the follow-ups.
Then turn it into something you keep. Ask it to write up what you learned:
- A plan you can follow: "Turn this conversation into a 4-week plan to fix my sourdough."
- A one-page report you can come back to later
- A note for a friend who shares the interest: "Write a short message to my fishing buddy about the three best tips from this chat."
By the end you will have a feel for where AI is strong, where it slips, and how to steer it. The rest of this guide builds on that.
Prompts that help you understand. If an answer goes over your head, don't give up. Ask it to explain again, a different way. Each of these works with any topic. Add it after any answer you didn't fully get.
| Ask this | What you get | Example |
|---|---|---|
| "Explain this to me in layman's terms." | The same answer without jargon | After a reply about mortgage rates: "Explain this to me in layman's terms." |
| "Explain this to me in middle school graduate terms." | Simpler still: short sentences, everyday words | "What is inflation? Explain it in middle school graduate terms." |
| "Explain this step by step, assuming I don't understand the industry. Use middle school graduate terms, and give me some examples and scenarios so I can see different perspectives and viewpoints." | A walkthrough with real-life situations, from more than one angle | "How does a restaurant make money? Explain it step by step, assuming I don't understand the industry..." |
| "Assume there are two sides to this: one that agrees and one that disagrees. Show me both sides." | The case for and against, so you can decide for yourself | "Should I lease or buy my next car? Show me both sides." |
The two-sides prompt is especially useful because AI tends to agree with whatever you seem to believe. Asking for both sides gets you a real second opinion instead of an echo.
Research
The best research results come from three habits: turn on web search, give context, and ask for sources.
-
Turn on web search for anything current. Without search, a model answers from memory, and its memory stops at a cutoff date. Prices, news, laws, and "who runs what" all change. Most chat apps have a search or web toggle. Perplexity searches by default.
-
Give context, not just a topic. Compare these two:
- Weak: "Tell me about heat pumps."
- Strong: "I live in central Texas in a 1,800 sq ft house built in 1995. Compare a heat pump to my current gas furnace on upfront cost, monthly bills, and lifespan. Assume hot summers and mild winters."
The strong version tells the AI who you are, what you are deciding, and what to compare.
-
Ask for sources, then open them. Ask: "List your sources with links." Then click at least the ones that support the numbers you care about. AI sometimes cites a real page that does not say what it claims, or invents one entirely.
-
Go back and forth. Research is a conversation. Ask follow-ups like "What would someone who disagrees say?" or "What is the weakest part of this answer?"
-
Use deep research for big questions. Several apps offer a "deep research" mode that runs dozens of searches and writes a long report. It takes minutes, not seconds, and is best for questions with many parts.
Data
Modern AI tools can open a spreadsheet, do the math with real code, and draw a chart. Your job is to ask a clear question and check the result.
- Upload the file. Attach a CSV or Excel file to the chat. Remove anything private first: account numbers, ID numbers, other people's personal details.
- Ask it to describe the data first. "What columns are in this file, how many rows, and is anything missing or odd?" This catches problems before you build on them.
- Ask one question at a time. "Which month had the highest spending?" works better than "analyze everything."
- Ask it to run the numbers, not guess. Say "calculate this with code." Models are good at reasoning and bad at mental arithmetic. Most chat apps can run code behind the scenes when asked.
- Ask for a chart. "Make a bar chart of spending by category." Then ask: "What stands out?"
- Spot-check one number yourself. Pick one total and add it up in your own spreadsheet. If it matches, trust the rest more.
What AI is good at here: cleaning messy data, explaining what a formula does, writing spreadsheet formulas for you, and summarizing patterns. What it is bad at: knowing whether your data is complete, and catching errors in the data you gave it.
Prompts you can copy
Replace the words in [brackets] with your own.
| Goal | Prompt |
|---|---|
| Quick overview | "Explain [topic] to someone with no background in 5 short paragraphs. End with the 3 questions I should ask next." |
| Compare options | "Compare [A], [B], and [C] for [my situation] in a table: cost, pros, cons, who it suits. Use current sources and link them." |
| Fact check | "Is this claim true: [claim]? Give the best evidence for and against, with links, and say how confident you are." |
| Summarize a document | "Summarize the attached file in 5 bullet points, then list anything that seems wrong or missing." |
| Read a spreadsheet | "Describe this file's columns and any data problems. Then tell me [question], calculated with code." |
| Challenge yourself | "Here is what I believe: [belief]. Give me the strongest case against it." |
Before you trust it
- Did I turn on web search for anything that could have changed recently?
- Did I open the sources behind the numbers I plan to use?
- Did I check one calculation myself?
- Did I ask what the answer might be missing?
- For health, money, or legal decisions, did I confirm with a qualified person?
Part 3 covers in detail why these checks matter.
Found this helpful? Raya.ai is built on the ideas in this series. Learn more at raya.ai (opens in a new tab), or download the app on iOS (opens in a new tab) or Android (opens in a new tab).