- 1. The mistake that started this list
- 2. It predicts words, it doesn't "know" things
- 3. Confidence is not a signal of accuracy
- 4. Vague in, vague out — I tested this directly
- 5. It forgets you the moment the chat ends
- 6. Round two is where the good answer shows up
- 7. The free tier is not a bait-and-switch
- 8. It doesn't know what happened last week
- 9. Assume what you type isn't fully private
- 10. The skill that actually matters
The mistake that taught me most of this list: I once handed a client a project timeline that ChatGPT had summarized from a scope doc, without re-reading the original. It had quietly turned "3-4 weeks" into "3 weeks flat." I had to send an awkward correction email an hour later. Nothing was catastrophic, but it was completely avoidable, and it's exactly the kind of thing this list would have prevented on day one.
2. It predicts words, it doesn't "know" things
AI chatbots work by predicting the most likely next words based on patterns learned during training. That produces genuinely useful answers most of the time, but it means the AI isn't looking facts up in a database the way a search engine does — which is exactly how my "3-4 weeks" became "3 weeks."
3. Confidence is not a signal of accuracy
This is the one I'd tattoo on the inside of the chat window if I could. Wrong answers arrive in exactly the same tone as right ones — no hedge, no hesitation. Our fact-checking guide covers the specific habit that catches this.
4. Vague in, vague out — I tested this directly
"Write me a product description." Result: generic, could describe almost anything, needed a full rewrite.
"100-word description for a $40 soy candle, warm and a little funny." Result: usable in one pass, no rewrite needed.
Same tool, same model, same day — the only variable was how much I told it. Our guide to writing good prompts walks through exactly what to include.
5. It forgets you the moment the chat ends
Most AI conversations start fresh with no memory of your last chat, unless the tool specifically offers memory or you're using a custom assistant with saved instructions. If context matters, plan to restate it, or set up a Custom GPT that remembers the basics for you.
6. Round two is where the good answer shows up
I almost never use the first response as-is anymore. Treat it as a draft: tell the AI exactly what to change rather than starting over. Two or three rounds of refining consistently beats trying to write the perfect prompt on the first attempt.
None of this is meant to make AI sound unreliable. It's meant to save you the awkward correction email — knowing where the rough edges are is what lets you use the tool well from day one.
7. The free tier is not a bait-and-switch
You don't need to pay anything to get real value. I use the free tiers of at least two major assistants for most everyday writing and research, and only feel a limit during unusually heavy days. Upgrade once you actually hit a wall — not before.
8. It doesn't know what happened last week
Every AI model is trained up to a certain date and, unless it can search the web, won't know anything after that. If an answer seems oddly out of date or shaky about something recent, that's almost always why — worth checking whether web access is even turned on before you trust a current-events answer.
9. Assume what you type isn't fully private
Depending on the tool and plan, what you type may be used to improve the product, and reviewers can occasionally see flagged chats. If you're working with anything sensitive, check the privacy settings first, and never paste in passwords or other people's private information.
Put it into practice
The free AI Prompt Builder helps you write clear, well-structured prompts and assistants — no experience needed.
Open the AI Prompt Builder →10. The skill that actually matters
Almost everything above comes back to one idea: AI reflects the quality of what you give it. Learning to ask clearly, with real context and a clear goal, matters more than picking the "best" AI tool — and it's the fastest skill on this list to build. That's the whole point of our prompt-writing guide.
Frequently asked questions
Trusting a specific number, date, or quote without checking it — not because AI is unreliable generally, but because wrong specifics sound exactly as confident as right ones.
Generally yes, with normal caution: verify anything factual, avoid sharing sensitive data, and check your company's AI policy if you have one.
Pick one small, real task you actually have this week (see our first AI project guide), rather than reading about AI indefinitely. You'll learn more from one real attempt than from ten more guides.