Use your browser's print dialog — set paper size to Letter, margins to Default. Prints 3 pages.
AI Prompt Cheat Sheet
goldengeeks.org
The Prompt Formula
1Context— Who you are or what situation you're in
2Task— What you want AI to do
3Tone— How it should sound (professional, friendly, brief)
4Format— How you want the answer (list, email, paragraph)
+Constraints— Any guardrails ("keep it under 200 words," "avoid jargon")
Tips That Make a Difference
Give AI a roleStart with "You are a [doctor / lawyer / travel agent]…" for more expert-sounding answers.
Free to try, but with limitsAll four tools are free to start — no credit card required. But free plans use older, less capable models and have daily usage caps. Paid plans (~$20/month) make a noticeable difference.
AI knowledge expiresModels are frozen in time. For current info, paste in a URL or tell it today's date.
Privacy firstTreat AI like a public forum — no passwords, Social Security numbers, or private medical records.
4 Free Tools to Try
Claude
claude.ai
Best reasoning & writing · privacy-friendly · great for long documents
ChatGPT
chatgpt.com
Image generation built in · name recognition · widest app integrations
Gemini
gemini.google.com
Best for Google users · Gmail & Drive built in · image generation
Copilot
copilot.microsoft.com
Best for Microsoft users · Outlook & Word built in · built into Windows
4 Stages of AI Use
Stage 1 — Apps
Ask questions, get answers. Typing, voice, photo uploads.
Stage 2 — Apps+
Projects with memory, image generation, deep research.
Stage 3 — Connected AI
AI reads your email and calendar and acts when you ask.
Stage 4 — Autonomous Agents
AI works in the background on a schedule, without being asked.
How AI Actually Works
goldengeeks.org
Under the Hood
How AI Generates AnswersAI predicts the next word, one at a time, based on patterns in billions of training examples. It's completing a very sophisticated sentence — not researching, not thinking, not looking anything up. This is why it sounds confident even when it's wrong.
HallucinationWhen the pattern-completion produces a plausible-sounding but false answer. AI doesn't know it's wrong — it has no internal fact-checker. It can cite a made-up book, invent a statistic, or name a doctor who doesn't exist, all with complete confidence. Always verify facts that matter.
Why Long Conversations Get FuzzyAI can only "see" a limited amount of text in any one conversation. In very long exchanges, it may lose track of what you said at the beginning. Starting a fresh chat for a new topic keeps its attention focused and its answers sharp.
Not a Search EngineUnless it has web browsing turned on, AI generates answers from memory — it doesn't look anything up. A generated answer is not a retrieved fact. Think of it as a very well-read friend, not a database.
Using It Well
It's a conversation, not a search boxAsk follow-up questions. Say "explain that differently" or "give me an example." Most people get better results on the second or third exchange than the first.
Push back if you don't like the answer"Try again," "make it shorter," "use simpler words" — all of these work. You're not stuck with the first response.
New topic = new sessionEvery message adds to the conversation's running context. Old context can bias new answers. When you switch topics, start a fresh chat.
Verify what mattersDates, names, statistics, medical and legal advice. AI can be wrong with complete confidence. Cross-check anything important against a real source.
Give it a sourcePaste in a URL or a document when accuracy matters. Grounding the AI in a real source dramatically reduces the chance of a made-up answer.
AI Companies: Safety & Ethics Focus
1Anthropic (Claude)— Founded specifically to make AI safe; uses written principles to guide behavior; operates as a public benefit company
2Google DeepMind (Gemini)— Strong safety research team; has published clear policies on responsible AI development
3OpenAI (ChatGPT)— Started with a strong safety focus; commercial growth has created some tension with the original mission
4Microsoft (Copilot)— Enterprise-focused safety standards; relies on OpenAI's models with Microsoft's oversight added on top
Reflects public information and reasonable expert assessment — not a definitive judgment.