How to use AI: a practical beginner's guide

What AI can and cannot do

Learning how to use AI starts with a healthier expectation: tools like ChatGPT, Claude, and Gemini are drafting and reasoning machines, not answer machines. They can turn rough notes into a clean outline, compare options, rewrite copy in a specific tone, explain a confusing concept, extract fields from a document, generate code examples, or help you pressure-test a plan. Used well, they give you a faster first pass and a sharper second pass.

The catch is that generative AI is probabilistic. It sounds equally confident whether it has understood your context or invented half the details, and its tone never warns you which one happened. That is why beginners should avoid asking, "What should I do?" as a stand-alone prompt. Better prompts give the model a role, task, context, constraints, and output format. Better workflows also include verification before anything important gets shipped.

A useful beginner workflow has four moves: define the job, give the model enough context, ask for a structured output, then review the result against sources or your own standards. The rest of this guide is that framework, with the checks attached.

The beginner prompt template

Use this four-part template whenever you prompt an AI chatbot: Persona, Task, Context, Format. It is simple enough to remember and specific enough to improve almost every output.

Persona: Tell the model what perspective to use. Examples: "Act as a senior product marketer," "Act as a careful research assistant," or "Act as a patient tutor." Persona should shape judgment, not create fake authority. If the output needs legal, medical, financial, or safety review, say so and keep a human expert in the loop.

Task: State the action clearly. Weak task: "Help with this." Strong task: "Turn these notes into a 600-word client update with three sections: progress, risks, and next steps." The task should contain the verb: summarize, compare, rewrite, extract, critique, plan, debug, classify, or generate.

Context: Give the model the material it needs to avoid guessing. Include audience, goal, source text, background, constraints, examples of good output, and anything it should not do. For long inputs, put instructions before the source material and label the source material clearly so the model can separate your request from the content.

Format: Specify the output shape. Ask for a table, checklist, email, outline, JSON-like fields, decision memo, bullets, or a draft with headings. Format is one of the easiest ways to get better results because it turns vague assistance into a deliverable.

Copy-paste master template: "Act as [persona]. Your task is to [specific task]. Context: [audience, goal, source material, constraints, examples]. Requirements: [must include, must avoid, tone, length]. Output format: [table/checklist/email/outline/memo]. Before finalizing, list any assumptions or missing information."

25 prompts by scenario

Use these beginner prompts as working starters. Replace the bracketed details, paste your source material where needed, and run the same prompt across more than one model when the decision matters.

1. Work plan: "Act as an operations lead. Turn this goal into a one-week execution plan. Goal: [goal]. Context: [team, deadline, constraints]. Output a table with day, priority, owner, deliverable, and risk."

1. Meeting notes: "Act as an executive assistant. Summarize these meeting notes into decisions, open questions, owners, and deadlines. Keep action items specific and flag anything ambiguous. Notes: [paste notes]."

1. Email draft: "Act as a concise business writer. Draft an email to [audience] about [topic]. Goal: [outcome]. Tone: warm, direct, not salesy. Include a subject line and a clear next step."

1. Email rewrite: "Rewrite this email so it is clearer, shorter, and more confident without sounding harsh. Keep all factual details unchanged. Email: [paste email]."

1. Brainstorm: "Generate 20 ideas for [project]. Constraints: [budget, audience, channel]. Group ideas by low effort, medium effort, and bold bets. Add one sentence explaining why each could work."

1. Decision memo: "Act as a strategy partner. Compare these options: [options]. Use criteria: cost, speed, risk, user impact, reversibility. End with a recommendation and the strongest counterargument."

1. Research questions: "I am researching [topic]. Create a research plan with the 10 questions I should answer, the evidence needed for each, and likely source types to check."

1. Source summary: "Summarize this source for a busy operator. Include thesis, key evidence, useful statistics, caveats, and what I should verify independently. Source: [paste text]."

1. Market scan: "Act as a market research assistant. Build a competitor scan for [category]. Columns: company, target customer, positioning, pricing signal, strongest claim, weakness, source needed."

1. Customer language: "Extract customer language from these reviews or calls. Group phrases by pain, desired outcome, objection, and buying trigger. Do not invent quotes. Text: [paste text]."

1. Content outline: "Create an SEO-friendly outline for an article targeting [keyword]. Audience: [audience]. Include search intent, H2s, examples to include, and what would make the article genuinely useful."

1. First draft: "Write a first draft of [asset] for [audience]. Use this outline: [outline]. Make it specific, practical, and free of generic filler. Ask for missing details before making claims."

1. Editing pass: "Edit this draft for clarity, structure, and usefulness. Do not change the meaning. Return: biggest issue, revised draft, and five suggested cuts. Draft: [paste draft]."

1. Tone match: "Rewrite this in the style of the example below. Preserve facts and structure, but match sentence length, directness, and level of detail. Example: [paste example]. Draft: [paste draft]."

1. Slide outline: "Turn this idea into a 7-slide outline. Each slide needs a title, one key point, supporting evidence needed, and speaker note. Idea: [paste idea]."

1. Explain simply: "Explain [concept] to a smart beginner. Use an analogy, then a practical example, then three common mistakes to avoid."

1. Learning plan: "Create a 14-day learning plan for [skill]. I can spend [time] per day. Include daily practice, a checkpoint, and a small project by day 14."

1. Code explanation: "Explain what this code does, where it may fail, and what tests would increase confidence. Keep the explanation accessible to a junior developer. Code: [paste code]."

1. Debugging: "Act as a careful debugging partner. Given this error, code, and expected behavior, list the most likely causes, the smallest next diagnostic step, and a safe fix. Error/context: [paste]."

1. Code review: "Review this change for correctness, security, edge cases, and missing tests. Prioritize real risks over style preferences. Diff: [paste diff]."

1. Data cleanup: "Turn this messy list into a clean table. Infer categories only when obvious, mark uncertain fields as unknown, and list cleanup rules used. Data: [paste data]."

1. Spreadsheet formula: "I need a spreadsheet formula for [goal]. Columns are [columns]. Explain the formula and include one example row."

1. Personal productivity: "Plan my day from this task list. Constraints: [meetings, energy, deadlines]. Group tasks into deep work, admin, quick wins, and defer. Task list: [paste]."

1. Responsible use check: "Review this AI-assisted output for risks. Check for unsupported claims, privacy concerns, bias, missing caveats, and places that need human review. Output: [paste]."

1. Model comparison: "I am testing AI models for [task]. Score this output from 1-5 on accuracy, completeness, clarity, usefulness, and risk. Explain the score and suggest one better follow-up prompt. Output: [paste]."

How to verify outputs

The most important skill in using generative AI is not prompt cleverness. It is verification. Treat AI output like a strong draft from a fast assistant: useful, but not automatically true. The higher the stakes, the more verification you need.

Run this checklist before you publish, send, or act on AI output. It takes a couple of minutes and it catches almost everything that goes wrong.

1. Check the facts. Identify every factual claim and check it against a primary source or a trusted internal document.
2. Hunt for hidden assumptions. Dates, prices, policies, customer segments, technical constraints, and legal requirements are where models quietly guess.
3. Check the format. Did the answer actually follow the structure you asked for, or did it drift back to prose?
4. Remove invented certainty. Add caveats wherever the evidence is incomplete, because the model will not.
5. Check privacy. Do not paste customer, employee, health, financial, credential, or proprietary data into a tool your organization has not approved.
6. Get a second opinion. Ask a different model, or the same model with a critique prompt, to find the weakest claim.
7. Read it yourself. A final human pass for tone, judgment, and context, which is the part no model can do for you.

For research, require citations or source notes and verify the cited source yourself. For writing, compare the draft to your voice and remove generic claims. For code, run tests and review the diff. For strategy, ask for counterarguments. For long documents, ask the model to quote or point to the section that supports each important conclusion, then check the source material directly.

Picking the right model for the job

You do not need to pick one AI model forever. The better habit is to match the model to the job, then compare outputs when quality matters. OpenAI, Anthropic, and Google all publish prompting guidance, and it points at the same practical idea. Different models suit different tasks, and the gap widens as you move between fast drafting, hard reasoning, long documents, coding, and multimodal work. Even the free tiers differ in shape: Claude's free plan meters usage in a session window that resets every five hours, while ChatGPT's free plan caps flagship-model messages before dropping you to a lighter fallback.

The job	What to reach for	What to judge it on
Writing, outlining, email, ideation	ChatGPT for range, Claude for drafts that need less editing	Specificity, tone control, and whether it followed your format
Coding	Claude, which reviews code as carefully as it writes it	Does it ask for missing context, propose tests, and prefer a small fix to a rewrite
Research	Perplexity or any search-first workflow with links attached	Can you follow every claim back to something you can check
Long documents	Gemini, with context windows documented at a million tokens and above	Ask for extraction before synthesis, then verify against the original
Images, PDFs, screenshots	Gemini or ChatGPT, where camera and file input are first class	Ask for structured extraction before interpretation

A simple decision rule: use a fast model for drafts and brainstorming, a stronger reasoning model for decisions and debugging, a long-context model for large source material, and a multimodal model for images or PDFs. Then A/B test the same prompt. If one answer is more accurate but another has better structure, ask the stronger model to revise using the better structure. That is often faster than forcing one model to do every job. Buying that flexibility directly is the expensive route: ChatGPT Plus, Claude Pro, and Google AI Pro each list at about $20 a month in the United States as of August 2026, and OpenAI's cheaper Go tier still adds $8. If you are still choosing where to start, the best AI apps for beginners guide compares the main options by task.

Your first 15 minutes in Whizi

The fastest way to make these habits stick is to run small, concrete comparisons. Start in Whizi with five prompts from this guide: one email rewrite, one meeting summary, one research plan, one decision memo, and one verification check. Use the same prompt across models and compare the outputs side by side.

Minute 1-3: choose one real task from your day, not a fake demo. Minute 4-6: paste the master prompt template and fill in persona, task, context, and format. Minute 7-10: run the prompt across models and mark which output is clearest, most useful, and least risky. Minute 11-13: ask the best output for a revision with one specific constraint, such as shorter, more evidence, or more direct. Minute 14-15: save the prompt pattern so you can reuse it. AI workflow templates has seven of these routines already written out if you would rather start from one than build your own.

For research-heavy work, try the Founder Research Stack at /templates/founder-research-stack. For more learning paths, browse /resources. When you are ready to turn experimentation into a repeatable workflow, create an account at /register and compare plans at /pricing.