We may earn a commission when you buy through links on this page. It doesn't change what we recommend.
Last checked . Picks are based on the prices, rates, fees and terms each provider publishes, checked at the source. We never rank products by commission. See how we’re funded.
Ask a chatbot a lazy question and you’ll get a lazy answer. Knowing how to write AI prompts mostly comes down to being specific, and OpenAI, Anthropic and Google all say so in their own documentation. Below is the advice they share, boiled down to a routine for email, summaries, trip planning and bills. The before-and-after examples are ours.
What AI prompts are, and why the details matter
A prompt is just the instructions and background you hand the chatbot. OpenAI’s help center compares it to briefing a coworker: say what the task is, give the background, set the tone, and split big jobs into smaller asks. Anthropic’s guidance puts it another way. Picture a smart new colleague who knows nothing about your situation. Would they be confused by what you wrote? If so, the chatbot will be too.
Tip one: write AI prompts that say exactly what you want
Vague prompts get vague answers. OpenAI’s developer guide says to spell out what you want done and what you don’t want. Google’s Gemini documentation asks for the same, plus limits like length. Want 120 words? Write 120 words.
Tip two: give your AI prompts the context the model lacks
The chatbot can’t see your manager’s name, your budget or the file sitting on your desktop, so tell it. OpenAI says to hand over any information the model wouldn’t otherwise have. Anthropic adds that explaining why a rule matters helps the model apply it sensibly, and that one sentence naming a role, such as “you’re a careful editor,” can steer the tone.
Tip three: state the format in your AI prompts
Tell the tool what the answer should look like: a table, a short list, an email with a subject line, a single word. Google suggests specifying a table, list or outline. Anthropic adds that it’s usually better to say what you want than what you don’t, such as asking for flowing paragraphs instead of saying “no bullet points.”
Tip four: show an example in your AI prompts
Examples help, and all three companies say so. Anthropic suggests three to five that differ from each other, and OpenAI warns that near-identical examples get copied. Google notes that piling on too many can backfire. Got a style in mind? Paste a short sample and write “match this tone.”
Tip five: break big jobs into steps
Big jobs go better in pieces. OpenAI suggests asking the model to plan and check itself as it works. Anthropic’s pattern is draft, review against your list, then refine, and Google recommends chaining prompts together. Easiest version: ask for an outline, approve it, then ask for the draft.
Four before-and-after examples
These are our own examples, not taken from any vendor.
Work email. Before: “Write an email about the project delay.” After: “Write me a 120-word email to my manager, Dana. The website launch is slipping from Nov 3 to Nov 17 because the vendor is late with the logo files. Keep it calm, own the delay and don’t blame anyone. One sentence on how I’ll make up the time, then ask for a 15-minute call Thursday. Give me a subject line and the body, with no bullet points.”
Summarizing a document. Before: “Summarize this.” After: “Here’s a six-page vendor contract. Explain it to me like I’m not a lawyer, in under 200 words. I want to know what I’m agreeing to, what it costs and when I pay, how I get out of it, and anything odd. Quote the exact sentence that covers cancellation. If the text doesn’t say, answer ‘not stated’ rather than guessing.” Then paste the text. Anthropic’s guidance recommends putting long documents first and your question after, and asking for quotes, which helps keep answers tied to the text.
Planning a trip. Before: “Plan a trip to Chicago.” After: “We’re two adults heading to Chicago for a long weekend in mid-May, with about $1,200 to spend before flights. We like food and architecture, skip the tourist traps and won’t rent a car. Give me a day-by-day table with morning, afternoon and evening, rough costs, and three questions that would make the plan better. Tell me what to confirm on official sites, like hours and prices.”
Explaining a bill. Before: “Why is my bill so high?” After: “Below is the text of my electric bill, with my account number and address removed. Walk me through each line in plain English. I’m a first-time renter. Which charges are fixed, and which move with usage? Give me three questions to ask the utility. If a rate isn’t on the bill, say ‘not shown’ instead of guessing.”
That last example shows a habit worth keeping: remove personal details before you paste.
Check the answers before you rely on them
Chatbots get things wrong. OpenAI’s own safety guidance for users says the model may sometimes produce incorrect information, so double-check important facts with trusted sources and ask a qualified expert before legal, medical or financial decisions. Anthropic’s guidance on reducing errors suggests letting the model say “I don’t know,” asking for a supporting quote behind each claim and dropping any it can’t support, and comparing answers from several runs. Those steps cut down on mistakes. They don’t remove them.
Be careful with private information
Google’s Gemini privacy hub says not to enter confidential information you wouldn’t want a human reviewer to see, since some conversations are reviewed by people. The page says turning off activity saving stops future chats from being reviewed for service improvement, but chats are still used to respond and protect users. OpenAI advises getting consent before sharing someone else’s data and following your workplace’s AI rules. Settings vary by product, so read the privacy page of whichever tool you use.
Iterate on your AI prompts, don’t restart
If the first answer is close, don’t start over. Google’s documentation recommends rephrasing, reframing or reordering your prompt, and OpenAI’s help center says to keep refining as in a conversation. Reply with what to change: shorter, more formal, add a cost column, remove the third paragraph.
A simple checklist for AI prompts
Before you hit enter, ask whether you’ve stated the task, the background, the audience, the format and the length, and whether you’ve told the tool what to do when it doesn’t know. Those five details do most of the work. For a look at the tools themselves, see our comparison of ChatGPT, Claude and Gemini.
Editor’s take
Strong AI prompts come from habits, not tricks. Treat the first answer as a draft. The people who get the most from these tools write one extra sentence of context, ask for a format and check the facts. That habit works the same no matter which chatbot you use, and it will keep working as the models change.
