Practical AI guideHow to Use Artificial Intelligence Effectively
What artificial intelligence actually is, how it works, and how to write requests that get useful results.
Before you startHow to make AI work for you
AI tools like Textie run on large language models, that are in public referred as Artificial Intelligence. These models predict the next word or sentence based on everything written in your request. They generate text that fits the pattern of what was asked - based only on what you put in. Whatever you do not tell the model, it will guess or skip. While some tools like Textie allows AI to consult the web for the latest information or to access databases and other tools, most of the time it will generate text based on the knowledge it has been trained on.
That means, that output quality follows input quality. A vague request gets a generic response. A specific request with context gets something usable. Most people who are disappointed with AI are writing requests that leave too much for the model to guess. When output is bad, the problem is almost always in the prompt.
AI works best on tasks where both the input and the expected output are clear: a recording turned into meeting notes, bullet points rewritten as an email, a document translated, a PDF condensed to five key points. These tasks go from 30 minutes to 3. For anything more open-ended like strategic decisions, predicting how someone will react, assessing context you have not described, your judgment needs to do most of the work.
Writing good AI requests is a skill that takes hours to pick up, not weeks. You do not need technical knowledge. The tips on this page come from real usage patterns. Pick one, try it on a task you are already doing, and adjust from there.
"Summarize this for me."
AI has to guess what to summarize, who it is for, how long it should be, and what format to use. The result will be generic and will need heavy editing.
"Summarize the key decisions from this sales call in 5 bullet points for our VP who wasn't there. Focus on what was agreed and who owns each next step."
AI now knows the source, the audience, the format, the length, and what to emphasize. The result will be usable with little or no editing.
- Source - what is being summarized
- Audience - who will read it
- Format - 5 bullet points
- Focus - decisions and owners
10 tips for better resultsHow to properly write request for AI
These apply to any task. Each one is something you can use right away.Give context before the task
Don't just say what you want, but also say who you are, what the output is for, and who will read it. "Summarize this meeting" gives AI almost nothing. "Write a 5-bullet summary of the key decisions from this sales call for our VP who wasn't there" gives AI everything it needs. The more context you provide, the less the model has to guess.
Specify the format explicitly
Without a format instruction, AI picks a structure it assumes fits. That is often wrong. You get a long paragraph when you wanted bullets, or a flat list when you needed sections with headings. Say exactly what you want: a numbered list, a table, three short paragraphs, a formal email, one sentence. Without this, you will spend time reformatting.
Set a length constraint
Without a length instruction, responses tend to run longer than you need. If you want brevity, say so: "in under 100 words", "in 3 bullet points", "in one sentence". If you need depth, say "cover each point in a short paragraph". Putting a limit forces the model to decide what matters.
Show an example
Showing an example is more effective than describing what you want. "Write it in this style: [paste your example]" gets you closer to the target than most other instructions. Even a rough example "something like this but shorter and more direct" narrows the output significantly.
Break complex tasks into steps
Asking AI to summarize, analyze, and draft a follow-up email in a single prompt gives you a thin answer to each. Use separate prompts in the same conversation. Handle one thing at a time, check the result, then move on.
Assign a role or perspective
"You are a senior editor reviewing this for clarity and concision" produces better edits than "make this better." A role gives the model a specific frame to work from. Use this when reviewing, critiquing, adjusting tone, or writing for a particular reader.
Refine, don't restart
When the first answer is 80% right, stay in the conversation and say what needs to change: "The tone is too formal. Keep the same structure, but written the way you'd explain it to a colleague." Starting a new conversation loses all context. Refining in place takes less time and produces better results.
Verify facts independently
AI produces text that sounds plausible. It does not verify what it writes. Numbers, dates, names, and specific claims can be wrong even when the rest of the response looks solid. For anything that matters, check the source.
Use follow-up messages
Each follow-up message can reference everything said before. "Shorten the third bullet" or "Make the conclusion more direct" takes seconds. Rewriting the whole prompt from scratch takes much longer. Use the conversation thread instead of starting over.
Read the output before you use it
AI does not know your relationships, your organisation's context, or what is sensitive in your situation. Reading the output takes 30 seconds. Catching a factual error or wrong tone before you send is always faster than fixing it after.
What to avoidTypical mistakes when writing requests for AI
These patterns show up in most bad AI requests. Fixing any one of them will improve your results.Being too vague
"Write me a summary" gives AI almost nothing to work with. What are you summarizing? Who is it for? How long? In what format? Each missing piece makes the result less useful. When output is bad, it is usually because the prompt left too much out.
Not specifying a format
Without format instructions, AI picks a structure it thinks fits. That is often wrong. You get prose when you wanted bullets, or a flat list when you needed headings. Specify the format, especially for anything you will paste or send directly.
Accepting the first answer
The first output is a draft. People who get good results from AI spend a minute refining: adjusting tone, cutting filler, strengthening one section. Using the first response unchanged almost always means settling for less than what the tool can produce.
Trusting AI with facts
AI produces text that sounds authoritative. Specific numbers, dates, legal requirements, and medical claims can be wrong. The confidence of the prose has nothing to do with whether the content is accurate. For anything that matters, check the source.
Putting everything in one prompt
One prompt that asks for four things "summarize this, find the problems, suggest improvements, draft a follow-up" will give you a thin answer to each. Break it up. One request per message, check the result, then move on.
Ignoring audience and tone
An email about a project delay to your manager, to a client, and to a teammate should read very differently. AI does not know which it is writing for. Tell it who the reader is and what your relationship is. Without that, the tone will be generic and you will need to rewrite it anyway.
Sending one request in pieces
"I'll send you something in a moment, don't do anything yet", then a second message, then a third. AI processes each message as it arrives and tries to respond. Splitting your request across multiple messages creates confusion, fills the context with partial instructions, and often produces output you did not ask for. Put everything in one message. If you are working with a document, attach the file directly instead of pasting fragments one by one.
Typing keywords instead of sentences
People used to search engines often type into AI the same way: "project delay email manager" or "meeting summary template". A search engine finds pages that match those words. AI needs to know what you want done with them. "Write a short email to my manager explaining a two-week project delay" gives the model an actual task. Keywords alone give it nothing to act on.
Frequently asked questionsLet's clarify some of your questions
Try it in Textie