Free tips
Practical ways to use ChatGPT, Claude, Gemini, or Microsoft Copilot in your own work. Pick a tip and try it. Copy a prompt when one is provided, then answer the AI’s questions with your own context. Some tasks need files, a transcript, or a connected account; features vary by app and plan.
1. Stop typing. Start talking.
Use your voice to explain what you need. Speak naturally and include the details that matter.
Speaking can help you explain what you need and include details you might leave out when typing.
Help plan a practical AI workshop for our team. Include a short demo, a hands-on exercise, and clear next steps for everyone.
Okay, um, help me plan a practical AI workshop for our team. We have thirty minutes, and most people are beginners. I’d like a short demo, then a hands-on exercise using a real task, you know, something they do every week. Keep it simple and avoid jargon. We haven’t picked the task yet, so flag that as missing. Finish with a short checklist they can use afterward. Don’t invent details that I haven’t given you.
You can include more context in the same amount of time.
Use the dictation microphone in your AI app’s message bar.
Talk through the problem. Let AI organize your thoughts.
Use a voice conversation model to talk through a real problem in a live back-and-forth. Explain it in whatever order it comes to mind, and let the AI ask useful follow-up questions.
When you’re finished, ask the AI to organize only what you said, mark assumptions and missing details, and suggest a practical plan.
2. Give your prompt a job to do
A prompt is an instruction you give the AI. Be specific about the task and the output you want: a decision with reasons, a draft email, a plan, or three useful observations. Add the context it needs to do the job.
Try this in an AI conversation that already knows something about you. The task is to give you candid feedback. The output is three points, each with an observation to support it and an action you can take.
Three things you might need to hear
Use an ongoing conversation or project with some context about you.
Based on everything you know about me, tell me 3 things you think I need to hear but don't want to hear. Be direct and ground your answer in what I've actually shared. For each point, give me the thing I need to hear, the specific observation that supports it, and one practical action I can take. Don't invent motives or personality traits. If you don't have enough context, ask me up to three questions first.
3. Give it context and keep that context current
Let the AI interview you about your work. You can answer out loud. Start with your role, then work through what you actually do, how you do it, and what you want to achieve: the people you work with, the decisions you make, the tools you use, and the work that keeps coming back.
It should follow up on your answers, use relevant files or connected accounts where available, and build a working profile you can check. Save that profile somewhere the app can actually use again, such as project instructions or a reusable file, and keep developing it as you work together.
Build a working understanding of my work
Start this in a project or ongoing conversation. No preparation needed; answer the questions by voice or text.
Interview me so you can understand my work and help me do it better. Start with my role and responsibilities. Ask one or two questions at a time, choosing follow-ups from my answers. Learn what I do, how I do it, who I work with, the tools and information I use, what good work looks like, and what I want to achieve. Ask for a real example when it would help. Use relevant context and connected sources already available to you. Build a concise working profile with confirmed facts, priorities, preferences, and open questions; let me correct it. Suggest a few practical ways you could help. Help me save the approved profile using this app's supported instructions, project files, or another reusable file. Explain where it was saved, what will be available next time, and how to keep it current.
4. A good example is worth a thousand words
Show the AI what you mean. Attach a document, image, spreadsheet, or other example you like, and explain what you want it to learn from it. That could be the layout, the tone, the level of detail, or a particular visual style.
A reference gives it something concrete to work with. You could even start with a photo of yourself and ask for a polished profile picture for your work email. The photo provides your likeness; the prompt explains the result you want.
Make a profile picture for your work email
Attach a clear photo of yourself in an AI app that can edit images.
Turn this photo of me into a professional profile picture for my work email. Keep my face, natural expression, and distinctive features recognizable. Use soft, natural-looking light, a simple background, and a head-and-shoulders crop that works in a small circle. Give me a clean, high-resolution square image.
5. Let AI learn your voice from your email
Your sent email already contains years of examples. Connect the account you want the AI to learn from, with read-only access where available. Ask it to study a substantial range of your writing: different people, subjects, situations, and levels of formality.
The useful result is a reusable set of instructions for drafting and reviewing your inbox. Some apps call this a skill. It should capture how your voice changes with the person and situation, then improve with your corrections. Your history can guide a recommendation; it does not give the AI permission to send messages or make decisions for you.
Build an email skill from my sent mail
Use an AI app that can search your connected Sent mail, or attach an authorized Sent-mail export. You do not need to select and paste individual examples.
Build a reusable email skill from my own writing. First verify which connected account and Sent-mail history you can read. Use read-only access. Review a substantial, varied sample from the last six months, aiming for at least 100 messages if available. Select across recipients, dates, topics, languages, and situations, rather than simply taking the latest messages. Work in batches and tell me the date range, number reviewed, variety covered, and any access limits. Analyze only text I authored; exclude quoted replies, forwarded material, signatures, automated messages, and copied templates. Identify recurring patterns in greetings, sentence length, vocabulary, warmth, directness, requests, explanations, disagreement, follow-ups, and sign-offs. Explain how they change for different relationships and situations. Separate repeated patterns from tentative observations, with brief supporting examples. Create a portable skill file called email-working-guide.md containing when to use it, general style rules, context-specific variations, exceptions, a drafting checklist, and suggested inbox-review criteria grounded in my history. Treat inferred priorities as suggestions for me to confirm. Do not turn past messages into permission to send, archive, label, make commitments, or decide for me. Keep unnecessary personal details and copied email content out of the file. Show me the guide and test it on three varied threads, producing draft replies and suggested next actions for my review. Ask for missing facts or decisions. Use my corrections to improve the file. Provide it as a reusable attachment; if this app supports skills, help me install it after I approve and verify that it works. Do not claim it is saved, installed, or active unless you can verify that.
Use the skill on a real reply
Open the relevant thread, or attach it, with your approved email skill available.
Use my approved email skill to draft a reply to this thread. Read the full relevant exchange and choose the style that fits this person and situation. Briefly state what needs an answer and which facts or decisions are missing. Ask me only for those missing details, then draft the reply for review. Do not invent my position, make a new commitment, or send the message. After my edits, suggest any useful correction to the skill.
6. Let AI keep track of what you need to do
Let the AI do the work of keeping track: find commitments in the sources it can access, keep tasks up to date, notice what is waiting on someone, and bring you the next useful action. You should not have to keep checking a list just to make sure it gets maintained.
Keep the list in a task app or shared file you control, so the AI assistants you use can work from the same place when their connections support it. I use Todoist. Check that each AI app can access your chosen list, then agree on the routine and permissions for keeping it current.
Take task tracking off my plate
Start with your existing task app and connected work sources. No need to retype your task list.
Set up a task-tracking system you can maintain for me. Start with the task app and work sources you can already access. Keep one list in a place I control that my other AI apps can also access, where supported. If we need a new place, suggest the simplest option. Find current commitments, deadlines, and things I'm waiting on; check later replies and existing tasks so you don't add duplicates or completed work. Keep each task's next action, status, source, and any stated owner or deadline. Ask only for missing access, important uncertainties, or decisions. Agree with me on which sources to check, a maintenance routine, and permissions for future updates. Use the list to suggest priorities and help me move work forward. Tell me what is set up and which ongoing checks are actually enabled. Bring me useful changes and decisions, rather than making me review the whole list every time.
7. Give AI more than text
A multimodal AI can work with different kinds of input, such as images, audio or video. Supported formats depend on the model and app. Try a photo of a whiteboard to organize its ideas, or a short video of a process to ask what steps it shows. Check the result against the original.
A recording is another example. When your workplace permits it and everyone agrees, record a conversation or meeting so you can stay involved without trying to write everything down. A client call, a project catch-up, or a handover can all become useful working context.
Attach the audio directly to an AI app that supports it. Ask for notes, actions, and a follow-up draft from the same recording. You do not need to summarize the conversation first.
Turn this recording into notes, actions, and a follow-up
Attach a conversation or meeting recording. If your app does not accept audio, use an exported transcript.
Turn this conversation or meeting recording into three things: concise notes, an action list, and a follow-up email draft. Capture the important context, agreed decisions, and unresolved questions. For each action, include what needs doing, the agreed owner, and the stated deadline; mark anything that was not agreed or is unclear. Keep suggestions separate from commitments. Use timestamps where available so important points are easy to find. Write a short follow-up summarizing the decisions and actions, with questions for anything unresolved. Return all three together. Flag any part of the recording you could not process, and leave the email as a draft.
8. Give feedback. Remember to iterate.
Tell the AI what you would keep and what you would change, and explain why. For a monthly report, give it approved data and let it produce a draft. Then compare that draft with a good human-written report based on the same data. What is different, and why does it matter?
Use those differences to improve the instructions and revise the draft. Keep the instructions and approved example. Next month, repeat with fresh data and review the result again. You’re improving the workflow through feedback; each report still needs human review.
9. Use more than one AI
For important questions, get a second opinion: ask two AI apps from different labs. Give each the same brief before sharing either answer. Then ask: “Where do these answers disagree? Which assumptions need checking? Find sources that could prove either answer wrong.” Check those sources yourself. Agreement is useful, but both can repeat the same mistake. Share only information you’re allowed to upload.
Which AI should you try?
All four can help with writing, reasoning and code. These are useful starting points, not a ranking:
ChatGPT · OpenAI. Try mixed work with documents, data and images. The app offers web search, voice and image creation/editing. Supported video uploads can be analyzed with tools, with limitations.
Claude · Anthropic. Try long documents, drafting and coding. It understands images and offers web search and voice. It makes diagrams with code, but has no native photo or video generator.
Gemini · Google. Try work combining documents, images, audio and video. Gemini Live supports spoken conversation; Nano Banana creates images and Gemini Omni creates videos.
Grok · xAI. Try questions involving current web and X discussions. It handles documents and images, offers voice, and creates images/videos through Grok Imagine.
Training data, post-training and tools differ; complete dataset recipes aren’t public. Search access is separate from training. Test the same real task rather than assuming one brand is best. Features vary by plan, region and app.
Capabilities checked 2 October 2026 · Sources
- ChatGPT model availability
- ChatGPT image and video inputs
- Images in ChatGPT
- ChatGPT Voice
- Searching the web with ChatGPT
- OpenAI model development and data
- Sora discontinuation
- OpenAI API deprecations
- Claude models overview
- Can Claude produce images?
- Claude voice mode
- Claude web search
- Anthropic Transparency Hub
- Gemini 3.8 Flash model card
- Gemini file inputs
- Gemini image generation
- Gemini video generation
- Gemini Live
- Gemini Deep Research
- Gemini 3 Pro family model card
- Grok 4.7 model specifications
- Grok model catalogue and separate media models
- Grok consumer capabilities
- Grok app overview
- Grok X Search and video understanding
- Grok Web Search
- Grok 4.20 system card: disclosed training categories (historical example)
Why different AIs can give different answers
AI companies have different routes to information. Training teaches a model patterns before you ask; search and connected tools fetch material while it answers. Owning a platform or licensing its content creates access, but does not reveal exactly what went into each model.
ChatGPT · OpenAI
OpenAI has commercial Reddit access, including a real-time Data API. Its publisher partners include the Financial Times and News Corp, whose titles include The Wall Street Journal and the UK’s The Times. Agreements with Axel Springer, Le Monde and Prisa explicitly include model training. Shutterstock licenses images, video, music and metadata for training. The New York Times is a different case: its confirmed AI licensing deal is with Amazon; its lawsuit against OpenAI was still active in September 2026.
Gemini · Google
Google also has a Reddit Data API agreement, explicitly covering training and content display. YouTube confirms that uploaded content supports AI development across Google and YouTube under creator terms. Gemini’s disclosed training mix includes video, but this does not establish that every YouTube video trained every Gemini model. Separately, Gemini can find public YouTube videos and answer questions about them.
Grok · xAI
Grok can search public X posts in real time. X also permits public posts, associated metadata, public Spaces and profiles to support training, subject to controls. Grok 4.7’s disclosure confirms some X video training and also lists public-web, commercial, private and synthetic data. “All of X” overstates this: protected posts are excluded, and users can opt out of training.
Claude · Anthropic
Anthropic describes public-web data, datasets acquired under commercial agreements, user and crowd-worker contributions, and synthetic data. It does not publish a complete supplier-by-supplier recipe. Claude can also search the web and retrieve scholarly material through Wiley’s Scholar Gateway, subject to access conditions. That connector is evidence of research access, not proof that Wiley’s whole collection trained Claude.
These are concrete differences in access, not a ranking of intelligence. Sources overlap: both Google and OpenAI license Reddit. Data selection, filtering, post-training, tools and update dates also shape the answer. For an important question, ask two labs independently, compare their evidence, and open the sources. Agreement alone is not verification.
Checked 2 October 2026. Public announcements and provider disclosures describe access and intended uses; they rarely reveal complete contracts, payment amounts or every item used in each model. No unpublished deal values are assumed.
Sources
- Reddit and OpenAI partnership
- Reddit complaint against Anthropic
- Financial Times content partnership
- A landmark multi-year global partnership with News Corp
- Partnership with Axel Springer
- Global news partnerships with Le Monde and Prisa Media
- Shutterstock expands OpenAI partnership with six-year training-data agreement
- The New York Times Company and Amazon announce licensing agreement
- Defendants memorandum supporting summary judgment in NYT litigation
- An expanded partnership with Reddit
- New tools to protect creators and artists
- Public summary of training content for the Gemini 3 Pro family
- Find and ask about YouTube content in Gemini Apps
- About Grok
- Public summary of training content for Grok 4.7
- How Anthropic uses personal data in model training
- Anthropic Transparency Hub
- Claude can now search the web
- Wiley launches interoperable AI Gateway
- Scholar Gateway training hub
- Claude for Life Sciences
- How ChatGPT and OpenAI foundation models are developed
10. Get meta
Ask the AI to examine how you use AI. Give it access to relevant conversations, files, and tools, and ask where you could get more value: tasks you could hand over, tools you are missing, better ways to give context, or work you could turn into a repeatable process.
The useful result is a practical report grounded in the work it can actually see. It should explain what to improve, why it matters, and how to start. A few well-chosen questions can fill the gaps.
Find the value I am missing with AI
Start in an ongoing conversation or project with relevant work available. The AI can only assess the history and sources it can access.
Where am I leaving the most value on the table with AI? Audit how I use it based on the conversations, files, and connected tools you can actually access. Briefly state what you reviewed. Look at the tasks I hand over, how I give context and feedback, how I use files and examples, and which tools or repeatable workflows could help. Identify what is working, missed opportunities, and habits that limit the results. Ground your findings in specific examples. Produce a practical report, ranking improvements by likely impact and effort. For each recommendation, explain the benefit, show how the workflow or interaction would change, and give me a clear first step. Check the current support for any app or feature you recommend. Finish with the three changes worth trying first. Ask up to three focused questions if missing facts would materially change the report; otherwise get started.
Start with a free 30-minute call
Tell me what you spend your week on, and I’ll tell you which tasks I’d start with.
Sources behind these tips · Continue learning
OpenAI Academy · Claude Academy · Grow with Google AI training
