How to use AI to be more productive: daily workflows that save time

You get more done with AI by giving it the repeatable, text-heavy parts of your day (first drafts, summaries, triage, research outlines) while you keep the judgment calls. This guide shows daily workflows for your inbox, meetings, planning, research and writing, with copy-paste prompts and the habits that decide whether AI saves time or wastes it.

How do you use AI to be more productive?

You use AI to be more productive by handing it the parts of your work that are repetitive, text-heavy and easy to check, and keeping the parts that need judgment, relationships or accountability. The best candidates are first drafts, summaries, reformatting, triage and the first pass of research. The worst are decisions, sensitive messages and anything you cannot verify quickly.

The evidence backs a measured view. In a study of 5,179 customer support agents, access to an AI assistant raised issues resolved per hour by 14% on average, with the largest gains for newer staff. In a 2025 randomized trial by METR, 16 experienced open-source developers took 19% longer on real tasks when using early-2025 AI tools, even though they believed afterward that AI had sped them up by about 20%. METR has since said those results are out of date for newer tools, but the lesson holds: perceived speed and measured speed differ, so measure your own.

A simple filter before you hand anything to AI:

Set up your AI assistant once so every request is faster

The biggest time sink with AI is retyping context: who you are, what tone you write in, what format you want. Save that once.

Most assistants have a place for standing context. ChatGPT has custom instructions, and Claude and ChatGPT both offer projects, where you attach reference files and instructions that apply to every chat inside. See the guide to Claude Projects for a worked setup.

A good standing instruction is short and specific:

I'm a product manager at a 40-person B2B software company. I write for engineers andcustomers. Style: plain US English, short sentences, no buzzwords, no exclamation marks.Default formats: bullet points for summaries, tables for comparisons, max 150 wordsunless I ask for more. If my request is ambiguous, ask one clarifying question first.When you're unsure of a fact, say so instead of guessing.

Then create a project per recurring area of work (weekly reporting, hiring, a client account) with the documents it needs: templates, past examples, style notes. A project for weekly reports with three past reports attached will produce a fourth in the right shape on the first try.

How to use AI for email and your inbox

AI helps most with email in three places: summarizing long threads, drafting routine replies, and turning a messy brain dump into a clear message. It helps least with deciding what matters, which is still your job.

Thread summaries

Paste a long thread (or use the summarize button many email clients now include) and ask for decisions and open questions, not a retelling:

Summarize this email thread in this format:- Decisions made (who decided, what)- Open questions and who owns each- What is being asked of me specifically, with deadlinesQuote exact dates and numbers from the thread. Do not add anything not in the thread.

Drafting replies

Give the facts and the outcome you want, and let AI handle phrasing. Compare:

The strong version takes 20 seconds longer to write and saves three rounds of edits.

Rules that keep email safe

• Read every AI draft before sending. Check names, dates and numbers line by line.
• Write apologies, bad news and negotiation emails yourself, then ask AI only to tighten them.
• Do not let AI reply automatically to people outside your company.

For a fuller setup, see AI for meeting notes and email.

How to use AI for meetings

Use AI before a meeting to prepare, during it to capture, and after it to turn talk into tasks. The after part saves the most time.

Before: paste the agenda and relevant docs and ask, "What are the three decisions this meeting needs to make, and what information is missing for each?" That question alone shortens many meetings.

During: transcription and summary features are built into most video call tools now. Turn them on only with everyone's knowledge; recording and transcription consent rules vary by country and company.

After: feed the transcript to your assistant with a strict format:

From this transcript, produce:1. Decisions (one line each)2. Action items as a table: task | owner | due date | quoted evidence from transcript3. Unresolved disagreementsIf an owner or date was not stated, write "not assigned" rather than guessing.

The "quoted evidence" column is the important part. It makes the model show where each action item came from, so you can spot invented ones quickly.

How to use AI for planning your day and week

AI is good at turning a pile of commitments into a draft plan, and bad at knowing your real priorities unless you tell it. Give it constraints along with the list.

A weekly planning routine that takes about 10 minutes:

1. Dump everything on your plate into one message: tasks, deadlines, meetings, half-formed worries.
2. Add your constraints: hours available, fixed meetings, energy patterns ("deep work only before noon").
3. Ask for a plan with explicit trade-offs.
4. Edit the plan, then move it into your calendar or task app yourself.

Here is everything on my plate this week and my calendar constraints.1. Group tasks into: must do this week, should do, can wait.2. Flag anything with a hidden dependency on someone else.3. Propose a schedule using 2-hour focus blocks before noon.4. Tell me what I should explicitly say no to or postpone, and why.

The last line is where the value is. A good plan says what will not happen. If you want an assistant that also reads your calendar and tasks directly, see setting up an AI personal assistant.

How to use AI for research without getting burned

Use AI to map a topic quickly, find what to read, and compare options, then verify any fact you will rely on. Models still state wrong things confidently, so research is where checking matters most.

A safe research workflow:

1. Map: "Give me the 8 questions an expert would ask about X, and the terms I should search for." This gives you a structure before you read anything.
2. Gather: use a search-enabled or deep research mode and ask for sources with links for every claim.
3. Verify: open the sources for the three to five facts your decision depends on. Check that the source actually says what the summary claims.
4. Synthesize: paste the verified material back and ask for a comparison table or a one-page brief.

Watch for these failure signs: citations that do not contain the quoted claim, numbers without a source, and summaries of documents that are more confident than the documents themselves. The guide to AI hallucinations covers how to spot and reduce them.

How to use AI for writing and documents

AI speeds up writing most at the two ends: getting from a blank page to a rough draft, and tightening a finished draft. The middle (deciding what you think and what to cut) is still yours.

Three patterns that work:

• Outline from notes. Paste your rough notes and ask for three alternative outlines with a one-line thesis each. Pick one and write.
• Reformat. Turn a long doc into a one-page summary, a slide outline, a FAQ or an email update. Reformatting is low risk because the content is already yours.
• Edit for clarity. Ask for edits as a list of suggestions with reasons, not a full rewrite, so your voice survives.

Edit this draft for clarity. Don't rewrite it. Give me a numbered list of:- sentences over 25 words, with a shorter version- claims that need a number or example- jargon a new hire would not understand- anything I say twice

If AI drafts often sound generic, read how to make AI writing sound human. Better prompts help too; the prompt engineering guide covers the techniques that still matter with current models.

Which habits decide whether AI saves time?

The habits matter more than the tool. People who get lasting value from AI tend to do the same few things.

• Ask for structure. Tables, checklists, ranked options and fixed word limits are faster to use than paragraphs.
• Give examples. One example of a good output beats three sentences describing it.
• Iterate in the same chat. "Shorter, and lead with the price" is faster than starting over.
• Keep a prompt library. Save prompts that worked in a note or snippet tool. Your best ten prompts will cover most of your week.
• Stop when it's slower. If you have rewritten the output twice, write it yourself.
• Check the risky parts. Names, numbers, dates, legal terms and quotes get checked every time.

There is a privacy habit too. Consumer and business plans handle data differently, and your employer may restrict which tools you can use. Before pasting client data, contracts or personal information, check the policy, or remove identifying details first.

What should you not hand off to AI?

Some work gets worse, or riskier, when AI does it, even if it looks faster. Keep these yourself and use AI only as a sounding board.

• Decisions with consequences. Hiring, firing, pricing, strategy and anything legal. AI can list options and risks; you choose and you own the outcome.
• Relationship messages. Condolences, apologies, feedback to a struggling colleague, a difficult client call. People can tell when a message was not written by the person who signed it.
• Work you need to learn. If you are new to a skill, doing it by hand a few times builds the judgment you will later need to check AI output.
• Facts you cannot check. If you have no way to verify a claim, a confident AI answer adds risk without saving time.
• Confidential material outside approved tools. Even a quick summary of a board deck in a personal account can break policy or contracts.

A useful test: if the outcome goes wrong, would "the AI wrote it" be an acceptable explanation to your manager or client? If not, the task stays with you, with AI helping at most on a draft you rewrite.

How to start today

Pick one workflow and run it for two weeks before adding another:

1. Write your standing instructions (role, style, default formats) and save them in your assistant.
2. Choose one recurring task from this guide: thread summaries, meeting action items or weekly planning.
3. Time it by hand once, then with AI for the next few times. Count checking time.
4. Save the prompt that works in a note you can copy from.
5. After two weeks, keep the workflow if it saves time, adjust it if it almost does, and drop it if it does not.
6. Add a second workflow only after the first one is routine.

If your work includes software, the same principle scales up: well-scoped tasks with clear checks are what you can hand to coding agents, and services like Orkio let those agents keep working on a cloud server after you close your laptop.

Frequently asked questions