Start Here: The 4 AI Features That Actually Save You Time

If you're new to AI and want time back, skip the overwhelm. You only need these four features to start. Set them up once and the tool starts working for you.

Takes 15 minutes for the first one. You need: ChatGPT or Claude, free or paid. You end with: every future chat starting from your context instead of a blank page.

1. Custom instructions: teach it how you like answers

This is the one that quietly improves every chat you will ever have. You write it once and it applies everywhere, so you stop repeating yourself.

In ChatGPT go to Settings, then Personalization, then Custom Instructions. You get two boxes. In Claude go to Settings, then Profile, and you get one box, so paste both blocks into it one after the other. Menu names move around between updates — if yours doesn't match, look for anything called personalization, profile, or preferences.

Box 1: what it should know about you. Fill in the brackets with your real answers. Two minutes here saves you months of generic responses.

I run [what you do] and most of my week goes to [the 2 or 3 things that eat the most time]. I am not technical, so skip the jargon or explain it in one line. When I come to you it is because I want time back, not because I want more to do. Assume I am busy and slightly behind.

Box 2: how it should answer you. This is the part most people skip, and it is where the difference shows up.

Here is how I want you to work with me.

Ask me questions before you answer if you are missing context. Do not guess and hand me something generic.

Tell me the truth over what I want to hear. If my plan has a hole in it, say that first, then help me fix it.

Keep answers short. Lead with the answer. Save the reasoning for when I ask.

Give me steps I can act on today, not the theory behind them.

When something is likely to go wrong, name it and tell me what it would cost me.

If you are not sure, say you are not sure, and tell me what would settle it.

Write in plain language. Short sentences. Active voice. Talk to me as you, not as one.

No hype words, no filler openings, no em dashes.

One optional add-on. If you use AI for decisions rather than tasks, add this to Box 2. It turns a helpful assistant into something closer to a second opinion.

When I bring you a decision, argue the strongest case against it before you agree with any of it. Ask me what I am assuming that could be wrong. Tell me who this helps and who it costs.

Both boxes cap out around 1,500 characters, so if you run out of room, cut from Box 1 first. How it answers matters more than what it knows about you.

What the difference actually looks like. Same question, "should I send a newsletter this month?":

BEFORE  "Great question! Newsletters can be a powerful tool
        for engagement. Here are 7 benefits of email
        marketing... [five paragraphs]"

AFTER   "Probably not this month. You told me you're behind
        on follow-ups, and those make you money now.
        What's the newsletter for - staying visible, or
        selling something specific?"

That's the whole point. It stopped performing helpfulness and started asking the question that mattered.

2. Upload files: stop retyping

Drag in a document, PDF, or screenshot and ask about it. This is the fastest win on the list because there's nothing to set up.

The three that earn their keep every week:

"Summarize the 5 things that matter and skip the rest."
"Pull out every date, deadline and dollar amount, as a list."
"What's my action item here, and what happens if I ignore it?"

Screenshots count, which is the part people miss. A photo of a paper invoice, a confusing error message, a competitor's pricing page — all readable. If you can see it, you can usually ask about it.

Two limits worth knowing: it can misread messy handwriting and dense tables, so spot-check anything numeric. And don't upload something you wouldn't email — the rules on training and retention differ by tool and plan, and they change.

3. Saved prompts: write it once, reuse forever

Anything you ask more than twice, save it. Weekly planning, email drafts, meeting summaries — perfect it once, reuse it forever.

Where to put it, best first:

A project or custom GPT, if your tool has them. The context tags along automatically, so you don't paste it every time.
A note titled with the task, not with the word "prompt." You'll search for "follow-up email," never for "prompt 3."
Your phone's text replacement, for the two or three you use daily. Type ;;follow and the whole prompt appears.

The habit that makes this stick: when an answer comes back genuinely good, stop and save the prompt right then, with one line about why it worked. Thirty seconds. Nobody remembers a month later what they changed.

4. Connect it to your tools

Look for "Connectors," "Apps," or "Integrations" and link your calendar, email, or docs. This is the jump from AI that talks about the task to AI that does it.

Start with the calendar. It's the lowest risk and the most immediately useful:

Look at my calendar for the last two weeks. Total my hours by type of meeting, then tell me which recurring blocks have the weakest case for existing at their current length. Ask me what I spend time on that never makes it onto the calendar before you conclude anything.

Before you connect anything: your calendar and inbox hold client names, interview slots and personal appointments. Say what's off limits up front — "refer to people by role, not name, and skip anything marked private." Connect read-only where you're given the choice, and connect one thing at a time so you can tell what changed. Availability varies by plan and shifts often; if you don't see the option, you're not doing it wrong.

Your first 15 minutes

Do #1. Only #1. It's the one that quietly improves every chat afterward, and the other three are useless if the answers are still generic.

Then use it normally for a week before you touch #4. You'll write much better instructions on day seven than you can today, because you'll have watched it get things wrong.

When it goes sideways

"It's ignoring my instructions." Usually one of three things: you're in a chat that started before you saved them, the instructions contradict each other (short answers and thorough explanations), or you asked for something that needs length. Start a fresh chat and ask: "What instructions are you currently working from? List them back to me."

"It got blunt and now it's a bit much." The honesty line is doing more work than you wanted. Soften it: "Tell me the truth, but lead with what's working before what isn't."

"I ran out of characters." Cut Box 1 to one sentence about what you do and one about what eats your week. Box 2 is what changes the answers.

"Answers are still generic." Your Box 1 is probably too abstract. "I run a small business" tells it nothing. "I run a Pure Barre studio with 6 instructors and 200 members" changes everything, because now it knows about schedules, retention and instructor cover without being told.

Make it yours

Client service: add to Box 1 that you're writing to clients most days, and that your voice matters more than polish.
Studio, shop, scheduled business: name the rhythm — weekly schedule, monthly billing, seasonal swings. It'll stop suggesting things that don't fit your calendar.
Solo: add "I have no team, so don't tell me to delegate to anyone. If it needs another person, say so plainly."
Small team: add who's on it and who does what, so it stops proposing work you'd have to do yourself.

Do this next

Set Box 1 and Box 2 now, then close this page and use it for something real today.

Next guide: 5 Hours Back: The One-Task Method — pick the task you dread and hand it over properly, now that the answers are finally aimed at your business.