# Building your personal AI workflow
Last Monday, a marketing manager named Priya opened ChatGPT and pasted the same prompt she pastes every Monday: "Summarize these five customer emails, flag anything urgent, and draft replies in my voice." Ninety seconds later she had her morning sorted. Her colleague, meanwhile, was still typing a fresh request from scratch, re-explaining his job to the AI for the hundredth time.
That is the whole game. The biggest gains from AI do not come from clever one-off chats. They come from saving your best prompts and running them again and again.
This lesson shows you how to build three things: a personal prompt library, custom instructions, and a repeatable weekly routine.
Most people treat AI like a search bar. They ask, get an answer, close the tab, and lose the prompt forever.
The problem: a *prompt* (the instruction you give the AI) is something you crafted. When it works well, it is an asset. Throwing it away is like writing a great email template and then deleting it.
A *workflow* is just a set of saved prompts and settings you reuse on a schedule. Build it once, benefit every week.
A prompt library is simply a place where you keep prompts that worked. It can be a Google Doc, a note in your phone, or a spreadsheet. Do not overthink the tool.
Save a prompt the moment it gives you a great result. Give each one a clear name and a category.
Here is a starter structure for a simple doc:
## EMAIL
- Triage inbox: "Summarize these emails, flag urgent ones, draft short replies in my voice."
- Polite decline: "Rewrite this to say no warmly but firmly. Keep it under 80 words."
## WRITING
- Tighten draft: "Cut this by 30%. Keep my tone. No new ideas."
- Headline options: "Give me 10 headlines for this post. Mix curious and direct styles."
## ANALYSIS
- Meeting notes: "Turn this transcript into decisions, action items, and owners."Two tips that make a library far more powerful:
Use placeholders. Write [PASTE TEXT HERE] or [TOPIC] so you can drop in new content fast.
Note what worked. Add a line like "works best when I give 3 examples first." Future-you will thank you.
For ready-made inspiration, the Anthropic Prompt Library is free and full of well-structured prompts you can adapt.
You no longer have to keep everything in a separate doc. In 2026, the major tools have native ways to store prompts:
Pick whichever tool you already use most. The concept is identical across all three.
*Custom instructions* are background notes the AI reads before every single chat. You write them once, and the AI applies them automatically. This is the single highest-leverage setting most people ignore.
In ChatGPT, find them under Settings, then Personalization, then Custom Instructions. Claude calls this "Personal preferences." Gemini uses "Saved info."
There are usually two boxes: what the AI should know about you, and how it should respond.
A strong example:
ABOUT ME:
I'm a project manager at a small software company.
I write a lot of status updates and client emails.
My audience is non-technical executives.
HOW TO RESPOND:
Be concise. Lead with the answer, then the reasoning.
Use plain language, no corporate jargon.
When I share a draft, suggest edits inline, don't rewrite the whole thing.
Ask me one clarifying question if my request is vague.Now you never have to re-explain your role or your style. Every chat starts halfway done.
Keep instructions short and specific. "Be professional" tells the AI nothing. "No exclamation marks, use British spelling, keep replies under 150 words" is something it can actually follow.
🎬 [VIDEO: "ChatGPT Custom Instructions: A Practical Setup" - youtube.com - a step-by-step walkthrough of writing custom instructions that change every future chat]
A single prompt is useful. A *chain* of prompts (running several in sequence, where each builds on the last) is where real time savings appear.
Think of it like a recipe. Each step takes the output of the previous step.
Example chain for writing a weekly client update:
1. "Here are my raw notes from the week. Group them into Wins, Risks, and Next Steps."
2. "Now write a 200-word client email from that summary. Calm, confident tone."
3. "Give me a one-line subject and a single follow-up question to end with."
You could do this in one giant prompt, but breaking it up lets you check the work at each stage and fix course early.
Let us build a full routine for a freelancer named Marco who juggles clients, content, and admin. Here is his actual AI week.
Marco pastes his task list into Claude with a saved prompt:
> "Here are my tasks this week. Sort by urgency and effort. Suggest a realistic order. Flag anything I should delegate or drop."
Five minutes, and he has a prioritized week.
He uses a Custom GPT he built for his newsletter. It already knows his topic, tone, and audience. He just types:
> "Draft this week's issue. Topic: how small teams use AI for hiring. Use my usual structure."
Because the instructions live inside the GPT, he skips all the setup.
Marco runs a two-step chain:
1. "Summarize what I completed this week from these notes."
2. "Turn it into three short client update emails, one per client, each tailored to their project."
Then a final admin prompt:
> "Draft my invoices summary: hours and tasks per client from this log."
What used to eat his Friday afternoon now takes twenty minutes.
The point is not the exact prompts. It is the *structure*: same days, same saved prompts, predictable output.
Knowledge check
1. According to the lesson, where do the biggest gains from using AI actually come from?
2. How does the lesson distinguish a 'prompt' from a 'workflow'?
3. Why does the lesson compare throwing away a good prompt to deleting a great email template?
4. Select ALL practices the lesson recommends for making a prompt library more powerful.
Select all the correct answers.
5. Select ALL statements that reflect the lesson's advice about building a personal prompt library.
Select all the correct answers.
If you find yourself running the same prompt on the same data constantly, you can connect AI directly to your tools so it runs without you typing at all.
No-code tools like Zapier or Make let you trigger an AI prompt when something happens, for example "when a new email arrives, summarize it and post the summary to Slack." You build it by clicking, not coding.
If you are slightly technical, calling the AI directly is only a few lines. Here is a tiny Python script that runs your "tidy these notes" prompt on demand:
from openai import OpenAI
client = OpenAI() # uses your API key from the environment
notes = """
- call with Acme, they want timeline by Friday
- fixed login bug
- need to invoice Beta Corp
"""
response = client.chat.completions.create(
model="gpt-4.1",
messages=[
{"role": "system", "content": "You organize messy notes into Wins, Risks, Next Steps."},
{"role": "user", "content": notes},
],
)
print(response.choices[0].message.content)The system message is just your custom instruction, written in code. Run this and it processes your notes instantly. You could schedule it to run every Friday and email you the result.
Do not feel you need to do this. Most people get 90% of the value from saved prompts and custom instructions alone. Automation is the cherry on top.
Hoarding without naming. A library of 50 unnamed prompts is a junk drawer. Name and categorize as you go.
Overstuffing custom instructions. If you list 30 rules, the AI follows few of them. Keep it to your top 5 or 6.
Never revising. Your best prompts will get better. When you tweak one and it improves, save the new version over the old.
Building for a routine you do not have. Start with one painful weekly task. Automate that. Add more later.
These actions are compiled in the role's Playbook.