If you already have a ChatGPT account you barely use, start there — not with a new tool. Spend your first two days turning off the settings you don't want on, then write one page that explains your business to the AI and save it somewhere it will always be read. That single page, not the tool you choose, is what turns generic answers into answers you can actually send.
Most people's real history with AI goes like this. You signed up, asked it something about your business, got back an answer that was technically fine and completely generic, felt a small flat disappointment, and closed the tab. Maybe you're still paying twenty dollars a month for it. There's a low-grade guilt about that subscription. This guide is written for exactly that spot, and the fix is smaller and less technical than you'd expect.
Start With the Bottleneck, Not a Tool List
There are a hundred articles ranking the best AI tools for business owners, and they're all answering a question that isn't your problem. Your problem was never that you picked the wrong assistant. Your problem is that the assistant doesn't know anything about you — not your prices, not your services, not the way you write to clients, not the fact that you never take Friday afternoon calls. It's answering as if you were the average business on the internet, because that's the only thing it has to go on.
That's a writing problem, not a technology problem. The gap between a useless AI answer and a genuinely useful one is almost entirely the context you supply. And context is something you can write down once and reuse forever, which makes it the highest-return hour you'll spend on any of this.
So the order in this guide is deliberate. Days one and two are configuration and writing, with almost no prompting. Days three through five connect the assistant to one real, repetitive task so it stops being a chat window you have to remember to visit. Week two gives your business a memory. By the end you'll have built one concrete thing and configured one workspace that knows your operation — and every prompt you type after that is better because of work you already did.
Week 1, Days 1–2: Set Up ChatGPT or Claude — Before You Type a Single Prompt
Which one, and which plan
Pick ChatGPT or Claude. Not both. The honest truth is that at your stage the two are close enough that the choice will not determine your outcome, and spending a week comparing them is a way of not starting. As of now, ChatGPT connects to more other software, which matters slightly later. Claude tends to be steadier when you hand it a long document and ask it to stick to what's in there, which matters for the context file you're about to write. Either is a defensible choice.
Start on the free tier. It's genuinely enough for week one, and it's the honest way to find out whether you'll actually use this. You'll feel the ceiling when you hit message limits mid-task, or when you want a saved workspace that always carries your business context — that's your signal to move to the roughly twenty-dollar personal plan. The team and business plans (ChatGPT Team, ChatGPT Business, or Claude's team plans) only start earning their price when several people need to share the same setup and you want the company, rather than individuals, to control the accounts. Don't buy that on day one.
Turn on your privacy settings before you use it for work
Do this before you paste anything from your business into a chat. It takes five minutes and most guides skip it entirely.
By default, some plans use your conversations to help train future versions of the model. That's the setting you want off. In ChatGPT, open Settings → Data controls and turn off the option about improving the model for everyone. In Claude, open Settings → Privacy and check the equivalent toggle for using your chats to improve the model. Business and team plans generally exclude your data from training by default, but check anyway rather than assuming.
Plain-English version of the risk: until you've confirmed those settings, treat the chat box like a public forum. Don't paste customer lists, social security or tax ID numbers, medical details, card numbers, or a signed contract with someone's home address in it. Even after you've turned training off, a good permanent habit is replacing real names with initials whenever the AI doesn't need the name to do the job. "Draft a follow-up to J.M., who asked about pricing for the twelve-week program" works exactly as well as using her full name.
Set up persistent context — this is the part that matters
Now the core of the whole two weeks. You're going to create a saved workspace and give it a document about your business, so you never explain yourself from scratch again.
In Claude, click Projects → Create project, name it after your business, and use the project instructions field plus the project knowledge area to hold your context. In ChatGPT, either create a Project and set its instructions, or build a Custom GPT (Explore GPTs → Create) and fill in the instructions and the knowledge upload. Functionally they're the same idea: a room you walk into where the AI already knows who you are.
Then write the one page. Open a blank document, answer these six prompts in plain sentences, and keep the whole thing to a single page:
- What do you sell, and to whom, in one sentence? Be specific about the buyer, not the market.
- What are your actual services or products, with real prices or price ranges? Include what's included and what isn't.
- How do you sound? Two or three sentences on your voice, plus a rule list — words you use, words you never use, whether you're formal with clients or first-name casual.
- What does a typical job or client look like from first contact to finished? The steps, in order, even if it's messy.
- What questions do people ask you constantly, and what's your standard answer? Five is plenty to start.
- What does "good" look like? What makes an email, a proposal, or a post something you'd actually send — and what makes you reject a draft.
Save it, then attach it to the project or paste it into the instructions. That's your deliverable for day two. It will not be perfect, and that's fine — you'll add a line to it every time the AI gets something wrong, and within a month it's the most valuable document in your business.
One warning worth hearing now: it will still get things wrong. Read the first ten outputs closely before you trust anything unread. You're not looking for perfection, you're looking for the point where the drafts stop being generic and start sounding like you.
We learned this the slow way. The first version of this at sagulabs was an ugly document titled something like "here's how our business actually works," and we pasted it at the top of every single conversation because we didn't know there was a better way. The day we stopped re-explaining ourselves to the AI was the day the tools became genuinely useful. Not a better model. A saved page of context.
Week 1, Days 3–5: Connect One Automation So the AI Isn't Just a Chat Window
A chat window only helps when you remember to open it. The next step is putting the AI inside one thing that happens whether you think about it or not. Zapier is the easiest place to do this, and its free plan covers a first automation.
Pick one task that is small, repetitive, and slightly embarrassing in how long it takes you. A reliable first choice: a new form submission comes in, and a personalized follow-up email gets drafted and waiting for you. If you want help choosing a different one, we wrote a longer piece on which workflow is worth automating first — but if you're unsure, use the form-to-follow-up example and move on.
The literal steps:
- Create a Zapier account and click Create → Zaps.
- Choose your trigger app — the thing that starts it. Search for your form tool (Google Forms, Typeform, Squarespace, HubSpot, whatever you actually use), pick the event "New form submission," and connect the account. Zapier will ask you to load a recent submission so it has real data to work with. Do that; it makes everything after this easier.
- Add an action step and choose ChatGPT or Claude from Zapier's app list. Pick the "Conversation" or "Send prompt" action.
- Write the prompt, and paste your context page into it. This is the step people skip, and it's why their automations produce bland email. The prompt should read something like: "Here is context about my business: [paste your one-pager]. A new inquiry came in with these details: [insert the form fields from step 2]. Draft a short, warm follow-up email in my voice that answers their question and proposes one clear next step."
- Add a second action: your email tool, with the action "Create draft" — not "Send email." Map the AI's output into the body. Drafts, not sends, for at least the first two weeks.
- Click Test step on each action, then turn the Zap on. Submit your own form once as a fake customer and watch what lands in your drafts folder.
Read the first ten drafts before you send them. You'll catch two or three phrasings you'd never use, and each one goes back into your context page as a rule. That loop — output, correction, context updated — is the entire skill.
Week 2: Add the Tool That Remembers What You'd Otherwise Forget
By now you'll notice something annoying: you keep re-typing the same background information into different places. The fix is a notes tool with AI built in, where your recurring business knowledge lives in one searchable place. Notion AI is the common choice; Google Docs with Gemini or Microsoft's Copilot in OneNote work the same way if you're already living in those.
Keep this simple. Create three pages and nothing else. One page for your frequently asked questions and your standard answers. One page for your how-we-do-it steps — onboarding a client, closing out a job, sending an invoice — written the way you'd explain it to a new hire. One page per recurring client with the details you always have to look up.
The point isn't tidiness. It's that when someone asks a question you've answered forty times, the answer already exists in writing, the AI can find it, and neither of you has to reconstruct it from memory. It also quietly becomes the raw material for anything you build later, because a business with its knowledge written down is a business that can hand work to someone else.
The Prompts to Actually Type First
Once your context file is loaded into your project, these work as-is. Paste them, fill in the brackets, and read the result critically.
A client follow-up in your voice:
Using the business context in this project, draft a follow-up email to
[first name or initials], who [what they asked about / where the
conversation left off] on [date]. Keep it under 120 words, match my
voice exactly, don't discount anything, and end with one specific next
step and a proposed time.
Messy meeting notes into action items:
Here are my raw notes from a call. Turn them into three lists: decisions
made, action items with an owner and a due date, and open questions that
need an answer before we can move. If an owner isn't clear from the
notes, mark it UNASSIGNED instead of guessing.
Notes:
[paste your notes exactly as you typed them, typos and all]
A week of social captions from three bullets:
Using my business context and voice rules, write five short social
captions based on these three points: [bullet one], [bullet two],
[bullet three]. One caption per day, Monday to Friday. Vary the opening
line so they don't read like a series. No hashtags, no em dashes, no
exclamation marks.
A complaint turned into a calm reply:
A client sent this message and I'm annoyed. Draft a reply that stays
calm and professional, acknowledges the specific thing they're upset
about without accepting blame we don't own, and offers one concrete
resolution consistent with my policies in the context file. Then, below
the draft, tell me anything in their message I should verify before I
send it.
Their message:
[paste]
That last instruction — "tell me what to verify" — is worth adding to almost every prompt you write. It's how you keep a fast draft from becoming a confident mistake.
When These Tools Stop Being Enough
This setup has a ceiling, and it's useful to recognize it rather than assume you're doing it wrong. Three things you'll notice:
You're copying and pasting between the assistant and your real systems every day. Every draft gets moved by hand into your inbox, your CRM, your scheduler. That's a sign the AI is sitting next to your workflow instead of inside it, and the copying is now the bottleneck rather than the thinking.
You built something that works for you, but nobody else on the team can get the same result. Your prompts depend on your judgment about what to fix in each output. That's a personal habit, not a tool, and habits don't transfer. Making it work for other people means turning it into something with the rules built in.
The tools help you internally, but your website converts visitors exactly as it did before. That's a different problem than more internal AI, and it needs a different fix.
If any of those describe you, the useful next step is a look at your operation rather than another subscription — which is what an operation audit in our AI consulting engagements is for: mapping where the work actually slows down before anyone decides what to build.
And if you want the thinking behind that distinction, the difference between AI as decoration and AI as infrastructure is the natural next read now that your tools are running.
Common Questions About Getting Started With AI
Which AI tool should I start with?
Either ChatGPT or Claude. At day one the difference between them is far smaller than the difference between a version that knows your business and one that doesn't, so pick one, open an account, and stop comparing. If you want a tiebreaker: ChatGPT has more third-party integrations, and Claude tends to hold a long document of context more faithfully. Both are fine for everything in this guide.
How long does it actually take to set this up?
About four hours of real work spread across two weeks. Roughly 30 minutes for the account and privacy settings, 60 to 90 minutes to write your one-page business context file, another 60 minutes to build your first automation, and an hour in week two to move your recurring answers into a notes tool. The writing is the slow part, and it's also the part that does the work.
Is it safe to use AI tools with client or customer information?
It's safe once you've configured it, and risky before that. Turn off the setting that lets the company use your conversations to train their models, and until you've done that, don't paste anything you wouldn't email to a stranger — no social security numbers, no medical details, no card numbers, no full client lists. A good habit even afterward is to swap real names for initials when the AI doesn't need the name to do the job.
Do I need a paid plan to get started?
No. The free tier of ChatGPT or Claude is genuinely enough for your first week, and it's the right way to find out whether you'll use this at all. You'll know you've outgrown it when you start hitting message limits in the middle of a task, or when you want a saved workspace that always carries your business context. Paid team or business plans only start to matter when more than one person needs the same shared setup.
What's the single most important thing to do before I start typing prompts?
Write a one-page file describing your business: what you sell, who buys it, how you price, how you talk, and what a good result looks like. Then attach or paste it into a saved workspace so it's present in every conversation. This one document is the difference between generic advice and advice that fits your operation, and you only have to write it once.
How do I know when I've outgrown these starter tools?
Three signs. You're copying and pasting between the AI and your real business systems every single day. You've built something that works for you but nobody else on your team can get the same result. Or the AI is helping you internally while your customer-facing numbers haven't moved at all. Any one of those means the next step is changing how the work flows, not adding another subscription.
Two Weeks From Now
At the end of this you won't have an impressive AI stack. You'll have one workspace that knows your business, one automation that runs without you, and one page of writing that makes every future prompt better. That's a much better position than a folder full of trial accounts, and it's the base everything else gets built on.
At sagulabs, we started in exactly this spot — business owners with a messy context doc, figuring out where the AI stopped helping. When you hit that ceiling, an audit of your operation is the honest next move: finding where the time actually disappears before deciding what to build.