Most people “use AI for their business” by opening a chat, typing a vague question, and getting a vague answer back. The tool isn’t the problem — the questions are. An AI assistant is only as useful as what you ask it, and a few habits separate people who get real leverage from people who get a shrug. Here’s how to actually put one to work.
Give it a role and context, not just a question
“Write me a marketing plan” gets you a generic template. “You’re a marketing strategist for a one-person course business making $4k/month from an email list of 900 — what are the three highest-leverage things to do this month?” gets you something usable. The pattern: role + real numbers + a specific decision. Vague in, vague out.
If you want the technique straight from the model makers, both publish free guides: Anthropic’s prompt engineering overview for Claude and OpenAI’s prompt engineering guide. They’re written for developers, but the core advice (be clear and specific, give context and examples) applies to anyone typing into a chat.
The four jobs AI is genuinely good at for a solo business
- Thinking partner. Pressure-test a decision. “Here’s my pricing and my reasoning — argue the other side.” It’s a tireless devil’s advocate.
- First-draft engine. Emails, outlines, job posts, FAQs. Never ship the first draft, but a first draft in ten seconds beats a blank page in ten minutes.
- Translator of your own data. Paste a messy export and ask “what changed, and what should I worry about?” It’s fast at turning numbers into a plain-English read.
- Explainer. “Explain this contract clause / tax term / metric like I run a small business, not like I’m a lawyer.” Great for the things you’re expected to just know.
The questions to ask (and the data each one needs)
Every good business question has two halves: the question, and the numbers that make it answerable. Here are ten worth asking, with what you need to give the assistant for each. Without the data column, you get the generic answer.
| Ask this | Data it needs | Where that data usually lives |
|---|---|---|
| “Which product or service brought in the most revenue this month versus last month, and what changed?” | Revenue by product or service for both months | Stripe, Shopify or your accounting export |
| “Here are my customers who haven’t bought in 60 days, with what they last bought. Who should I contact first, and what should I say?” | Last purchase date and amount per customer (use IDs, not names or emails) | Stripe, Shopify, your CRM |
| “Which traffic source sent visitors who actually converted this month, and where should my next hour of marketing go?” | Sessions and key events (conversions) by channel | Google Analytics 4 |
| “Here’s my price, my cost per sale and my conversion rate. Argue against raising my price 15%.” | Current price, unit cost, conversion or churn rate | Your pricing page, payment processor, analytics |
| “Here’s my cash, my monthly fixed costs and my last three months of revenue. How many months of runway do I have, and what’s the first cost you’d cut?” | Bank balance, list of recurring costs, revenue trend | Bank, accounting tool, subscriptions list |
| “These are my open tasks with due dates. Which three move revenue this week, and what can wait?” | Task list with due dates and what each is for | Trello, ClickUp, Notion, your calendar |
| “Here are my last 10 posts or videos with views and watch time. What do the winners have in common, and what should I make next?” | Per-post stats (views, watch time, engagement) | YouTube Studio, Instagram insights |
| “Here’s a customer complaint, their order history and my refund policy. Draft a reply that keeps them.” | The message, order history, your policy text | Inbox or WhatsApp, store admin |
| “Explain this contract clause (or tax notice) in plain English, and list the questions I should ask my lawyer or accountant.” | The exact text, with names and account numbers removed | The document itself |
| “Here are the recurring tasks I want off my plate and the hours they take. Write a job post for a part-time role.” | Task list, hours per week, budget | Your notes, time tracking |
Then push on the answer with follow-ups. These four do most of the work:
- “What are you assuming that I didn’t tell you?”
- “Which number in what I gave you matters most to your answer?”
- “What would change your recommendation?”
- “Argue the opposite, then tell me which side is stronger.”
For the job-post question, our guide on when to hire your first person covers what to check before you post it.
Where it will quietly fail you
Two limits worth knowing. First, a general chat assistant doesn’t know your business as it is today — it knows what you’ve told it and what it can fetch from a connected app when you ask, but it doesn’t keep your live numbers in view, and what you paste is stale the moment you send it. So it gives advice that could apply to any company. Second, it will state a confident wrong answer as easily as a right one. Treat it as a sharp intern, not an oracle: use it to draft and think, verify anything that matters.
Before you paste your numbers: what happens to your data
Most of the questions above need real business data, so know where it goes. Here’s what the two biggest providers say today, on their own pages:
- ChatGPT on a personal plan (Free, Go, Plus, Pro): go to Settings, then Data controls, and turn off “Improve the model for everyone”. OpenAI says new conversations then won’t be used to train its models, and they still stay in your history. Temporary chats don’t appear in history, don’t create memories and aren’t used for training, but may be kept for up to 30 days for safety. One catch: if you give a thumbs up or down, OpenAI says the whole conversation may be used for training (OpenAI: Data controls in ChatGPT).
- ChatGPT Business, Enterprise and the OpenAI API: OpenAI says it doesn’t train on inputs or outputs from these by default (OpenAI: Business data privacy).
- Claude on a personal plan (Free, Pro, Max): you choose in your Privacy Settings whether your chats can be used to improve Claude. Anthropic says that if you allow it, data is kept for five years; if you don’t, the 30-day retention period applies (Anthropic: Updates to Consumer Terms). Incognito chats aren’t used for training even if that setting is on, and thumbs up/down feedback stores the whole conversation for up to five years (Anthropic Privacy Center).
- Claude for Work (Team, Enterprise) and the Anthropic API: Anthropic says it doesn’t use inputs or outputs from its commercial products to train models by default (Anthropic: commercial data).
Three habits that cost nothing: send totals and trends rather than raw customer lists; replace names, emails and card or account numbers with IDs; and keep anything about customers or employees on a business plan or the API, where no-training is the default. For a framework on the “verify what matters” part, NIST’s AI Risk Management Framework is free and intended for voluntary use.
If you use Vexlynk: the assistant reads the cards on your board instead of what you paste. On the desktop app your boards are stored on your machine; the web workspace keeps them in the cloud (desktop vs web). On Plus and Pro you can bring your own key (OpenAI, Anthropic, DeepSeek or xAI), and then your requests are billed and handled by that provider under its API terms. The AI agent docs list exactly what it can read.
Links checked September 27, 2026.
The fix for the “doesn’t know your business” limit is context. The more of your real situation the assistant can see — your revenue, your traffic, your customers — the less generic it gets. That’s the difference between advice for “a business” and advice for yours, and it’s the whole reason a board that an assistant can actually read beats pasting screenshots into a chat.
Frequently asked questions
How do I use AI for my small business?
Use it as a thinking partner, a first-draft engine, a translator of your own data, and an explainer — and always give it a role plus your real numbers plus a specific decision, not a vague question. Verify anything that matters, since it can be confidently wrong.
Why does AI give me generic advice about my business?
Because a general chat assistant only knows what you’ve told it or what it fetches when you ask — it doesn’t keep your actual revenue, customers, and traffic in view. The more real context it can read, the less generic the answer. Specific inputs get specific outputs.
What are good AI prompts for a business?
The best ones give the AI a role, your real numbers, and a decision to make — for example, “You’re a pricing strategist for a $4k/month service business; here are my costs and current price — what should I test?” A free pack of 50 business prompts gives you ready-made ones.
Is it safe to paste business data into ChatGPT or Claude?
It depends on the plan and your settings. On personal ChatGPT plans you can turn off “Improve the model for everyone” in Data controls; on personal Claude plans you choose in Privacy Settings whether chats are used for training. Business plans and the APIs from both OpenAI and Anthropic don’t train on your data by default. Either way, strip names, emails and account numbers and send totals where you can.
Vexlynk
Vexlynk Team