To use AI well as a founder or small business owner, stop treating it like a search engine and start treating it like a fast, capable new hire on their first day. Give it context about your business, hand it one clear task, show it an example of what good looks like, then review what comes back and tell it what to fix. That loop (context, task, example, review) is most of the skill. The tool you pick matters far less than the habit.


In practice, that means choosing one general assistant such as ChatGPT, Claude or Gemini, loading it with a short brief about your company, and using it daily for three kinds of work: thinking through decisions, drafting the writing you already do, and eventually running repetitive tasks without you. The rest of this guide shows how to do each one, including the prompts, the limits and when to graduate from chatting to automating.


Step 1: Pick one assistant and stop shopping


Founders lose weeks comparing tools. Don't. The major assistants can all help with the core jobs a small business has: writing, summarizing, brainstorming and working with documents you upload. What each one can do with files, how much you can upload and which features you get depend on the plan and change often, so check the current details before you commit.


A simple way to choose:



  • You live in Google Workspace (Gmail, Docs, Drive): Gemini is a natural place to start, since it comes from the same company as your files.

  • You write a lot of long documents (proposals, reports, policies): Claude is worth trying first. This is my editorial view, not a ranking, so test it on your own writing.

  • You want a broad set of features in one place: ChatGPT is a reasonable default.


Start on a free tier to learn the basics, and upgrade if you hit usage limits or need stronger models. Paid plans and their limits change regularly, so check current pricing directly, for example on the ChatGPT pricing page and the Anthropic pricing page. If a paid plan saves you even a couple of hours a month, the subscription may be easy to justify.


Commit to one tool for 30 days. Switching constantly means you never build the setup that makes AI useful.


Step 2: Write a founder brief once, reuse it forever


The biggest reason AI gives generic answers is that it knows nothing about you. Fix that once.


Most assistants have a place to store standing context. ChatGPT has custom instructions and Projects. Claude has Projects, where you can add instructions and files. Gemini has Gems, which are saved assistants with their own instructions. Feature names and availability vary by plan, so look in your settings. Put a one page brief in there that covers:



  1. What the business does in two sentences.

  2. Who the customer is: industry, size, what they worry about.

  3. Your offer and price range, so suggestions stay realistic.

  4. Your voice: "plain, direct, short sentences, no corporate language, Canadian spelling."

  5. What you never want: "don't invent numbers, don't use exclamation marks, ask me when you're unsure."

  6. Your current priorities: "this quarter we're focused on getting five new accounting firm clients."


Add two or three samples of your best emails or posts as reference files. Now every conversation starts from your business instead of from zero.


Step 3: Use a four part prompt


Most bad AI output comes from a vague request. Here's the shape that works almost every time:



  1. Context: who it's for and why.

  2. Task: the exact thing you want.

  3. Example or constraints: what good looks like, length, tone.

  4. Format: bullet list, table, email, three options.


A weak prompt: "Write a follow up email."


A strong prompt: "I met the owner of a 12 person bookkeeping firm at a networking event on Tuesday. She said month end close is chaos because client receipts arrive by email, text and paper. Write a follow up email under 120 words that reminds her of that conversation and suggests a 15 minute call next week. Match the tone of the sample emails in this project. Give me two versions: one warm, one more direct."


The second prompt takes a little longer to write and usually saves you rounds of edits. When the result is close but not right, don't start over. Reply with what to change: "Shorter. Cut the second paragraph. Make the ask more specific." Treat it like editing a junior's draft.


Step 4: Use AI as a thinking partner, not an oracle


The most underrated way founders use AI isn't writing. It's thinking. A good assistant is a tireless sparring partner that will poke at a decision at 11pm when nobody else is around.


Prompts that work well for decisions:



  • Ask questions first: "I'm deciding whether to hire a part time salesperson or spend that money on ads. Before you give any advice, ask me the ten questions you'd need answered."

  • Run a premortem: "Assume this pricing change failed badly six months from now. Write the three most likely reasons why."

  • Argue the other side: "Here's my plan. Make the strongest possible case against it."

  • Compare options in a table: "Lay out these three options with cost, time to results, risk and what I'd need to believe for each to be the right choice."


The rule: AI widens your view, you make the call. It doesn't know your cash position, your gut read on a client or what your team can actually handle. Use it to surface what you missed, then decide yourself.


Step 5: Find the tasks worth handing over


Once the basics feel natural, look at your week. Write down every task you did over five working days, then run each one through three filters:



  1. Is it repetitive? You do it more than a few times a week.

  2. Is it mostly text or data? Reading, writing, sorting, extracting, summarizing.

  3. Is a mistake cheap to catch? You or someone can review the output before it matters.


Tasks that pass all three are your first candidates. For many small businesses the list looks something like this:



  • First drafts of proposals and quotes

  • Turning call notes into a summary and a task list

  • Replying to common customer questions

  • Rewriting one blog post into social posts and a newsletter

  • Pulling key fields out of invoices, contracts or forms

  • Writing job descriptions and standard operating procedures


Tasks that fail the filters, such as a sensitive conversation with an unhappy client or a final legal decision, stay with you. AI can help you prepare for them, but it shouldn't run them.


Step 6: Check everything that matters


AI is confident even when it's wrong. Knowing its limits is part of knowing how to use AI.



  • It makes things up. Models can invent facts, statistics, quotes and even sources that look real. Never publish a number or a citation from AI without checking the original.

  • It can be out of date. Unless it's searching the web, its knowledge stops at a training date. Ask it to search, or paste in current information yourself.

  • It's shaky with precise math. For anything financial, have it build the formula in a spreadsheet rather than trusting a number in a chat reply.

  • It agrees with you too easily. If you lead with your opinion, it tends to back it. Ask for criticism explicitly.

  • It is not your lawyer, accountant or doctor. AI output can help you prepare questions and understand a topic, but it does not replace professional legal, financial or medical advice. Take anything with real consequences to a qualified person.


Before you paste in client data


Your data goes somewhere. Before you paste in client names, financials or personal information, check three things in your vendor's current privacy and data settings:



  1. Training: whether your conversations can be used to train the model, and whether you can switch that off.

  2. Retention: how long chats and uploaded files are stored, and whether you can delete them.

  3. Controls and contracts: whether your plan includes admin controls for your team and whether the vendor offers a data processing agreement.


These answers differ by vendor and by plan, and they change, so read the policy for the exact plan you're on rather than assuming. If you handle personal information in Canada, check your obligations with the Office of the Privacy Commissioner of Canada before putting that information into any outside tool.


A simple habit: the higher the stakes, the more you review. A social caption gets a quick read. A contract summary gets checked line by line against the source.


Step 7: Turn repeat prompts into workflows


Here's the signal that you've outgrown the chat window: you're pasting the same prompt, with the same kind of input, several times a week. At that point the job is no longer "ask AI". It's "build a small system that asks AI for you".


The basic shape of an AI workflow has four parts:



  1. Trigger: a new email arrives, a form is submitted, a file lands in a folder.

  2. AI step: the model reads the input and extracts, classifies or drafts something, using the same prompt you've already refined in chat.

  3. Check: a person approves, or rules flag anything unusual for review.

  4. Output: the result goes into your CRM, a spreadsheet, a document template or a reply.


Automation platforms like Zapier, Make and n8n are built to connect apps into steps like these with little or no code. Before you build, check that the platform has a current connector for the AI model you use and that your plan includes it.


A real workflow example: a Toronto process serving firm had staff preparing affidavits by hand. Once the work followed a clear pattern, it became a workflow. The result was 1,500 affidavits per month automated with zero manual touches, and 90+ hours saved monthly for that firm, more than half a full time hire. On a separate document project, per document handling time was cut from 4 to 6 minutes to about 30 seconds, with 500 documents processed in under a month for a live paying client.


The lesson isn't the numbers. It's the order. Both started as a repetitive, text heavy task with a predictable structure. The prompt and the rules were worked out first, then the automation was built around them. If you can't get a task to work reliably in a chat window, automating it will only produce bad output faster.


A 30 day plan to actually start


Week 1: Set up. Pick your assistant. Write your founder brief. Add three samples of your writing. Use it for every email you'd normally write from scratch.


Week 2: Think with it. Take one real decision you're facing and run it through the four decision prompts above. Note what it caught that you missed.


Week 3: Audit and delegate. List a week of tasks, apply the three filters, and hand AI the top three. Save the prompts that work in a document or inside your Project.


Week 4: Measure. Estimate the time saved per task. Find the one prompt you used most often. That's your first automation candidate.


By the end of the month you'll have a tuned assistant, a small library of prompts that work for your business, and a clear idea of which task to automate first.


Mistakes founders make with AI



  • Asking once and giving up. The first answer is a draft. The value is in the second and third rounds.

  • No context. Generic input gives generic output. Your founder brief fixes most of this.

  • Publishing without reading. One invented statistic in a client proposal costs more trust than the time AI saved.

  • Automating a broken process. If the manual version is messy, AI makes the mess faster. Clean up the steps first.

  • Trying everything at once. One tool, one brief, three tasks. Depth beats breadth.


The short version


How to use AI as a founder or small business owner comes down to this: give it context, give it a clear job, review its work, and keep the decisions for yourself. Start with one assistant and one brief. Use it to think, then to draft, then to run the repetitive work you shouldn't be doing anymore. The people who get the most from AI aren't the ones with the most tools. They're the ones who use one tool every single day.