Most small business owners I meet are paying for at least one AI assistant, usually ChatGPT, Claude or both. Most are also getting a fraction of the value. They use it like a search engine, get generic answers, and decide it’s “fine for emails”.
Used properly, these tools can take hours out of your operations every week. Used carelessly, they can send a client the wrong price or write an SOP that quietly invents a step. This guide covers how to get the first without the second.
ChatGPT or Claude?
Both are excellent, and both improve constantly, so don’t overthink it. In my day-to-day work:
- ChatGPT is a strong all-rounder with a wide set of features, including custom GPTs, image generation and voice.
- Claude is particularly good with long documents and careful, natural writing, and its Projects feature keeps your reference files and instructions in one place.
Pick one as your team’s main tool so everyone builds shared habits, prompts and reference material. Use a business or team plan rather than personal accounts, so you control the data settings and nothing is lost when someone leaves.
The one idea that changes everything: give it context
Generic input gets generic output. The biggest upgrade is to stop starting from a blank chat.
Both tools let you save standing context. ChatGPT has custom instructions and custom GPTs. Claude has Projects. Set one up per area of your business and load it with:
- A short description of your business, your services and your ideal client.
- Your tone of voice, with two or three real examples of emails or posts you’ve written.
- Key facts: prices, packages, policies, turnaround times.
- Your SOPs and templates for that area.
- Rules, such as “Never quote a price that isn’t in the price list” and “Use British spelling”.
Now every chat starts from your business, not from the internet’s average.
Six practical uses, with prompts
1. Turning how you work into SOPs
Record yourself doing a task on a screen recording or voice note, get a transcript, and paste it in with this prompt:
Turn this transcript into a step-by-step SOP for a new team member. Include: purpose, when to use it, tools needed, numbered steps, common mistakes, and what “done” looks like. Flag any step where I was vague or seemed to skip something, rather than guessing.
The last line matters. It tells the AI to point out gaps instead of inventing steps.
2. Drafting client communication
Use it for first drafts of proposals, onboarding emails, project updates and difficult conversations:
Draft a reply to this client email. Context: [paste the thread]. Goal: [e.g. push the deadline by one week without sounding defensive]. Keep it under 150 words, warm and direct, in my tone (see examples in the project).
Always read before sending. Check names, dates, prices and promises line by line.
3. Meeting notes to actions
Paste a transcript and ask:
List: decisions made, action items with owner and due date, open questions, and anything a client asked for that we haven’t agreed to. Use only what’s in the transcript. If an owner or date wasn’t stated, write “not stated”.
“Use only what’s in the transcript” stops it filling gaps with plausible guesses.
4. Weekly reporting
Export last week’s completed and overdue tasks, or your metrics sheet, and ask:
Write a five-bullet weekly operations summary for the founder: what moved, what’s at risk, what needs a decision, and one thing to improve. Be specific and use the numbers.
5. Sorting the inbox and enquiries
Paste a batch of enquiries and have it categorise them by service, urgency and fit, with a suggested next step for each. For ongoing volume, connect this to Zapier or Make so it runs automatically, and send the results to a review list rather than straight to clients.
6. Thinking partner for decisions
It’s useful for pressure-testing a plan:
Here’s my plan for [e.g. raising prices for existing clients]. Act as a sceptical operations advisor. What could go wrong, what am I not considering, and what would you want to know before deciding?
It won’t make the decision for you, but it’ll surface questions you hadn’t asked.
The rules I follow with every client
These are non-negotiable in any business I run operations for.
Rule 1: Nothing AI-written reaches a client without a human check
Every AI output that goes outside the business passes through a person first. In automations, that means a Needs review status or a draft folder, never a direct send.
Rule 2: Verify facts, numbers and names every time
AI can state wrong things confidently. Prices, dates, names, legal or financial statements and anything quoted from a document all get checked against the source.
Rule 3: Be careful with sensitive data
Don’t paste passwords, bank details, ID numbers or client personal data into AI tools. Use business plans with training on your data switched off, and agree a simple team policy on what can and can’t go in.
Rule 4: Save what works
When a prompt produces great output, save it in a shared prompt library: in your Claude Project, a custom GPT, or a page in Notion or ClickUp. The value compounds when the whole team uses your best prompts instead of reinventing them.
Rule 5: Review your AI setup regularly
Your prices, services and processes change. Update the context files, instructions and saved prompts at least monthly, or the AI will keep working from last quarter’s business.
A simple starting plan
If you’re not getting much from AI yet, do this over the next two weeks:
- Choose one main tool and move your team onto a business plan.
- Build one project or custom GPT for your most repetitive area, usually client communication.
- Load it with your services, prices, tone examples and rules.
- Use it for every draft in that area for two weeks, checking every output.
- Save the five prompts that worked best and share them with the team.
Then move on to the next area.
When it becomes an operations job
Once AI is woven into several workflows, someone has to own it: keeping the context current, checking output quality, fixing the automations it feeds, and deciding where it should and shouldn’t be used. That’s the part of the job I take on for clients. If you’d like help setting it up properly, start with an Operations & AI Audit.