Tool
Claude, ChatGPT or another AI assistant
Price
Free with the free plans
Time
10 min read
Level
Beginner

You asked AI for a quote, a reply to a customer or a LinkedIn post. The result was generic, or contained something false, and you closed the tab thinking it was not made for your trade. This guide covers five classic mistakes freelancers and small businesses make when they start, and how to fix each one.

The 5 mistakes and how to fix them

  1. Asking in one line, with no context"Write a follow-up email" gets you any follow-up email. Anthropic's documentation advises treating the AI like a brilliant new employee who does not know your norms, and offers a simple rule: if a colleague with no background on the task would be confused by your request, the AI will be too. Explain the why as well: Anthropic says that giving the reason behind an instruction helps the model aim better. The request template below handles this.
  2. Believing every answerAI can state something false in a very confident tone: this is called a hallucination. Anthropic gives three simple safeguards: allow the AI to say "I don't know", ask it to rely on exact quotes from the documents you provide, and forbid it from using anything other than those documents when it matters. The same documentation says these techniques reduce the problem without eliminating it: check every figure, date or rule yourself before sending it.
  3. Pasting sensitive data without thinkingA client contract, a bank statement, a password in a screenshot. First, look at your account settings: on Claude's consumer plans, the "Model Improvement" setting decides whether your chats may be used to improve the models. Then get into the habit of replacing names with roles ("the client", "the supplier") and never pasting passwords or bank details.
  4. Starting from scratch in every conversationExplaining your business every time is tiring, so you end up skipping it, and the answers turn generic again. Store your information once in a Claude project (or your tool's equivalent): the project's documents and instructions apply to all its conversations. The list of documents to prepare is in the tutorial prepare your documents for AI.
  5. Judging on a single tryA mediocre first answer is not a verdict. Reply to the AI as you would to a colleague: "too long", "more direct", "keep the second paragraph". And show it an example of what you want: Anthropic describes examples as one of the most reliable ways to get the right format and tone. My rule: do a task several times by hand with AI before thinking about automating it.

The request template to copy

It fixes mistakes 1, 2 and 5 in one go. Keep it in your notes and fill in the brackets for each new task.

Prompt
Context: [my business in one sentence, my target customers, the exact situation]
Task: [what I want to get]
Why: [what it will be used for, who will read it]
Format: [length, tone, formal or informal, structure]
Example of what I like: [paste a text of yours that you like]
Avoid: [words, promises, topics]

Rules:
- Rely only on what I give you here. If information is missing, ask me instead of making it up.
- If you are not sure about a fact, a figure or a rule, say so clearly.
- At the end, list the points I need to check before using this text.

A real-life example: a plumber wants to follow up on a quote that got no reply. With a one-line request, the AI writes a generic email. With the template (quote sent ten days ago, private customer, friendly tone, one of his past follow-ups as an example, no discount allowed), he gets a short message that sounds like him, plus a list of two points to check: the amount and the quote's validity date.

If you would rather have the AI interview you first, the tutorial a prompt that makes AI ask its questions first does the same job the other way round.

Good to know

  • This advice applies to every assistant, even though the sources quoted are from Anthropic, the company behind Claude.
  • Used well, AI saves time on a first draft; it does not replace your review or your responsibility toward the customer.
  • Before automating a task, make sure you can describe it in a few lines and that the manual result satisfies you.

Be careful. For team use or with customer data, read your plan's terms and privacy settings before you start, and agree together on what must never be pasted into an AI assistant.

Sources

Once a task works well by hand, it can run on its own: that is the kind of automation I set up.