AI for freelancers — writing quotes and client emails in half the time

As a solo freelancer you write more than you'd like. A quote here, a follow-up email there, a polite rejection to an inquiry that doesn't fit. I once counted for a week and came to eleven written customer contacts that together cost almost four hours. Not because the texts were hard, but because I started from zero every time.

That's exactly what AI is good for in everyday life, not for the one brilliant text, but for the twenty solid standard texts a week you have to write anyway. This playbook shows you a fixed setup with which you finish quotes, emails and rejections in 10 to 15 minutes instead of an hour. It's for freelancers, coaches, tradespeople with their own business, consultants and everyone who works without a team. No terminal, no API, a normal ChatGPT or Claude account is enough.

Step 1, Collect your three most common writing tasks

Before you open any tool, write down which texts you really need often. For most freelancers it's the same three, a quote after a first conversation, a follow-up email when nobody responds, and a friendly rejection. Maybe an appointment confirmation or an invoice reminder gets added for you.

Take the three most common. For each you pull two or three real examples from your sent folder that worked well. These examples are worth gold later, because the AI learns your tone from them instead of inventing one.

Step 2, Set up a style profile the AI reads along every time

The biggest time-waster is explaining to the AI in every new chat who you are. You do that exactly once. Write a short style card, five to eight lines, with your role ("I'm a self-employed graphic designer, clients are small businesses in the DACH region"), your tone ("friendly, direct, no advertising speak"), and your anti-patterns ("no platitudes like Dear Sir or Madam, no in today's fast-paced market").

In Claude you store this as a Project, in ChatGPT as Custom Instructions or as a custom GPT. Both read it along automatically in every chat. If you don't want to use that, a text file on the desktop that you paste at the top of the chat also does. The only thing that matters is that you don't type it out anew every time.

Step 3, Always give the AI the facts, never the wording

The mistake almost everyone makes, they write "Formulate me a friendly quote for a client". Then comes a generic text you have to completely rewrite. Turn it around. Give the hard facts in bullet points and let the AI build the prose from them.

So, client is called Krüger Bakery, wants a new online shop, budget roughly 4,000 euros, timeframe six weeks, three feedback rounds included. From these five lines the AI makes a clean quote. You only check numbers and tone, instead of writing it yourself.

Step 4, Build yourself three fixed prompt templates

Now it gets concrete. For each of your three tasks you write a template you always reuse. A quote template looks like this, for example, "Write a quote based on these facts: [FACTS]. Use my style from the profile. Structure: short greeting, what I deliver, price, next step. Maximum 200 words."

The square brackets are the only spot you swap out per client. Save the three templates where you find them in two seconds, in a notes app, as a text snippet or as saved prompts. From now on a quote is a copy-paste plus five lines of facts.

Step 5, Always have two variants delivered

Append to every template the sentence "Give me two variants, one factual-terse and one warmer". That costs the AI no extra time, but saves you the round where you say "a bit friendlier please". You read two versions, take the better one or mix two sentences. In practice I take the terse variant in about half the cases and adjust two words.

Step 6, Check every number and every name by hand

This is the one spot where you must never cut corners. AI models occasionally invent details, a wrong date, a price you never quoted, a company name with a typo. That's called hallucination and happens even to the best models. If that ends up in a quote, it's embarrassing or expensive.

Make yourself a fixed rule, before an email goes out you read numbers, names and dates once against your notes. That takes 30 seconds and is your safety net. More on the topic is in the lesson on hallucinations.

Step 7, Keep customer data out of the prompt where you can

As a freelancer you work with other people's data, and for that GDPR applies to you too. For a quote you rarely need the full real name plus address plus phone number in the chat. Use placeholders like "the client" or just the first name, and only insert the real data during fine-tuning in your mail program.

If you process sensitive data, check whether your tool offers a business tier with a data-protection guarantee, where your inputs aren't used for training. The details are in the playbook Data protection with AI tools.

Step 8, Set up a text snippet for follow-up emails

The follow-up email is the task that saves the most time, because it's the most annoying. Nobody likes to follow up. Build yourself a template that turns the last contact into a short, non-pushy reminder, "The client received a quote eight days ago, hasn't responded. Write a short friendly inquiry, maximum 80 words, no mention of a discount, one clear next step."

For me this now runs every Friday morning in one go. I open the open quotes, throw in three lines of context per client and have five follow-up emails in ten minutes. Before, I put it off for weeks, which cost me real gigs.

Step 9, Turn the rejection into a 2-minute task

Nobody likes writing rejections, so they pile up and then the inquirer feels ignored. A good, friendly rejection keeps the door open for later. Template, "Write a polite rejection to this inquiry: [INQUIRY]. The reason is [at capacity / doesn't fit the profile / budget too small]. Tone warm, short, offer to get back in touch in three months if it makes sense."

The trick is the reason in brackets. You give it honestly, the AI wraps it diplomatically. That way even a budget rejection doesn't sound dismissive, and you're through in two minutes.

Step 10, Keep a folder with your best results

Every time a text was really good, copy it into a folder or a note "Good templates". After four weeks you have a small collection of your best quotes and emails. You feed those back into your style profile as needed, then the results get more and more like you and less and less like AI over time.

That's exactly the difference between "I sometimes use ChatGPT" and a real system. The system keeps building itself out week by week, without you putting in extra work.

What comes next

When this setup is in place and you notice you keep refining your prompts, the playbook Building a prompt library is the logical next step. And if you want to understand why the two-variant rule from step 5 works so well, you'll find the craft behind it in Level 2, the lessons on prompting and system prompts.

Sources

• Claude Projects, official docs, https://support.anthropic.com/en/articles/9517075-what-are-projects
• ChatGPT Custom Instructions, official docs, https://help.openai.com/en/articles/8096356-custom-instructions-for-chatgpt