← Marketing with AI

Thank-you notes with AI: a saved skill that drafts, so you only sign

Writing thank-you notes with AI is a Skill job: you set it up in one sitting, about thirty minutes, and then each note takes two or three minutes. You save one set of instructions that knows your voice and your rules, feed it a single detail after each purchase, referral or review, and it drafts the note. You read it, fix anything wrong, and sign it. The AI drafts; you sign.

That split matters because a thank-you note fails in two ways. It fails when it never gets written, which is the usual case for a busy owner. It also fails when it reads like a form letter, which is the usual case when someone tries to batch them. A saved skill solves the first problem, and the one detail you add each time solves the second.

Shape
Skill
Sitting
one sitting

The shape: Skill

This job is a Skill because the work repeats, but every instance is small and personal. There is no campaign to plan and no tool to connect. There is a trigger, such as a sale, a referral or a review, and a short piece of writing that should follow it within a few days.

A skill here means a saved set of instructions the AI follows every time. In Claude that is a Project with instructions; in ChatGPT it is a custom GPT or a saved project. If you prefer, it can be a block of text in your notes app that you paste at the top of a fresh chat. The format does not matter much. What matters is that you write the rules once, so the tone, the length and the things to avoid stay the same from the first note to the hundredth.

The payoff is speed at the moment it counts. The hard part of a thank-you note is starting it, and the skill removes that. You open the chat, type two lines about what happened, and a draft is ready before you lose the impulse to send it.

What you give the AI

The skill needs two kinds of input: the standing rules you give once, and the one detail you give each time.

Give these once, when you set up the skill:

  • Three thank-you notes or warm emails you have already sent, with names removed, so the AI can hear how you write when you mean it.
  • One sentence on what your business does and for whom, in the words you use out loud.
  • How you sign off, exactly as you write it, including your first name or your team name.
  • The three triggers you thank people for: a purchase, a referral and a review, plus any others that fit your business, such as a long-time client renewing.
  • Your no-go list: no discounts or gifts tied to reviews, no request for another review, no upsell, no exclamation-heavy cheer if that is not how you talk.
  • The format you send in, whether that is a handwritten card, an email, or a text message, since each one has a different natural length.

Give these each time you need a note:

  • The person's first name.
  • The trigger: what they did.
  • One specific detail, such as what they bought, who they referred, or a phrase from their review.
  • Anything to avoid, such as a delay in their order that you have already handled.

The specific detail is the whole difference between a note that works and one that does not. Without it, the AI can only write a polite, interchangeable thank-you. With it, the note proves you noticed this person.

What comes back

Each time, you get one short note of three to five sentences, written in your voice and sized for the format you named. If you ask, you also get a shorter version that fits on the inside of a small card.

Paste this into the instructions of a Claude Project or custom GPT, or keep it in a note and paste it at the top of a new chat:

You draft short thank-you notes for my business. The owner reviews and signs every note.

About the business: [one sentence on what we do and for whom]
How we sign off: [exact sign-off]
Default format: [handwritten card / email / text]

Examples of how we write when we mean it:
[paste three past notes or warm emails, names removed]

Every time you get a request, it will include a first name, the trigger
(purchase, referral, review, or other), and one specific detail.

Rules:
- Three to five sentences. A card version is under 50 words.
- Open with the person's name and what they did, in plain words.
- Use the specific detail in the second sentence, exactly as given.
- Say plainly, in one sentence, that it matters to you. Do not invent a reason or a feeling.
- No discount, gift, coupon, upsell, or request for another review.
- For a review thank-you, thank them for taking the time. One phrase from the review is fine; do not quote the rest or praise its rating.
- For a referral, name the person they referred only if the request
  says it is fine to.
- Never invent a detail that was not in the request. If the detail
  is missing, ask for it instead of writing the note.
- Match the tone of the examples. No exclamation marks unless the
  examples use them.

Then, each time a trigger happens, send two lines like this:

Name: Dana. Trigger: referral. Detail: she sent her neighbor Luis, who booked a spring cleanup. Fine to name Luis. Format: card.

The draft comes back in a few seconds. Read it once, change anything that is not true, and send it or copy it onto a card.

How to judge it

A good note reads as if you wrote it on a good day. You can reject a bad one in under a minute by looking for these:

  • It could go to anyone. If you can swap the name and send it to another customer unchanged, the specific detail is missing or buried. Add it or move it up.
  • It invents something. The AI mentions a feeling, an event or a fact you did not give it, such as "your lovely family" or "your second visit." Delete it. A thank-you note with a wrong detail is worse than none.
  • It asks for something. Any line that requests a review, a referral, a follow-up purchase or a social share turns the note into a pitch. Cut it.
  • It gushes. Three adjectives in a row, a string of exclamation marks, or phrases like "beyond grateful" read as a template. Trim it to one plain sentence of thanks.
  • It runs long. A thank-you note past five sentences starts to read like a newsletter. Shorter feels more personal.
  • It sounds like the website. If the note uses your marketing language instead of the way you speak, your example notes are too few or too formal. Swap in warmer ones and update the skill.

When the same problem shows up twice, fix the instructions, not just the note. That is the point of the skill: each correction makes every later draft better.

What stays with you

Three parts of this job no AI does, and the page would be dishonest to suggest otherwise.

The first is noticing. The AI cannot know a customer referred a friend, or that a review went up last night, unless you tell it. The skill only drafts when you notice and take thirty seconds to type the trigger. If you want the habit to stick, tie it to something you already do, such as checking new reviews once a week or closing out a job.

The second is the facts. You confirm the name is spelled right, the detail is true, and the referred person is fine with being mentioned. For a review thank-you, check whether a public reply is the better move, and what it should say; our guide to responding to Google reviews covers the careful cases. Keep incentives out of it entirely, since Google's review policy does not allow rewards in exchange for reviews.

The third is the signature, and the send. A handwritten card in your own handwriting carries something no draft can. An email still goes out from your account, after you read it. That last read is where the note becomes yours.

An editorial example

This example is invented to show the flow, not drawn from any customer.

Say you run a one-person dog grooming van. In one week, three things happen: a new client books a full groom for a nervous rescue, a regular sends her sister, and someone leaves a review mentioning that their dog came home calm.

You set up the skill on Sunday with three texts you have sent clients in the past, your sign-off ("Thanks again, Priya and the van"), and a default format of a short card left on the doorstep.

On Tuesday you type: "Name: Marcus. Trigger: purchase. Detail: first groom for Biscuit, a rescue who was nervous at the start and fell asleep on the table." The draft thanks Marcus for trusting you with Biscuit, mentions the nap, says plainly that it meant a lot, and signs off. You change one word and copy it onto a card.

On Thursday the referral note takes two minutes, and Friday's review thank-you goes out as a short email. Three notes, under ten minutes in total, each one specific to a single dog and a single owner.

Card, email or text

The format is a real choice, and the skill handles all three if you tell it which one you want.

TriggerBest formatWhy
A first purchaseHandwritten card or emailA card stands out; email is fine when you have no address.
A large or repeat purchaseHandwritten cardThe effort matches the size of what they did.
A referralHandwritten cardReferrals are the most generous thing a customer does.
A public reviewEmail or text, or a public replySpeed matters more than paper; when you have no contact for the reviewer, the public reply is the thank-you.

If you only pick one format, pick the one you will actually keep doing. A steady habit of short emails does more than a box of cards you never write.

For more on what happens around these moments, see how to get more Google reviews and how to reuse one review across website, social and email. If you want another job to fill the same spare credits, the full DIY marketing list has every one.

Of every job on this list, social media is the one that comes back every week, and Boomp is software that does that one for you. See ten posts made from your website, free

Frequently asked questions

When should a small business send a thank-you note?

Send one after the moments a customer chose to help you: a first purchase, a large or repeat purchase, a referral that turned into a conversation, and a public review. Send it within a few days, while the customer still remembers what they did. A note that arrives weeks later reads as a chore you finally got around to, and the warmth of the gesture fades with the delay.

Can AI write a thank-you note that does not sound generic?

Yes, if you give it one specific detail each time. A saved skill with your voice and your rules handles the structure and the tone. The detail you add, such as the color the customer picked or the friend who sent them, is what makes the note read as written for one person. Without that detail, any writer, human or AI, produces the same polite filler.

Should a thank-you note be handwritten or emailed?

Handwritten notes stand out because few businesses send them, so they suit referrals and bigger purchases. Email suits a review thank-you or a quick follow-up after a routine sale, since speed matters more there. The AI draft works for both. For a card, copy the draft by hand and trim it to fit. For email, paste it, read it once, and send it yourself.

Is it okay to offer a discount or gift as thanks for a review?

Be careful here. Google's review policy does not allow incentives in exchange for reviews, and a gift sent after a review can look like payment for it. Thank the reviewer in words only. If you want to reward loyalty, do it on its own schedule, for every customer who qualifies, and never tie it to whether someone left a review or what that review said.

How long should a business thank-you note be?

Three to five sentences is enough. Name what the person did, add one specific detail that shows you noticed, say what it means to the business in plain words, and stop. Skip the sales pitch. A note that ends with a coupon or a request for another review turns a thank-you into a transaction, and the customer can tell the difference in the first line.

What is an AI skill, and does it need a paid AI plan?

A skill here means a saved set of instructions the AI follows every time you call on it, such as a Claude Project or a custom GPT, or simply a prompt you keep in a note and paste in. Paid plans make saving it easier, but the instructions are the part that matters. If you can paste text into a chat window, you can use this approach today.

Of every job on this list, social media is the one that comes back every week, and Boomp is software that does that one for you.

See ten posts made from your website, free

Rather not do this yourself? Start with when a free scheduler is enough or what done-for-you social media should handle.

Save this to Pinterest

Thank-you notes with AI: a saved skill that drafts, so you only sign
KC

Written by Kathleen Celmins

Founder of Boomp. Helping local businesses stay visible on social media without doing the work themselves.