Dry cleaners can use AI for wording, never for the verdict. Staff dictate what they see at the counter and AI tidies it into a note. When a claim arrives, the owner decides the outcome first, then AI drafts the reply from that decision. It never judges fault, works out a settlement on its own or sends anything unread.
A claim is only as easy to settle as the note made at the counter, weeks before anyone complains. A ticket that says "stain on front" protects nobody. "Red wine, left lapel, about 3 cm, customer says Saturday, removal not guaranteed" protects the customer and the shop alike. That sets the order of work: better notes at drop-off, a sort of every claim before anyone replies, a decision by the owner, and only then a letter.
Drop-off: the note that settles the claim later
The Drycleaning & Laundry Institute's guide to home stain removal asks customers to tell the cleaner the location of specific stains and any procedures they have already used. That gives the counter note its shape. Said aloud in the same order for every item with a mark or a flaw, it has three parts. Where: the garment, the side, and the position and size of each mark ("left lapel, about 3 cm"). What: how it looks, what the customer says caused it and when, and anything they have tried since. Promise: what you told them about the result, then the care label as read from the garment.
Staff can say all of that into a phone in under a minute, and an assistant tidies it into text. The risk is a tidy-up that adds things nobody said, so the rules come first. Save them once as the instructions of a ChatGPT Project or a Claude Project (both allow this on free accounts; Claude caps Free at five projects), and nobody has to retype them.
The rules, as the assistant sees them:
You turn a dry-cleaning counter dictation into two pieces of text.
1. Use only what I dictate. If something is not said, write "not stated".
2. Describe condition in my words (worn, frayed, faded, loose). Never write
"damaged", "defect" or "pre-existing" unless I say that word.
3. Name a stain only as I did. "Looks like wine" stays "looks like wine".
4. Write "removal not guaranteed" only if I say I told the customer.
5. Output exactly:
TICKET NOTE: one line per mark or flaw, then the care label.
RECEIPT LINE: plain words for the customer, under 40 words.Here is a dictation for a navy wool coat, and the text that came back (both invented for illustration):
Navy wool coat. Left lapel, red mark, looks like wine, about three centimetres, customer says Saturday and dabbed it with a damp cloth. Told them removal is not guaranteed. Right cuff worn, loose threads. Second button down is loose. Care label: circle with a P.
Ticket note: Left lapel: red mark, looks like wine, about 3 cm; customer says it happened Saturday and was dabbed with a damp cloth. Right cuff: worn, loose threads. Second button from top: loose. Label: circle with P.
Receipt line: The mark on the left lapel will be treated, but removal is not guaranteed. Worn cuff and loose button noted at drop-off.
Two checks on every note. The sizes and positions match what was said, and the receipt line promises nothing the member of staff did not say. Here the assistant kept "worn" and "loose" as dictated. A weaker prompt turns "worn cuff" into "pre-existing damage", which sounds like a finding and was never one.
Set the note beside the usual ticket line, "coat, stain, button". Three weeks later, when the customer says the lapel was never marked, "stain, button" cannot tell you which button, which side, how big, or whether anyone warned them. For ready texts, pickup booking and uncollected garments, the tutorial on AI for dry-cleaning orders and collections covers the rest of the order.
Stain wording that promises effort, not results
A receipt line should promise what the shop controls, which is the attempt, and name what it does not control, which is the result. The table gives ticket and receipt wording for the situations that cause the most arguments. Adjust it to your own terms, because wording that suits one shop's terms may not suit another's.
| Situation | Ticket note | Receipt line |
|---|---|---|
| 1. Old or unknown mark | Mark, [position], [size], origin unknown, looks old. | We will treat this mark, but old marks do not always come out. |
| 2. Customer tried a home remedy | Customer applied [what they said]; may affect removal. | Home treatment can set some marks. We will do what we can. |
| 3. Spilled drink, food or oil that may not show | Spill reported: [what], area [where]; may not show until cleaned. | Spills that dry clear can turn yellow or brown after cleaning. Please tell us where. |
| 4. Label says do not dry clean | Label: crossed circle. Warned; cleaned at customer's request. | The label says not to dry clean this item. We have cleaned it at your request. |
| 5. No care label | No label; fibre not confirmed; method chosen by [initials]. | There is no care label, so the cleaning method is our best judgement. |
| 6. Beads, sequins or glued trims | Trim: [type], fixed with [thread or glue, if known]. | Beads and glued trims can loosen or change in cleaning. We cannot guarantee them. |
| 7. Strong dye beside a pale panel | Dark dye next to pale area; colour run possible. | Strong dyes can run in cleaning. We cannot guarantee against colour change. |
If your terms do not let a customer accept the risk on a "do not dry clean" item, drop row 4 and decline the job instead. Ask your insurer or an adviser which route suits you.
Counter staff should read the care label aloud, so that the note records what a person read and not what a photo seemed to show. A circle means professional cleaning is possible. A letter inside it, P or F, says which solvents may be used. A crossed circle means do not dry clean, and a W in a circle means professional wet cleaning. One bar under the symbol means a gentle process and two bars mean very gentle, as Wikipedia's article on laundry symbols sets out. Say "circle with P" or "crossed circle", never "dry cleanable", so the note records what the label shows and not a conclusion.
Row 3 comes from the same institute guide, which says that stains from oily substances, food and drinks can be invisible once dry and then turn yellow or brown with time or after a cleaning process. If a customer mentions a spilled drink, record where, even when nothing shows. That one line can later show that you were told, or that you were not.
Pick-up day: the coat is back and the customer is unhappy
The customer collects the navy coat and says the lapel mark is still faintly there and a button has gone. The coat cost $300 three years ago, they say, and they expect it replaced. The cleaning charge was $18. All the figures here are invented.
Before any AI is involved, look at the garment and the file. Is the mark the size the note says? Does the button turn up in the bag, the pressing area or the workroom? Ask when and where the customer first noticed the problem, and write down the answer in their words.
Then paste the note and the customer's message, with names removed, under this prompt:
Compare the customer's message with our intake note. Do not decide who is at fault.
List four things: MATCHES, DIFFERENCES, UNKNOWN, and TO GATHER (evidence to collect
before replying). Quote the note's exact words where you can.What the assistant returned for the coat (illustrative):
Matches: the lapel mark was recorded at drop-off with a size ("about 3 cm"), a cause given by the customer ("looks like wine", Saturday) and a no-guarantee warning. The second button was recorded as loose.
Differences: the customer says the coat is ruined, but our note has nothing on the mark's current size. The note says the button was loose, not missing.
Unknown: whether the button came off in cleaning or afterwards; whether the mark has faded; the price and age, which are the customer's statements.
To gather: a photo of the lapel now; a search of the workroom for the button; the intake photo, if one was taken.
Not concluded: fault.
That list does the sorting a person would do on the back of an envelope, and it points at the workroom instead of at the customer. Next, put the claim in a bucket. Every claim belongs in at least one of four, and the bucket decides which evidence matters:
| Bucket | What it means | What to pull from the file |
|---|---|---|
| A. Our handling | Lost item, another customer's garment, press burn, wrong process, machine damage | Ticket trail, who handled it, process record, photos before and after |
| B. The garment itself | Dye that runs, weak fabric, a label that did not match the fibre, trims that fail | Photo of the label, fibre content, what else was in the load, a second opinion |
| C. Already there | Worn, faded, loose or marked when it came in, and written on the note | Intake note and photo, compared word for word. If it was never noted, the note cannot help, so look at A and B |
| D. A mark we warned about | Treated but not fully removed after a no-guarantee warning | Intake note, receipt line, what was tried |
The coat's lapel is bucket D. The button is bucket C, because it was noted as loose, with a bucket A question hanging over it, since a button that came off in the machine is the shop's to put right. Buckets are questions, not rulings, and a claim can sit in two of them. When the file cannot settle one, such as colour loss on a dark garment or fabric that split along a seam, get a second opinion before you write. Some trade associations run a garment analysis service, so ask yours what it offers and what it costs.
The settlement sheet: you decide, then AI writes
Replies go wrong when the assistant is asked to decide and write in one step. Split the two. The owner fills in a short sheet, and the assistant sees only the sheet, so it cannot be generous or stingy on the owner's behalf. This is the sheet for the coat:
| Field | Entry |
|---|---|
| Claim reference | C-0412 |
| Garment | Navy wool coat |
| Customer says | Cost $300 three years ago; wants it replaced |
| Buckets | D (lapel mark); C (button loose at drop-off); A question (button now missing) |
| Findings | Lapel mark treated as warned and faded, faint outline remains. Button found in the workroom at 14:10. |
| Decision | No compensation for the coat. Goodwill offer. |
| Offer | Sew the button back on free; one further lapel treatment free; refund the $18 cleaning charge. |
| Approved by | Owner, with date |
Goodwill bands let counter staff act at once on small things while money decisions stay with a person. These are illustrative, so set yours against your margins; the dry-cleaning price list tutorial covers checking the margin on each price.
| Who | May approve | Limit |
|---|---|---|
| Counter staff | Free re-press or re-clean; a button sewn back on; a credit | Up to $10 |
| Manager | Refund of the cleaning charge; a credit | Up to $40 |
| Owner | Anything involving the value of a garment; any figure over $40; any claim where the customer says they will take it further | Owner's decision |
The coat's offer costs $18 plus a few minutes at the bench, which a manager could approve, but the customer asked for $300, so the owner signs. When the shop is at fault, someone has to do arithmetic, and it should not be the assistant. Suppose a cotton jacket was scorched by the press and the customer says it cost $240 two and a half years ago. The owner uses a straight-line rule with an expected life of five years for this kind of jacket. Unused life is 2.5 of 5 years, which is 50%, so $120, plus the $16 cleaning charge back, making $136. That figure goes on the sheet before anyone asks for any wording.
A straight line is only an illustration. Published claims guides set their own expected lives and methods. The Fair Claims Guide for Consumer Textile Products is one, and the Drycleaning & Laundry Institute's own page describes it as a standard giving guidelines for determining liability when adjusting claims for textile products. Read the guide your trade body or insurer follows before fixing a rule of your own. Check your insurance policy, or ask your broker, whether the insurer wants to hear about a claim before you reply. Some policies ask you not to admit fault or agree a payment until they have.
Five replies, filled in
One prompt serves all five replies. Staff name the reply type in its last line, so the assistant never chooses it, and every draft stays inside the sheet.
You draft replies for a dry cleaner. I give you a settlement sheet; you write only what it supports.
1. Do not admit fault unless the sheet says DECISION: ACCEPT.
2. Do not name a cause unless the sheet states it.
3. Use only figures, dates and offers that appear on the sheet.
4. Do not mention laws, policies, industry standards or guarantees.
5. Warm, plain and short: under 140 words, one apology at most, no blame.
6. Start with [first name]; end with [your name] and the shop name.
Write the reply type I name: ACK, ACCEPT, DECLINE, STAIN or MISSING.ACK, the same-day acknowledgement. It promises a time and nothing else.
Hello [first name], thank you for telling us about your coat. I'm sorry it isn't how you hoped to find it. I'm looking at the note we made at drop-off now. Could you send a photo of the lapel and of the place where the button was, or bring the coat in with your receipt? I'll come back to you by 5pm tomorrow with what we've found. Kind regards, [your name], [shop name]
ACCEPT, the shop is at fault. This is the scorched jacket, and the sheet carries the $136 figure.
Hello [first name], I'm sorry. When we examined the jacket we found it was scorched during pressing, and that is our responsibility. We'd like to settle this with $136: $120 towards the jacket, based on its age and condition, and a refund of the $16 cleaning charge. I can pay that to your card today or add it as credit, whichever you prefer. If you'd like to talk through how we reached the figure, please call me. Kind regards, [your name], [shop name]
DECLINE, the cause lies in the garment. A dark-green knitted dress lightened along the seams, the label said circle with P and the process matched. The sheet carries a $22 refund and an offer of independent analysis on request.
Hello [first name], I've examined your dress with its care label and our records. We cleaned it by the method the label allows, and I found no handling error on our side. In my view the lighter colour along the seams comes from the dye in the fabric, which can happen even when a label is followed. I know that's disappointing, and I'm sorry. As a gesture we'd like to refund the $22 cleaning charge. If you would like an independent view of the dress, we can arrange that, and I'm happy to talk it through. Kind regards, [your name], [shop name]
STAIN, a mark that was warned about. This is the coat, where the mark has partly lifted and the shop found the button.
Hello [first name], I've looked at the coat and at our drop-off note. The mark on the left lapel was recorded when you brought the coat in, and we told you then that removal couldn't be guaranteed. We treated it and it has faded, though a faint outline remains. We found the button in our workroom and will sew it back on at no charge. We'd also like to treat the lapel once more, free, and refund the $18 cleaning charge. If you'd like us to go ahead, reply here or call in whenever suits. Kind regards, [your name], [shop name]
MISSING, a garment the shop cannot find. The honest reply is a dated update.
Hello [first name], an update on your blouse: we haven't found it yet. We've checked the rails, the bags waiting for collection and the tickets either side of yours. I'm widening the search today and will write again by 5pm on Thursday. If we still haven't found it by then, I'll call you to talk about how to put this right. Kind regards, [your name], [shop name]
Counter staff also need one spoken line for a customer who is standing there: "I can see the note we made when it came in. May I photograph the coat now? [Owner's name] will look at it today and I'll phone you by 5pm tomorrow." Nothing else is promised, and nobody has to improvise. If the customer posts a public review, answer in two sentences that apologise for the disappointment, say you have written to them and give a way to reach you. Leave out the garment, the figure and the decision.
The promise check: what the draft added that nobody decided
Assistants try to reassure, and reassurance in a claim letter is where money leaks. This draft for the coat comes from a careless prompt that left out the rules (illustrative):
Hello [first name], I'm so sorry for the damage to your coat. We take full responsibility and guarantee we'll put this right. Our policy is to compensate fully for any damage caused in cleaning, so we'll refund the coat's value within 24 hours.
Five things went wrong in three sentences. "Full responsibility" admits fault the sheet does not. "Guarantee" is a promise. "Our policy" invents a rule that does not exist. "The coat's value" is a figure the sheet does not carry, and the opposite of the decision. "Within 24 hours" is a deadline nobody agreed. The draft reads well, which is the danger. Read every draft against the sheet with six questions:
- Does it admit fault anywhere the sheet says decline, or says nothing?
- Does it use "guarantee", "always", "never" or "full"?
- Is every figure on the sheet, to the cent?
- Is every deadline one you can keep?
- Does it quote a policy, a law or an industry standard you have not confirmed?
- Does it name a cause the owner has not concluded?
The tutorial on catching promises you never made to customers covers that last pass in more detail, and the tutorial on whether AI can handle complaints covers the holding reply, the one message automation may send.
What to paste into the chat window, and what to keep out
Paste the garment type, the ticket note, the customer's message with the name and contact details removed, and the sheet. Keep out the surname, phone number, email, address, payment details and any loyalty number. The ticket number is enough to link a draft to the file. The shop's till or customer system holds the contact details, and the draft needs only a first name, which you add when you send.
Photos are your evidence, so take them at the counter: one wide shot, one close-up of each mark, one of the care label, with the ticket in the frame. An assistant can describe a photo, but treat that description as a second pair of eyes and never as proof. If it mentions something your note does not, that is a question for a person. If it misses something your note records, believe your note, because a flat or dim photo can hide a mark the person saw.
On privacy, consumer ChatGPT and Claude accounts let you switch off the use of your chats for model training in the privacy settings, so check yours. Business plans, ChatGPT Business or Claude Team, do not train on business content by default, but both need two seats, which at list prices comes to about $40 a month on annual billing. Confirm the current terms on the vendor's pricing page before you buy.
A claims log that shows where garments get hurt
Keep one line per claim in a spreadsheet, filled in from the sheet. Names stay out, and the bucket letter does most of the work later:
| Date | Garment type | Bucket | Cause noted | Cost to shop | Approved by |
|---|---|---|---|---|---|
| 3 Oct | Wool coat | D, C | Mark warned; button loose at drop-off | $18 | Owner |
Every quarter, paste the log and ask the assistant to group it by garment type and bucket, list repeated causes and suggest three process checks, without naming any member of staff. A quarter's log might hold 26 claims: 9 on beaded or trimmed items, 7 on colour loss, 5 on buttons, 3 mix-ups and 2 others. The biggest group, trims, points to the first fix: a trim warning on the receipt and a tick box at intake. Colour loss and buttons point to two more, a dark-dye flag at intake and a button check at the pressing stage. None of these fixes would have been visible from any single claim. The tutorial on spotting patterns in complaints, returns and defects covers the same pattern-spotting more widely.
An afternoon to set up, four numbers to judge it by
A free account is enough to try this on a handful of claims a week, with the three prompts (counter note, claim comparison, reply) saved in a ChatGPT Project, or kept in a shared document and pasted in if you use Claude's free plan. Check the usage limits on whichever plan you choose. A paid plan, such as ChatGPT Plus or Claude Pro at about $20 a month list price, raises the usage limits. Allow an afternoon to write the sheet, the bands and the three prompts, and test them on five old claims before a live one. Do notes in the first two weeks and replies in the third, so staff learn one habit at a time.
Judge it by what the file shows after a month. Four numbers are enough: the time from claim to first reply, which should be the same day; how many claims had an intake note that matched what the customer described; how many claims were reopened after a reply; and the goodwill spent, which is the count of gestures times their average. Take 7 claims in a month, 4 of them settled with a goodwill gesture averaging $18: that is $72 for the month, and the sheet shows exactly who approved it.
Questions dry cleaners ask about AI and claims
What should happen to the garment while a claim is open?
I'd keep it as it came back, tagged and bagged, and photograph it before anyone touches it. Do not re-clean, press or repair it until the owner has decided and the customer has agreed, because a second treatment changes the evidence. If the customer takes it away, note the date and its condition in the file.
Can counter staff use AI while a customer is waiting?
For dictating a condition note, yes: a 30-second voice note takes about as long as writing one by hand. Live claim conversations are different. Staff should use the one-line spoken script, photograph the garment and leave the wording to the owner, who drafts the reply from the settlement sheet later the same day.
Should replies say that AI helped write them?
You don't need a line in every reply, but never present a draft as something you wrote personally if you did not read it. Read and edit every reply and send it under your name, because you are responsible for it either way. If a customer asks, say so plainly, and cover the general case with a line in your terms or on your website.
Would the same prompts work for a laundry or an alterations service?
Mostly. The note order (where, what, promise), the four buckets and the settlement sheet suit any service where you hold a customer's item. Change the bucket examples, the care-label wording and the goodwill bands to suit your own work, then test the replies on five old claims before using them for real.
Further reads
- AI in Delivery and Courier Services: Handling Damage Claims and Disputes — Damage claims in another business that holds customers' goods.
- Hotel Guest Complaints Handling: Service Recovery Scripts With AI — Service-recovery scripts with goodwill limits, from hotels.
- Should You Tell Customers You Use AI? A Disclosure Guide — How to word an AI disclosure for customers.
- How to Manage an Unattended Laundromat: Can AI Handle Fault Calls? — Refund and fault calls at a laundromat.
- AI App for Tailors: What Measurement and Fitting Apps Can Do — Another garment business where notes and fittings matter.
- Shoe Repair Software: Job Tickets, Pick-Up Texts and AI Replies — The ticket fields, quote texts and unclaimed-item wording a shoe repair shop needs, and where AI can draft the replies.
- AI for Laundromats: Upsell Wash-Dry-Fold From Your Transaction Data — Export your payment data, group customers by load size, price a bag and its delivery, then text a wash-dry-fold offer and test it against a hold-back group.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Drycleaning & Laundry Institute pages on home stain removal and on the Fair Claims Guide for Consumer Textile Products; Wikipedia's laundry-symbol article for care-label meanings; ChatGPT and Claude plan and privacy pages (checked September 2026). Every garment, customer, figure and reply in the examples is invented for illustration.