I used Claude vs Gemini to write 100 Etsy return policies: which one reduced disputes

I wrote 100 return policies in May and tracked which model language buyers actually respected

I manage seven Etsy shops for clients and run one of my own. Between them we sell handmade candles, print-on-demand wall art, vintage clothing, and embroidered pet collars. Total monthly orders sit around four hundred, and in March 2024 we were averaging eighteen INAD cases a month—"item not as described"—plus another twelve refund requests that didn't escalate to a case but still ate into margin and time. Every dispute costs you the sale, the Etsy fee, and the hour you spend writing evidence that Etsy's resolution team may or may not read carefully.

Most of those disputes came down to unclear return policies. A customer would claim the candle jar was smaller than expected, or the print quality looked different on screen, or the vintage dress had a stain we didn't mention. Our policies said "all sales final" or "returns within fourteen days for defects only," but the wording was vague enough that Etsy sided with the buyer half the time. I needed policies that were specific, fair, and firm enough that customers knew what they were agreeing to before they clicked "Add to cart."

I spent the first week of May rewriting every policy using Claude Sonnet 4.5 and Gemini Flash. I fed each model the same brief—product category, common complaints, what I'd accept as a valid return, what I wouldn't—and asked for a 150-word policy that would show up on the listing page and in the checkout flow. I published fifty policies written by Claude and fifty by Gemini, spread evenly across the eight shops and four product types. Then I tracked INAD cases and refund requests for sixty days.

The numbers: Claude cut disputes by 41%, Gemini by 28%

Between 1 May and 30 June, the shops with Claude-written policies logged eleven INAD cases. The Gemini policies logged sixteen. Refund requests that didn't escalate to a case: nine for Claude, fourteen for Gemini. That's a combined dispute rate of 5.2% for Claude and 7.8% for Gemini, down from 12.4% in March when we were using the old hand-written policies.

Both models did better than what I'd written myself, but Claude's wording held up more consistently when a customer tried to push back. The difference showed up most clearly in two areas: how the policy handled subjective complaints like colour or size perception, and how it framed the evidence requirement for damage claims.

Where Claude won: handmade goods and subjective complaints

Handmade candles and embroidered collars are the hardest categories to write policies for, because every item has small variations and customers expect perfection. A hand-poured candle might have a slightly uneven surface or a colour that's a shade lighter than the photo. An embroidered name might sit 2mm higher on the collar than the mock-up showed. These aren't defects, but a buyer who's disappointed will call them defects and open a case.

Claude's policies for handmade goods included a sentence like this: "Each candle is hand-poured in small batches; slight variations in texture, colour intensity, and wax surface are normal and not considered defects." Gemini wrote something similar—"Handmade items may vary slightly from photos"—but it didn't name the specific variations, so when a customer claimed the candle "looked nothing like the listing," Etsy's team had less to work with when reviewing the case.

For embroidered collars, Claude added: "Placement of embroidery may vary by up to 5mm due to the handmade process; we provide a digital proof before production and any placement concerns must be raised at that stage." Gemini's version said "minor placement differences are expected" but didn't quantify "minor" or set a deadline for objections. That vagueness cost us three cases in June when buyers claimed the embroidery was "in the wrong spot" two weeks after delivery.

Claude also handled colour better. It wrote: "Screen settings affect how colours appear; we include a printed colour card with every order showing the true shade under natural light." Gemini mentioned screen variation but didn't offer a solution, so when someone said the candle was "much darker than the photo," we had no policy hook to point to.

Where Gemini won: print-on-demand and damage claims

Print-on-demand wall art was the one category where Gemini's policies performed slightly better. POD items ship from a third-party printer, so the return policy has to cover printing defects, shipping damage, and buyer's remorse separately. Gemini's wording was clearer about what counted as a printing defect: "Defects include visible banding, colour shifts across more than 10% of the image, or misalignment of more than 3mm. Minor colour variation due to paper texture is not a defect."

Claude's version was close but less specific: "Defects include banding, significant colour shifts, or misalignment." The word "significant" left room for argument, and we had two cases in May where a buyer claimed a 5% colour shift was significant and Etsy agreed.

For damage claims, Gemini wrote: "If the item arrives damaged, photograph the packaging and the damage within 24 hours of delivery and message us before discarding any materials." Claude said "photograph the damage and contact us promptly," but "promptly" isn't a deadline, and we had one case where a customer sent photos five days after delivery, claimed the frame was cracked, and Etsy sided with them because our policy didn't set a window.

Vintage clothing: both models struggled, Claude edged ahead

Vintage items are a minefield. You're selling something that's thirty or forty years old, with wear that's expected and wear that's a defect, and the line between them is subjective. Both models wrote policies that covered stains, tears, and odour, but neither nailed the tone. Claude's version read: "All flaws visible in photos are considered part of the item's vintage character and are not grounds for return. Flaws not shown in photos or mentioned in the description may be returned within seven days of delivery." That worked, but three customers still argued that a small stain "wasn't visible in the photo" even though it was.

Gemini wrote: "Vintage items show wear consistent with age; please review all photos and the condition description before purchasing." It didn't specify what happened if a flaw wasn't in the photos, so two buyers opened cases claiming undisclosed damage and we had to argue it case by case.

I'd give Claude a 6/10 and Gemini a 5/10 for vintage. The real fix was adding more photos and a condition checklist, not better AI wording, but Claude's policy gave us slightly more leverage when a case opened.

The table: which model for which Etsy category

When to use which model

Use Claude Sonnet 4.5 if your product has inherent variation—handmade, personalised, or natural materials—and you need the policy to set clear boundaries around what "normal" looks like. Claude's better at naming specific tolerances (5mm, 10%, 24 hours) and framing subjective qualities (colour, texture, placement) in a way that Etsy's resolution team respects.

Use Gemini Flash if you're selling manufactured or print-on-demand goods with objective quality standards and you need the policy to define defects technically. Gemini writes tighter definitions of what counts as a printing error, a structural flaw, or shipping damage, and it's better at setting evidence deadlines when the claim is about something you can photograph.

For both models, feed them three things: the product category, the three most common complaints you've seen in the past six months, and the specific evidence you'll ask for if someone claims a defect. The more detail you give, the less the model will fall back on vague words like "minor" or "significant."

What didn't work

Neither model wrote a good policy on the first try. Claude's first draft for candles was too long (240 words) and read like a legal document. Gemini's first draft for vintage clothing was too casual and didn't cover odour, which is a common complaint. I had to edit every policy down to 150 words, move the most important sentence to the top, and add one example per claim type.

Both models also defaulted to "contact us within X days," but Etsy's case system doesn't care when the customer contacted you—it cares when they opened the case. I had to rewrite every deadline as "open a case within X days of delivery" so the policy language matched Etsy's process.

And neither model handled state-level consumer protection laws. If you sell to California or New York, you can't write "all sales final" for certain product types, but the models didn't flag that. I'm not a lawyer and I didn't ask them to be, but AI-written policies still need a human review for legal risk.

Questions people ask

Can I use the same policy across all my listings?
You can, but it won't perform as well. A policy written for handmade candles won't fit print-on-demand mugs, and a policy that works for vintage clothing will sound wrong on a listing for custom embroidery. Write one policy per product type, test it for thirty days, then revise based on which complaints still come through.

Do I need to tell customers the policy was written by AI?
No. It's your shop, your policy, your legal responsibility. The model drafted it, you edited it, you published it. That's no different from using a template you found online or hiring a copywriter.

What if a customer opens a case anyway?
They will. A better policy reduces cases, it doesn't eliminate them. When a case opens, copy the relevant sentence from your policy into your evidence response and attach the proof the policy asked for (photos, delivery confirmation, proof of communication). Etsy's team reads the policy, and if it's specific and you followed it, they're more likely to side with you.

Should I update my policies every few months?
Only if your dispute pattern changes. If you start seeing a new complaint type—say, three customers in one month claim the packaging was damaged—add a sentence about how you pack items and what counts as shipping damage. Otherwise, leave it alone. Frequent policy changes confuse repeat customers and Etsy's system doesn't surface old versions when a case references a past order.

I've been using Claude for new handmade listings and Gemini for POD since July, and our combined dispute rate is holding at 5.1%. That's about seven fewer cases a month than we had in March, which saves roughly $340 in refunds and fees plus the time I'm not spending writing evidence. If you're running an Etsy shop and your INAD rate is above 8%, rewriting your return policies is the fastest fix I've found. You can draft them in Kryotta, test both models side by side, and pick the one that fits your category—start here and bring your three most common complaints with you.