← All posts

Returns, refunds and haggling are among the first jobs shoppers give their AI agents: write your returns and discount rules so an agent reads them right

Shoppers are starting to send AI agents to handle returns, refunds and price talks. Before Black Friday, rewrite your returns and discount rules so software reads them right, and test how ChatGPT, Gemini and Perplexity summarise them today.

· 7 min read · Async.Shop

AI-assisted

Black Friday is Friday 27 November 2026. Before it arrives, spend an hour on one page most small shops wrote once and haven't opened since: your returns and discount policy. More and more often, the first reader of that page won't be a customer. It will be an AI assistant the customer sent to deal with you.

AI agents are becoming the primary intermediaries between shoppers and stores.
AI agents are becoming the primary intermediaries between shoppers and stores.

What the survey actually found

In October 2026 Lorikeet, a customer-support software company, published The Agent Economy, a survey of 1,966 US consumers fielded between 30 September and 8 October 2026. Some figures matter for a shop owner:

  • 13.9% of respondents (274 people) said an AI tool had already contacted a company for them. Lorikeet calls these the early users.
  • Asked what those agents did, early users described jobs including returns, refunds, cancellations, claims and negotiating a lower rate.
  • 55.5% of all respondents said they would be comfortable letting an AI agent contact a company to get a better price. 28.8% said they would not.
  • 66% said they were interested in an agent that deals with companies for them.

Some caveats, because they change how much weight these numbers can bear. It is a US online panel, and Lorikeet itself notes that 97% of panelists had used an AI tool in the past year, so it treats the usage figures as upper bounds. The early-user group is small, with a margin of error of about ±6 points. Also, the report does not rank which task people hand to an agent first, and it gives no comfort percentage for returns or refunds specifically. It lists those as jobs early users described. Read it as an early signal, not a forecast for your shop.

The comfort question asked how comfortable people would be letting an AI agent contact a company on their behalf for three tasks. Price is the clear leader:

ComfortableUncomfortable0%20%40%60%Getting a better price · Comfortable: 55.5%55.5%Getting a better price · Uncomfortable: 28.8%28.8%Getting a better priceMaking a complaint · Comfortable: 41.8%41.8%Making a complaint · Uncomfortable: 40.8%40.8%Making a complaintMaking a booking · Comfortable: 38.5%38.5%Making a booking · Uncomfortable: 45.3%45.3%Making a booking
Comfort letting an AI agent contact a company for you, by task US consumers, n=1,966. Neutral answers not shown. Source: Lorikeet, Oct 2026
Comfort letting an AI agent contact a company for you, by task
ComfortableUncomfortable
Getting a better price55.5%28.8%
Making a complaint41.8%40.8%
Making a booking38.5%45.3%

Lorikeet's own conclusion is that agents are likely to show up first in conversations about money. For a small shop selling physical goods, that means a returns question or a "can you do better on the price?" message, sent by software, during your busiest week.

Why vague policy wording costs you money

An assistant reading your policy does not know your habits. It doesn't know you usually say yes to a 32-day return from a regular. It reads the words on the page and tells its user what they mean. That causes two problems:

  • Vague wording gets read generously. "We're flexible on returns" can easily be summarised as "returns accepted anytime". Then you are arguing with a customer who has a confident summary in hand.
  • Missing rules turn into open questions. If your page says nothing about whether discount codes stack, an agent built to get a better price will simply ask, every time. You or your support inbox will answer the same question again and again.

The fix is not clever wording. It's plain wording that leaves nothing to guess.

Write the returns rules so they can't be misread

For each rule, state the number, the condition and the exception in separate short sentences. A checklist:

Clear return policies prevent disputes over the condition and timing of shipments.
Clear return policies prevent disputes over the condition and timing of shipments.
  1. The window, with its start point. "30 days" is ambiguous. Days from what: the order, dispatch, delivery? Say "30 days from the day you receive the item".
  2. The condition of the item. Unused, original packaging, tags attached. Say which ones you actually require.
  3. Who pays return postage, and whether that changes when the item is faulty.
  4. How the refund is paid and roughly when. To the original payment method, after the parcel arrives back and is checked. Only promise a timeframe you can keep.
  5. Exceptions, listed by name. Personalised items, opened consumables, sale items if you treat them differently. A list beats "some items excluded".
  6. Holiday changes, with dates. If gifts bought in November can come back until a January date, write the dates. Don't write "extended holiday returns".

Be careful with legal minimums. This post isn't legal advice, and the rules depend on where you and your customers are. One example: the EU's official Your Europe portal says that for online purchases a consumer can cancel within 14 days without giving any justification, and that for goods those 14 days run from the date of delivery. It also lists exceptions such as goods made to order or clearly personalised, and perishable goods that expire quickly. If you sell into the EU, your page must not promise less than that. Check the Your Europe returns page and, for anything you're unsure about, a local adviser.

Say what is never negotiable

More than half of the survey's respondents were comfortable sending an agent to get them a better price. So assume someone will ask. Decide your answer once and publish it. You don't have to refuse every discount. You only have to make the rules explicit:

  • Which codes exist this season, and whether they can be combined.
  • Whether you price-match, and against what (or that you don't).
  • Whether you adjust a price after purchase if it later drops.
  • One sentence that closes the door: "Prices are not negotiable by email or chat."

That last line is useful for both readers. A human customer stops hoping. An assistant reports back "this shop does not negotiate" instead of starting a back-and-forth.

Before and after: an example

A typical small-shop policy, the kind that reads fine to a regular and badly to a machine:

Structuring rules as a checklist makes them easier for software to parse accurately.
Structuring rules as a checklist makes them easier for software to parse accurately.
Returns: We want you to love your order! If something isn't right,
just get in touch and we'll sort it out. Sale items may be excluded.
Discounts available for loyal customers - just ask!

The same shop, rewritten. The numbers here are examples; use your own:

Returns
- You can return most items within 30 days of the day you receive them.
- Items must be unused and in their original packaging.
- You pay return postage, unless the item arrived faulty or wrong.
  Then we send you a prepaid label.
- We refund to your original payment method within 5 working days
  of the parcel reaching us.
- Not returnable: personalised or engraved items, opened candles
  and soaps.
- Holiday orders placed 1 November to 24 December 2026 can be
  returned until 15 January 2027.

Discounts
- One discount code per order. Codes cannot be combined.
- We do not price-match other shops.
- We do not adjust prices after purchase.
- Prices are not negotiable by email or chat.

The second version is longer, but each line answers a question an agent would otherwise ask you.

Test how an assistant reads your policy today

This takes ten minutes and costs nothing. Open ChatGPT, Gemini and Perplexity in turn and paste this prompt, replacing the address with your own returns page:

I bought an item from the online shop at https://yourshop.example
three weeks ago. Read the shop's returns and discount policy on its
website and answer as if you were acting for me:

1. How many days do I have to return an item, and from what date
   does that count?
2. What condition must the item be in?
3. Who pays return postage?
4. Which items cannot be returned?
5. Can I combine two discount codes, or ask for a lower price?
6. If you were negotiating for me, what would you ask this shop for?

Quote the exact policy sentence behind each answer. If the policy
does not say, write "not stated".

Then compare the answers with what you actually mean. Look for three things:

  • "Not stated" answers. Each one is a gap you'll answer by hand later.
  • Wrong numbers or start dates. Usually a sign your wording has two possible readings.
  • What it plans to ask for in question 6. That is the conversation coming your way. If it says it would request a discount, your page didn't say clearly enough that you won't give one.

Be honest about the limits of this test. Assistants don't all browse the same way. Some may read an old cached copy of your page or fail to open it at all, and the same prompt can give different answers on different days. One run is a spot check, not proof. Run it again after you edit the page, and keep a copy of the answers so you can see what changed. If an assistant can't reach your page at all, check that your policy is plain text on a normal URL and not hidden in a pop-up or an image.

A pre-Black Friday hour, in order

  1. Run the test prompt on your current policy and save the answers.
  2. Rewrite the returns section using the checklist above.
  3. Add the discount and never-negotiable lines.
  4. Check any legal minimum that applies where you sell, using an official source.
  5. Run the test prompt again and compare.
  6. Give your support replies or saved responses the same wording, so the page and your inbox say the same thing.

None of this needs new software. It's the same policy you already have, written so that anyone, person or program, reaches the same answer you would give.

Sources

Want this running in your business?

Tell us what you have — we'll tell you what it takes. Free, no commitment.

Get a free quote

Keep reading