Compare-At Prices and AI Agents: Is Your Discount Real?

Compare-At Prices and AI Agents: When Your Discount Gets Quoted

Compare-at prices used to be a storefront detail: a struck-through number next to the current price, designed to make a deal feel like a deal. They are about to travel much further. In October, Shopify announced that its catalog can now share a product's compare-at price with AI shopping agents, which previously received only the current price. Merchants control it with a toggle, and Shopify is clear that sharing the data does not guarantee how an agent will display or rank the product. The timing matters. Bain's holiday forecast, as reported, suggests close to a quarter of US online shoppers plan to start their holiday shopping on an AI platform. An agent that tells a shopper "this is 40 percent off" is repeating your claim, and reference pricing rules in most major markets already say that claim has to be true.

How a Discount Claim Travels

The same compare-at price carries different risks depending on who reads it and how it is presented.

What Changed With Agent Pricing Data

Until recently, many AI shopping agents saw a product's current price and little else. That made them blunt instruments for deal hunting. They could compare prices across stores, but they could not tell a shopper that a product was discounted, because they did not know what it was discounted from. Passing the compare-at price closes that gap. For merchants, the appeal is obvious: agents that recognise a saving may favour the product when a shopper asks for deals.

The risk is equally obvious once you say it out loud. Compare-at prices are among the most abused fields in e-commerce. Some merchants set them high and leave them there permanently, so every product looks discounted all year. On a storefront, that practice already draws regulatory attention. Repeated by an assistant to a shopper actively looking for a bargain, it becomes a more prominent and more scalable claim.

What Reference Pricing Rules Require for Compare-At Prices

The details differ by country, but the principle is consistent: a former price must be a genuine price at which the product was actually offered, for a meaningful period, recently.

Compare-At Prices Under EU, UK, and US Rules

In the European Union, rules introduced by the consumer protection Omnibus Directive require any announced price reduction to show the prior price as the lowest price charged in the 30 days before the reduction. In the UK, misleading pricing practices are covered by consumer protection law, now enforced more directly under recent legislation, with regulator guidance on reference prices. In the US, federal guides on deceptive pricing say a former price comparison should reflect a price at which the item was openly offered in good faith for a reasonable period, and some states add their own tests on how recent that price must be. None of these rules were written for AI agents. All of them apply to the claim the agent repeats.

Where Discount Claims Go Wrong

In our view the problems that bite tend to be ordinary rather than fraudulent:

• Permanent compare-at prices. A was-price set at launch and never updated, so the product has never sold at that price.
• Imported list prices. A manufacturer's recommended price used as the comparison when the merchant never charged it.
• Stale data after a sale ends. The sale price becomes the normal price, but the compare-at field remains.
• Market mismatch. A comparison that is valid in one country is shown to shoppers in another with stricter rules.
• Bulk edits. Apps or scripts that set compare-at prices as a fixed percentage above current price across a whole catalog.
• Variant confusion. A compare-at price from one size or colour applied to another that never carried it.

A Compare-At Price Audit for Sellers

1. Export every product with a compare-at price. Include variants and markets, not just headline products.
2. Check price history. Confirm each compare-at price was actually charged, and when, using order or price logs.
3. Apply the strictest market's rule. If you sell into the EU, test against the lowest price in the prior 30 days.
4. Remove permanent and imported comparisons. Clear any compare-at price you cannot evidence.
5. Automate expiry. Set compare-at prices to clear when a promotion ends rather than relying on memory.
6. Decide what to share with agents. Only enable agent sharing once the data would survive a regulator's question.
7. Keep the evidence. Store price history so you can show what was charged if a claim is challenged.

Pros and Cons of Sharing Compare-At Prices With Agents

• Pro: deals become visible. Genuine promotions can be surfaced to shoppers actively looking for savings.
• Pro: fairer comparison. Agents can distinguish a real reduction from a merely low everyday price.
• Pro: holiday reach. With more shoppers starting on AI platforms, discoverability during peak season matters.
• Con: amplified exposure. A misleading was-price reaches far more shoppers when an agent repeats it.
• Con: stale copies. An agent may keep showing a saving after the promotion has ended.
• Con: less control. You cannot fully control how an agent words or ranks the saving.

Real Scenarios Worth Thinking Through

These scenarios are illustrative, showing how compare-at prices play out in practice rather than presented as verified case studies.

A homeware store enables agent sharing ahead of the holidays. An audit afterwards finds that a third of its compare-at prices were set at launch and never charged. An assistant had been describing those products as heavily discounted. The store clears the fields, and its genuine seasonal promotions now stand out more clearly against the noise.

An electronics seller selling across Europe runs a weekend promotion. Its compare-at prices reflect a price charged six weeks earlier rather than the lowest price in the previous 30 days. A competitor complaint follows. A simple automated check against price history would have prevented it.

A fashion brand ends a sale but forgets to clear compare-at prices on a few variants. Its own site is fixed within a day, but an agent continues showing the old saving for longer. The brand adds expiry rules so the data clears at the source.

“After testing multiple AI models on coding, research, and business prompts, combined outputs produced more reliable results than any single model.” Internal multi-model evaluation, Talkory research team.

Pricing Data Is Now Agent Data

Compare-at prices are one example of a wider shift. Every field in your product data, price, stock, delivery time, returns policy, is becoming input for agents that speak to shoppers on your behalf. That raises the standard for accuracy across the whole catalog. We covered the broader mechanics of selling to agents in agentic commerce.

Pricing also attracts scrutiny from another direction. Regulators are increasingly interested in how algorithms set and personalise prices, a topic we explored in surveillance pricing. A merchant whose discount claims are clean and evidenced is in a far better position as both conversations develop.

Why Talkory Wins

Pricing rules vary by market and the wording matters, which makes them a natural fit for checking across several models. Talkory runs the same question across GPT, Claude, Gemini, Grok, Perplexity Sonar, and Kimi K3 in one pass. A seller can ask whether a specific was-now comparison is likely to meet the rules in the EU, UK, or a US state, and see whether six independent models agree. Agreement suggests a mainstream reading. Disagreement highlights the markets and claims where the seller should check the actual guidance or ask an adviser before turning on agent sharing.

Final Verdict

Compare-at prices are moving from your storefront into the answers AI shopping agents give shoppers, and that raises the stakes on a field many merchants treat casually. The rules have not changed: a former price must be genuine, recent, and evidenced, and in the EU it must reflect the lowest price of the previous 30 days. What has changed is reach. Audit every compare-at price, remove anything you cannot prove, automate expiry, apply the strictest market's rule, keep your price history, and only then share the data with agents. A real discount is a selling point. A fake one, repeated by an assistant, is a liability at scale.