Product Descriptions That Win Disputes — And Why AI Agents Make This Urgent

This is article 3 in our Agentic Commerce series. Article 1 covered the chargeback rules gap. Article 2 covered how to build Terms of Service that hold up as dispute evidence. This article covers a third layer of defense that most merchants overlook entirely: the product description itself.

The "Not as Described" Problem Is Bigger Than You Think

Ask most merchants which chargeback reason code worries them most and they'll say fraud. That's understandable—fraud chargebacks feel like an external attack. But industry data suggests that "not as described" disputes are responsible for somewhere between 20% and 50% of chargebacks depending on the product category. For merchants selling physical goods, apparel, electronics, or specialty items, that number sits toward the high end.

The reason this category is so large is structural. "Not as described" is one of the easiest disputes to file and one of the hardest to definitively disprove. A cardholder who claims a product didn't match the listing puts the burden squarely on the merchant to prove otherwise—with written evidence, in a 20 to 45 day window, to a reviewer who wasn't there.

That 53% win rate on physical goods disputes sounds mediocre, but the gap between merchants who win and those who lose often comes down to a single factor: whether their product page clearly and specifically described what was sold. The merchants who lose "not as described" disputes frequently cannot produce contemporaneous evidence showing exactly what the listing said at checkout.

The Reason Code Landscape: What You're Actually Fighting

Each card network handles "not as described" under a different reason code, and each has different evidence requirements and response deadlines. Understanding the specific code on your chargeback changes what you need to submit.

All four codes share a common evidentiary thread: to win, you need to demonstrate that what was shipped matched what was described at the time of sale. Not what your listing says today. What it said when the customer placed the order.

This distinction matters more than most merchants realize. If you have updated your product descriptions since the purchase date—even a minor edit to clarify something—a chargeback reviewer comparing your current listing to the dispute claim may see a discrepancy that wasn't there at the time of sale. Versioned evidence is critical.

What Makes a Product Description "Chargeback-Proof"

A chargeback-proof description is not about length or marketing copy. It's about precision, specificity, and the ability to be compared against a physical product later. Here is what that looks like in practice.

1. Specify Every Material Attribute

The word "material" has a legal meaning in dispute contexts: an attribute is material if a reasonable buyer would consider it important when making a purchase decision. For physical goods, material attributes typically include:

• Dimensions: Height, width, depth, and weight with units clearly stated (not "large" or "compact")
• Materials: Exact composition, not just "premium quality" or "durable construction"
• Color: Specific named color with hex or Pantone reference where relevant
• Compatibility: Exact model numbers, operating systems, or standards supported
• Contents: Exactly what is included in the package and what is not
• Condition: New, refurbished, open-box, or used—unambiguously stated
• Country of origin: Where regulations or buyer expectations may make this relevant

If a buyer receives a product and files a dispute claiming it wasn't as described, they will identify a specific attribute that differed. Your job is to ensure your listing already addressed that attribute with enough specificity that the description clearly covered it.

2. Use Comparables and Exclusions

One of the most effective techniques for "not as described" defense is explicit exclusion language. If your product is similar to a well-known item but differs in a key way, say so directly. If something commonly assumed to be included is not included, list it explicitly as excluded.

This is especially important for:

• Accessories that are commonly bundled but sold separately in your listing
• Software licenses or subscriptions that do not transfer with a product resale
• Regional or voltage variations that affect usability
• Products where "standard" configurations vary widely by market

3. Lock Your Descriptions at Order Time

Your order confirmation email should include a summary of the product attributes the customer purchased, not just a name and SKU. This creates a timestamped, customer-acknowledged record of the description at the moment of purchase.

At minimum, include in your order confirmation:

• Product name and full SKU or variant ID
• Key specifications (size, color, model, configuration)
• A permalink to the product listing as it existed at checkout (if technically feasible)
• The purchase price and what it covers

Combined with a screenshot archive of your product pages (timestamped via your content management system's version history or a third-party archiving tool), you can present a complete picture of exactly what was offered and when.

4. Align Photos and Text

Disputes frequently arise from a mismatch between product photos and text descriptions. A photo showing a larger version of a product than what's actually for sale, a lifestyle image that implies a use case the product doesn't support, or a color that photographs differently than it appears in person can all generate legitimate "not as described" claims.

Before dispute season arrives, audit your product pages for photo-text alignment. If your listing photo shows a product in a context that implies features or accessories not included, add a caption or note clarifying what is and is not depicted.

The AI Agent Amplifier: Why This Is More Urgent Than It Was Two Years Ago

Everything above applies to human buyers. Human buyers read descriptions, look at photos, and can ask questions before purchasing. When a description is vague or ambiguous, a human buyer may notice the problem and either clarify or abandon the cart. The vagueness causes friction, but it doesn't necessarily cause a sale.

AI shopping agents work differently. When an AI agent is tasked with finding and purchasing "a blue wool sweater in size M for under $150," it will interpret your product listing literally and programmatically. It will not notice the ambiguity in your color description. It will not pick up on the implication in your photo. It will parse your text and structured data, make a confidence determination, and either execute the purchase or move on.

This creates a new failure mode: an AI agent purchasing the wrong product from a correct description of an ambiguous listing, followed by a human cardholder disputing the charge because "this isn't what I wanted."

This isn't speculative. We covered the broader picture in our anchor article on agentic commerce chargebacks: AI agents are actively purchasing goods and services today, card networks have not yet built dispute rules that account for agent-initiated transactions, and the merchants who prepare now will be in a substantially stronger position than those who wait.

The connection to product descriptions is direct. A description that is clear enough for an AI agent to parse accurately is also clear enough to win a "not as described" dispute. The work is the same work—the stakes have just doubled.

Structured Data: The Bridge Between SEO, AI Agents, and Dispute Defense

Structured data—specifically JSON-LD schema markup using the Product schema type—is the mechanism that makes your product attributes machine-readable in a standardized format. It's the same markup that powers Google Shopping listings, enables Perplexity's product cards, and feeds AI agent shopping tools that retrieve product information programmatically.

Google Shopping, Amazon Rufus, and Perplexity's shopping features all prioritize structured data over unstructured prose when parsing product information. Research suggests that AI systems extract product attributes from HTML tables approximately 40% more reliably than from paragraph descriptions. If your product attributes live only in marketing prose, an AI agent reading your page has a significantly higher chance of misinterpreting them.

Here is what a basic product schema looks like for chargeback defense purposes:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Merino Wool Crewneck Sweater",
  "description": "100% Merino wool crewneck sweater. Machine wash cold. Not suitable for dryer.",
  "sku": "MWS-NAVY-M",
  "color": "Navy Blue",
  "size": "M (Chest 38-40 inches, Length 27 inches)",
  "material": "100% Merino Wool",
  "brand": {"@type": "Brand", "name": "YourBrand"},
  "offers": {
    "@type": "Offer",
    "price": "89.00",
    "priceCurrency": "USD",
    "availability": "https://schema.org/InStock",
    "itemCondition": "https://schema.org/NewCondition"
  }
}
</script>

This markup does three things simultaneously:

1. Improves search visibility through Google Shopping rich results
2. Reduces AI agent misreads by providing structured, unambiguous attribute data
3. Creates a versioned, machine-readable record of what your listing said at a given point in time

Point three deserves emphasis. When a dispute is filed, being able to provide a timestamped export of your structured data—showing that your listing described the product as "Navy Blue, 100% Merino Wool, Size M (Chest 38-40 inches)"—is substantially more persuasive than a screenshot of marketing copy that says "deep ocean tones in a luxurious natural fiber."

For merchants building toward agentic commerce readiness, we recommend also reviewing Visa's Transaction Aggregation Protocol (TAP) as it develops—a framework that will require agent-initiated purchases to carry verified product descriptions as part of the transaction record itself. Getting your structured data right now positions you ahead of that requirement.

The Description Audit Framework

Use this framework to audit your existing product listings before your next high-dispute period (typically Q4 post-holiday returns season, and the spring returns surge in apparel).

Step 1: Pull Your Top Dispute SKUs

Look at the last 90 days of chargebacks. Which SKUs appear most frequently in "not as described" disputes? Start there. The Pareto principle applies—a small number of products typically account for a disproportionate share of description-related disputes.

Step 2: Compare Your Description to the Dispute Claim

For each disputed SKU, read the chargeback claim carefully. What specific attribute did the cardholder allege was different? Now find that attribute in your product listing. Is it there? Is it specific enough to resolve the discrepancy definitively? If not, you've found a gap.

Step 3: The "Stranger Test"

Give your product description to someone who has never seen the product. Ask them to describe it back to you. If they can't accurately describe the key attributes—dimensions, materials, color, compatibility—your description is not specific enough to survive a dispute review.

Step 4: The Machine-Readable Test

Run your product URL through Google's Rich Results Test or Schema.org's validator. Does your structured data include all material attributes? Is your itemCondition specified? Is your color attribute a specific named color rather than a vague descriptor? Is size expressed in actual measurements rather than abstract designations like "M" or "Large" with no reference dimensions?

Step 5: Lock and Archive

Once you've updated your listings, create a system for archiving versions. Your CMS version history, combined with a periodic export of your product data feed, creates an evidence trail that can be produced in a dispute response. If you're on Shopify, the built-in version history serves this purpose. On WooCommerce, consider a plugin that logs product description edits with timestamps.

What to Submit When the Dispute Arrives Anyway

Even with excellent descriptions, disputes will happen. When a Visa 13.3, Mastercard 4853, Amex C31, or Discover RG dispute lands, your response package should include:

1. Archived product description as of the purchase date — Screenshot with timestamp, CMS version history export, or Wayback Machine capture
2. Order confirmation email — Shows the specific SKU and attributes acknowledged by the buyer at checkout
3. Shipping and delivery confirmation — With photos if available for high-value items
4. Structured data export — A copy of your product schema as it existed at the time of sale
5. Customer communication log — Any emails, chat transcripts, or support tickets showing the buyer did not raise a description mismatch prior to filing
6. Refund policy — Demonstrating the buyer had an alternative remedy available and chose the chargeback route instead

If you have a solid Terms of Service with explicit product description acknowledgment, include that ToS acceptance record as well. When a buyer clicks through a ToS that acknowledges the product description, your evidentiary position on "not as described" is substantially stronger.

The Layered Defense

Product description quality is one layer of a complete agentic commerce chargeback defense. The full stack looks like this:

• Understanding the agentic commerce dispute gap — Know what the card networks don't cover yet
• Building ToS that hold up as dispute evidence — Explicit acceptance, audit trails, and clickwrap
• Precise product descriptions with structured data — This article
• Bot management that distinguishes malicious bots from legitimate AI agents — Protecting your checkout without blocking legitimate buyers

Each layer addresses a different failure point. ToS handles authorization disputes. Product descriptions handle "not as described" disputes. Bot management handles the fraud and enumeration risk that comes with automated traffic. None of them works in isolation—but together, they create a defense posture that substantially outperforms merchants who address only one or two layers.

Frequently Asked Questions