Why ChatGPT Ignores 80% of Ecommerce Products

Your ecommerce catalogue may contain 10,000 products.

But when a customer asks ChatGPT to recommend the best product for a specific requirement, only a small portion of that catalogue may be understandable, comparable, and trustworthy enough to appear.

The problem is usually not that ChatGPT has deliberately rejected the other products.

The problem is that those products are effectively invisible to AI.

Important: The “80%” in this title is a diagnostic illustration, not a universal statistic published by OpenAI. The actual percentage will vary by website, catalogue quality, product category, technical accessibility, and availability of structured product data.

Quick Answer

ChatGPT may ignore ecommerce products when it cannot confidently determine:

• What the product is
• Who manufactured it
• Which customer requirement it satisfies
• How it differs from similar products
• Whether its price and availability are current
• Whether the product page is accessible and trustworthy
• Whether identifiers, attributes, variants, and offers belong to the same product
• Whether the seller is an authoritative source

A product page can look acceptable to a human while remaining difficult for an AI system to interpret.

The solution is not simply to add more marketing copy.

Businesses need accurate product attributes, clear identifiers, valid structured data, accessible product pages, reliable commercial information, product feeds, and consistent information across every channel.

Executive Summary

AI-powered shopping is changing product discovery.

Customers can now describe what they need in natural language:

“Find a waterproof safety shoe with a composite toe, electrical hazard protection, a wide fit, and next-day delivery.”

An AI assistant must translate that request into product requirements, identify relevant products, compare their specifications, assess source reliability, and present a small number of recommendations.

Products with incomplete, inconsistent, unstructured, inaccessible, or outdated information are less likely to survive that process.

OpenAI allows merchants to provide structured product information so that product discovery experiences can use current pricing, availability, seller information, and product attributes.

For ecommerce businesses, the implication is clear:

Publishing a product online is no longer enough. The product must also be machine-understandable, commercially accurate, technically accessible, and trustworthy.

Why It Matters

Traditional ecommerce discovery often starts with keywords.

AI-powered discovery starts with customer intent.

A customer may not search for:

ABC WorkSafe Pro 4200

They may ask:

“Which lightweight safety shoe is suitable for warehouse employees who stand all day?”

The AI system must understand that the customer may value:

• Lightweight construction
• Cushioning
• Slip resistance
• Breathability
• Safety certification
• Toe protection
• Suitable usage environments
• Available sizes
• Delivery location
• Current stock

A product with only a title, generic description, image, and price provides very little evidence for such a recommendation.

A competitor with richer, structured, and verifiable product information is easier to understand and safer to recommend.

The Business Problem

Most ecommerce businesses optimise product pages for one or more of the following:

• Internal site search
• Traditional SEO
• Marketplace templates
• Paid advertising
• Human readability
• Basic conversion

AI product discovery introduces another requirement:

Can an AI system understand the product well enough to match it to a detailed customer request?

Many catalogues are not ready.

Common catalogue problems include:

• Supplier descriptions copied without standardisation
• Missing technical attributes
• Inconsistent units of measure
• Product titles containing internal abbreviations
• Missing GTINs, MPNs, SKUs, brands, or manufacturer names
• Variants mixed together on one page
• Prices available only after login
• Availability loaded through inaccessible scripts
• Important specifications stored only in PDFs
• Generic descriptions reused across hundreds of products
• Product schema that contradicts the visible page
• Images with meaningless filenames and no descriptive context
• Discontinued products remaining indexable
• Product pages blocked from AI search crawlers
• No structured merchant product feed

These weaknesses affect more than ChatGPT visibility.

They can also undermine site search, marketplaces, comparison engines, Google Shopping, SEO, feed quality, customer confidence, and conversion.

Real-World Example

Consider two ecommerce businesses selling the same type of industrial safety shoe.

Product A

Title: Work Shoe 4200

Description:
High-quality work shoe. Comfortable and durable. Available in multiple sizes. Contact us for more information.

Available information:

• Product code: 4200
• Price: Login to view
• One image
• No brand displayed
• No safety certification
• No toe-cap material
• No slip-resistance classification
• No width information
• No structured variant data
• No visible stock status

Product B

Title: ProGuard X420 Composite-Toe Waterproof Safety Shoe – Black

Description:
Lightweight waterproof safety shoe for warehousing, logistics, maintenance, and light industrial environments. Features a composite safety toe, slip-resistant sole, cushioned footbed, and electrical hazard protection.

Available information:

• Brand: ProGuard
• Manufacturer part number: X420-BLK
• GTIN: 05012345678903
• Colour: Black
• Upper material: Water-resistant microfibre
• Toe protection: Composite
• Safety standard: EN ISO 20345:2022 S3S
• Slip resistance: SR
• Electrical hazard protection: Yes
• Width: Wide fit
• Sizes: UK 6–13
• Product weight: 540 g per shoe
• Price: £74.99
• Availability: In stock
• Delivery information: Next-day delivery available
• Returns policy: 30 days
• Product and variant structured data
• Multiple images
• Technical datasheet
• Current product feed record

Now consider this customer request:

“Recommend a waterproof, non-metallic safety shoe for a warehouse supervisor who needs a wide fit and slip resistance.”

Product B provides evidence for nearly every requirement.

Product A does not.

Even when Product A might technically satisfy the customer’s needs, the AI assistant lacks enough reliable information to confirm the match.

The product is not necessarily poor.

Its digital representation is poor.

Technical Explanation

ChatGPT shopping and search experiences may use information from public websites, merchant metadata, third-party providers, and product data supplied directly by merchants.

Several technical layers therefore affect whether a product can be discovered and recommended.

1. The Product Must Be Discoverable

The product page must be accessible to the systems used for search and retrieval.

Common accessibility problems include:

• Blocking relevant crawlers in robots.txt
• Requiring authentication
• Serving critical information only after user interaction
• Returning inconsistent HTTP responses
• Using canonical tags that point to the wrong page
• Creating product pages with no internal links
• Loading essential content only through unsupported client-side behaviour
• Publishing important attributes only inside images or downloadable files

A product cannot be selected if it cannot first be reliably found and processed.

2. The Product Must Be Identifiable

AI systems need to distinguish one product from another.

Useful identifiers include:

• SKU
• Manufacturer part number
• GTIN
• EAN
• UPC
• ISBN, where relevant
• Brand
• Manufacturer
• Model number
• Variant ID
• Product-group ID

A product titled simply “Blue Drill” is ambiguous.

A product identified as “Makita DHP484Z 18V LXT Brushless Combi Drill – Body Only” is more precise.

Identifiers also help connect information from manufacturers, distributors, retailers, marketplaces, reviews, documentation, and product feeds.

3. The Product Must Be Classified Correctly

AI systems need to know what kind of product they are evaluating.

Weak classification creates uncertainty.

For example:

Tools > Other > Miscellaneous

provides much less meaning than:

Power Tools > Drills > Cordless Combination Drills

Good classification supports:

• Product-type recognition
• Relevant attribute selection
• Comparison between similar products
• Query matching
• Filtering
• Variant grouping
• Category-specific recommendations

A strong taxonomy is therefore not merely an internal PIM structure. It contributes to external product understanding.

4. Product Attributes Must Be Explicit

AI systems should not have to infer every critical product characteristic from marketing copy.

A description such as:

“Built for demanding professionals.”

does not clearly state:

• Voltage
• Battery type
• Chuck size
• Maximum torque
• Speed
• Weight
• Included accessories
• Warranty
• Material compatibility

Structured attributes make products easier to filter and compare.

{
  "voltage": "18 V",
  "batteryIncluded": false,
  "maximumTorque": "54 Nm",
  "chuckSize": "13 mm",
  "weight": "1.8 kg",
  "warranty": "3 years"
}

Product attributes are particularly important for long, specific, conversational queries.

5. Variants Must Be Unambiguous

Many ecommerce sites place every size, colour, material, capacity, or pack quantity on a single page.

This becomes problematic when:

• Variants share one SKU
• The selected variant does not update the URL
• Images do not correspond to the selected colour
• Price changes are not reflected in structured data
• Availability is shown only at the parent level
• Variant identifiers are missing
• The parent description contains conflicting values

Each variant should have a distinct and consistent identity.

6. Commercial Information Must Be Current

A product recommendation is only useful when the product can actually be purchased.

Important commercial fields include:

• Price
• Currency
• Availability
• Condition
• Seller
• Delivery options
• Shipping cost
• Returns policy
• Promotion
• Effective dates
• Geographic availability

When website content, schema markup, and feeds disagree, confidence may decline.

| Source | Price | Availability | |---|---:|---| | Visible product page | £79.99 | In stock | | JSON-LD | £69.99 | Out of stock | | Product feed | £74.99 | In stock | | Open Graph metadata | £69.99 | Not provided |

This is not simply a technical error.

It creates commercial uncertainty.

7. Structured Data Must Describe the Visible Product

Schema.org provides types such as Product, Offer, AggregateOffer, Review, and AggregateRating for describing products and commercial offers.

A basic product JSON-LD implementation may include:

{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "ProGuard X420 Composite-Toe Safety Shoe",
  "sku": "X420-BLK-09",
  "mpn": "X420-BLK",
  "gtin13": "5012345678903",
  "brand": {
    "@type": "Brand",
    "name": "ProGuard"
  },
  "image": [
    "https://example.com/images/x420-black-front.jpg",
    "https://example.com/images/x420-black-side.jpg"
  ],
  "description": "Waterproof composite-toe safety shoe with slip-resistant sole.",
  "color": "Black",
  "size": "UK 9",
  "offers": {
    "@type": "Offer",
    "price": "74.99",
    "priceCurrency": "GBP",
    "availability": "https://schema.org/InStock",
    "url": "https://example.com/products/proguard-x420-black-size-9"
  }
}

Structured data is not a substitute for visible product content.

It should accurately reflect what customers can see on the page.

8. Product Feeds Provide a More Direct Signal

A product page is designed primarily for people.

A product feed is designed for systems.

A feed can communicate:

• Product ID
• Title
• Description
• Product URL
• Image URL
• Brand
• GTIN
• MPN
• Category
• Price
• Currency
• Availability
• Condition
• Variant relationships
• Seller information
• Shipping information
• Promotions
• Additional product attributes

A feed does not fix poor source data.

It distributes source data more efficiently.

9. The Product Must Demonstrate Trust

AI product recommendations involve risk.

A weak recommendation can waste the customer’s money, present incorrect specifications, or suggest an unsuitable product.

Trust may be strengthened by:

• Consistent manufacturer and seller information
• Verifiable technical specifications
• Recognised product identifiers
• Genuine customer reviews
• Clear warranty information
• Returns and delivery policies
• Regulatory and safety documentation
• Accurate stock information
• Authoritative product documentation
• Consistency across the website, feeds, schema, and external sources

Trust is particularly important for:

• Healthcare products
• Electrical products
• Automotive components
• Industrial equipment
• Safety products
• Food and allergens
• Children’s products
• High-value purchases
• Products requiring compatibility checks

Best Practices

Build Product Information Around Customer Questions

Do not structure the catalogue only around what the ERP already contains.

Identify the questions customers ask before purchasing:

• Will this fit?
• Is it compatible?
• Is it suitable for outdoor use?
• Does it contain allergens?
• What certification does it meet?
• What is included?
• What is the operating temperature?
• Is it available in my size?
• Can it be delivered tomorrow?
• How does it compare with the alternative?

Convert those questions into governed product attributes.

Use Category-Specific Attribute Models

A universal set of generic attributes will not adequately describe every product.

A laptop needs attributes such as:

• Processor
• Memory
• Storage
• Screen size
• Resolution
• Graphics
• Battery life

A chemical product needs:

• Chemical composition
• Concentration
• Hazard classification
• Storage requirements
• Safety datasheet
• Compatible materials

The attribute model should reflect how customers evaluate products in each category.

Separate Facts From Marketing Content

Both are useful, but they serve different purposes.

Facts:

• 18 V
• 54 Nm torque
• 13 mm chuck
• 1.8 kg
• Brushless motor

Marketing content:

• Compact performance for demanding installation work
• Designed for professional use in confined spaces

The best product page contains both.

Maintain a Single Trusted Product Record

Product data should not be independently maintained in:

• ERP
• PIM
• Ecommerce platform
• Marketplace spreadsheets
• Product feeds
• Schema templates
• Marketing documents

Define which system owns each field.

| Information | Recommended System of Record | |---|---| | SKU and base product identity | ERP | | Taxonomy and attributes | PIM | | Enriched descriptions | PIM | | Images and documents | DAM or PIM | | Price | ERP or commerce platform | | Real-time stock | ERP, OMS, or commerce platform | | Product schema | Generated from trusted commerce data | | AI product feed | Generated from governed source systems |

Validate Data Across Every Surface

Compare the same product across:

• Visible PDP content
• HTML metadata
• JSON-LD
• Product feed
• Sitemap
• Marketplace feed
• Manufacturer page
• Search result
• Mobile page
• Variant URLs

Differences should be intentional and explainable.

Comparison Table

| Area | AI-Invisible Product | AI-Ready Product | |---|---|---| | Product title | Generic or abbreviated | Clear, descriptive, and specific | | Product identity | Internal SKU only | SKU, brand, MPN, and GTIN | | Classification | Broad miscellaneous category | Accurate product taxonomy | | Attributes | Embedded in description or missing | Structured and category-specific | | Description | Generic marketing language | Benefits, applications, and specifications | | Variants | Mixed or ambiguous | Unique variant IDs and URLs | | Price | Hidden, stale, or inconsistent | Current price and currency | | Availability | Missing or parent-level only | Variant-level stock status | | Images | One generic image | Multiple product-specific images | | Schema | Missing or invalid | Valid Product and Offer markup | | Product feed | Not available | Structured, maintained feed | | Delivery | Not stated | Shipping and delivery information | | Returns | Difficult to find | Clear returns policy | | Trust signals | Minimal | Reviews, warranty, documentation, certifications | | Crawlability | Blocked or script-dependent | Accessible and internally linked | | Data consistency | Conflicting channel values | Governed from a trusted source |

Common Mistakes

Mistake 1: Assuming SEO Automatically Means AI Readiness

A page may rank for a product name but still fail to answer a detailed product question.

SEO visibility and AI product understanding overlap, but they are not identical.

Mistake 2: Writing Longer Descriptions Instead of Better Data

More words do not automatically create more meaning.

A 1,000-word description can still omit voltage, dimensions, compatibility, certification, or material.

Mistake 3: Generating AI Content From Incomplete Attributes

AI-generated descriptions can make weak source data sound polished without making it accurate.

Generation should happen after product facts are validated.

Mistake 4: Treating Schema as a One-Time Development Task

Product schema changes when:

• Price changes
• Stock changes
• Variants are added
• Ratings change
• URLs change
• Products are discontinued
• Shipping policies change

Schema must remain connected to live, governed data.

Mistake 5: Using the Same Description for Every Variant

Variant-specific information should be explicit where it affects the buying decision.

Examples include:

• Colour
• Size
• Capacity
• Voltage
• Material
• Pack quantity
• Compatibility
• Availability

Mistake 6: Hiding Important Information Inside PDFs

Technical documents are useful, but critical buying information should also be available as structured page content.

Do not require an AI system—or the customer—to inspect a 40-page PDF to discover the operating temperature.

Mistake 7: Publishing Conflicting Information

A product cannot be confidently recommended when:

• The title says 500 ml
• The image shows 750 ml
• The specification table says 1 litre
• The schema says 500 ml
• The feed says 750 ml

Consistency is a trust signal.

Mistake 8: Ignoring Product Feeds

Businesses preparing for AI commerce should evaluate feed readiness rather than relying exclusively on webpage crawling.

Practical Checklist

Use this checklist to assess a sample of your most important products.

Product Identity

• [ ] Every product has a unique SKU
• [ ] Brand is clearly stated
• [ ] Manufacturer is identified
• [ ] MPN is provided where applicable
• [ ] GTIN, EAN, or UPC is provided where applicable
• [ ] Product and variant identifiers are consistent across systems

Product Classification

• [ ] Every product is assigned to a meaningful category
• [ ] Categories reflect how customers search and compare
• [ ] Category-specific attributes are available
• [ ] Products are not placed in unnecessary “Other” categories

Product Content

• [ ] Titles clearly identify the product
• [ ] Descriptions explain applications and benefits
• [ ] Important specifications are structured
• [ ] Units of measure are standardised
• [ ] Abbreviations are explained
• [ ] Content is not duplicated across unrelated products

Variants

• [ ] Each variant has a unique identifier
• [ ] Variant options are clearly labelled
• [ ] Price is accurate at variant level
• [ ] Availability is accurate at variant level
• [ ] Images match the selected variant
• [ ] Variant URLs are stable and accessible

Structured Data

• [ ] Product schema is present
• [ ] Offer information is present
• [ ] Price and currency match the visible page
• [ ] Availability matches the visible page
• [ ] Brand and identifiers are included
• [ ] Structured data passes validation
• [ ] Discontinued products are handled correctly

Technical Discoverability

• [ ] Product pages return successful HTTP responses
• [ ] Product pages are internally linked
• [ ] Canonical URLs are correct
• [ ] Product URLs are included in relevant sitemaps
• [ ] Important content is not restricted to inaccessible scripts
• [ ] Relevant AI search crawlers are not unintentionally blocked
• [ ] Product pages can be accessed without authentication

Trust and Commerce

• [ ] Shipping information is clear
• [ ] Returns information is clear
• [ ] Warranty information is available
• [ ] Reviews are genuine and product-specific
• [ ] Regulatory information is included where relevant
• [ ] Safety documents are current
• [ ] Seller and manufacturer information are consistent

Product Feeds

• [ ] A structured product feed is available
• [ ] Feed identifiers match product-page identifiers
• [ ] Price and availability are refreshed regularly
• [ ] Variant relationships are represented
• [ ] Feed records link to the correct product pages
• [ ] Feed errors and rejected records are monitored

How AI Commerce Readiness Measures This

The Ingwalabs AI Commerce Readiness Assessment evaluates whether product information is prepared for AI-powered search, shopping, and recommendation experiences.

The assessment reviews areas such as:

Product Data Completeness

Does the page provide sufficient product facts?

This can include:

• Brand
• SKU
• MPN
• GTIN
• Product type
• Technical attributes
• Dimensions
• Materials
• Compatibility
• Certifications

Content Quality

Does the product content explain what the product is, who it is for, and why someone should choose it?

The assessment looks beyond word count to evaluate clarity, specificity, and usefulness.

Structured Data

Does the page contain valid, consistent Product and Offer markup?

The assessment checks whether structured information agrees with visible product content.

Discoverability and SEO

Can search and AI systems find, access, interpret, and reference the product page?

Product Detail Page Quality

Does the PDP provide the commercial and technical information needed to support a buying decision?

AI Trust Signals

Does the product provide enough consistent evidence to support a reliable recommendation?

The result is an executive-level readiness score with practical recommendations for improving product visibility.

FAQs

Does ChatGPT really ignore exactly 80% of ecommerce products?

No universal figure has been published.

The percentage is used here to illustrate a common business problem: a large part of an ecommerce catalogue may be technically online but insufficiently structured, complete, accessible, current, or trustworthy for AI-driven product discovery.

The correct percentage must be calculated by auditing the organisation’s actual catalogue.

Can I guarantee that ChatGPT will recommend my products?

No.

No optimisation method can guarantee inclusion or ranking.

However, complete product data, accessible pages, accurate commercial information, structured feeds, valid schema, clear identifiers, and strong trust signals can make products easier to understand and evaluate.

Is product schema enough?

No.

Schema helps systems interpret product information, but it does not compensate for poor source data, inaccessible pages, missing attributes, incorrect variants, stale prices, or weak trust signals.

Do I need a PIM system?

Not every business requires a PIM immediately.

However, organisations with large catalogues, multiple suppliers, complex products, many channels, frequent updates, or extensive variants usually benefit from centralised product information governance.

What is the difference between a product feed and product schema?

Product schema is embedded in or associated with an individual product page.

A product feed provides structured records for many products and is usually refreshed on a scheduled basis.

Both should originate from consistent, governed product information.

Does blocking GPTBot prevent products from appearing in ChatGPT search?

OpenAI distinguishes between different crawlers and user agents.

GPTBot relates to potential model training, while search discovery can use different systems and user agents. Businesses should review current OpenAI crawler documentation and configure access according to their policies.

Are reviews important for AI product visibility?

Reviews can provide useful evidence of product quality and customer experience, but they are only one signal.

Reviews should be genuine, relevant to the specific product, and consistent with the visible page and structured data.

Which products should we optimise first?

Start with products that have the greatest commercial impact, such as:

• Highest-revenue products
• Highest-margin products
• Strategic brands
• Products with strong stock availability
• Products receiving high search demand
• Products with high abandonment rates
• Products with incomplete specifications
• Products frequently compared with competitors
• Products suitable for long-tail conversational searches

Key Takeaways

1. Being listed online does not make a product AI-ready.
2. ChatGPT needs enough evidence to identify, understand, compare, trust, and commercially validate a product.
3. Generic descriptions are not a substitute for structured attributes.
4. Product identity, taxonomy, specifications, variants, pricing, stock, schema, feeds, and trust signals must work together.
5. Conflicting data across product pages, structured data, and product feeds creates uncertainty.
6. Product feeds provide merchants with an increasingly important route into AI-powered shopping discovery.
7. AI-generated content should be built on validated product facts—not used to hide missing data.
8. AI commerce readiness is a product-data and governance challenge, not merely an SEO task.

Practical Recommendations

Begin with a representative sample of 20 to 50 commercially important products.

For each product:

1. Write three realistic conversational customer queries.
2. Identify the attributes required to answer those queries.
3. Compare the required attributes with the current PDP.
4. Validate product identity and variant relationships.
5. Compare visible content with JSON-LD and feed data.
6. Confirm price and availability accuracy.
7. Review crawlability and internal linking.
8. Assess supporting trust information.
9. Document the enrichment gaps.
10. Prioritise fixes according to revenue, demand, risk, and effort.

Once the method is proven, scale it across the catalogue through PIM governance, automated validation, enrichment workflows, and feed monitoring.

Call to Action

Your products may be available for purchase—but are they understandable enough for AI to recommend?

The Ingwalabs AI Commerce Readiness Assessment evaluates your product pages across product data, content quality, structured data, discoverability, PDP experience, and AI trust signals.

In approximately three minutes, you can receive an executive-level readiness score and practical recommendations for improving AI product visibility.