ChatGPT Just Told Us More About How Products Get Selected – Article and Video 63

What Changed?

OpenAI recently updated its first-party documentation explaining how products are selected in Shopping with ChatGPT Search.

The disclosure is unusually useful because it describes parts of the platform’s own product-selection process rather than relying on outside correlation studies.

According to OpenAI, a product may appear when ChatGPT perceives it as relevant to the user’s intent. ChatGPT can consider the user’s query and context, including Memory and Custom Instructions. Depending on the request, factors such as price, reviews, availability and ease of use can also matter. (OpenAI Help Center)

OpenAI then identifies three inputs ChatGPT considers when determining which products to surface:

• Structured metadata from first-party and third-party providers, including information such as price and product description.
• Model responses generated by ChatGPT before it considers new search results.
• OpenAI safety standards and product policies. (OpenAI Help Center)

The weighting is intent-dependent.

OpenAI gives a simple example: if a user explicitly specifies a $30 budget, ChatGPT can place more importance on price. (OpenAI Help Center)

The Most Important New Disclosure

The key line is that ChatGPT may consider:

“Model responses generated by ChatGPT before it considers any new search results.”

That is meaningful because it provides direct evidence that live retrieval is not necessarily the first point at which potential products enter the decision process. (OpenAI Help Center)

A simplistic model would look like this:

User query → web search → product ranking

OpenAI’s disclosure supports a more nuanced model:

User query + context → model-generated product understanding → current evidence/retrieval → product selection

The exact internal architecture is not public.

And this disclosure should not be interpreted as proof that products appearing more often in OpenAI’s training data automatically rank better.

OpenAI does not make that claim.

A safer interpretation is:

ChatGPT can begin live retrieval with an existing model-generated view of potentially relevant products or product characteristics.

The Three Ecommerce Competitions

This creates a useful framework for understanding ChatGPT commerce visibility.

These are separate problems.

An ecommerce company can win one and lose another.

Competition 1: Candidate Formation

Before a product can be recommended, it has to become a plausible candidate.

OpenAI’s documentation suggests that this may begin before ChatGPT examines new live search results.

For brands, this means the question is no longer only:

“Does ChatGPT find our page?”

It is also:

“Does ChatGPT perceive our product as a relevant solution to the user’s request?”

That puts more emphasis on clear product identity and semantic relevance.

A product should be consistently understandable as:

• What it is
• Who it is for
• Which problems it solves
• Which use cases it supports
• Which competing products or categories it belongs near
• Which specifications distinguish it

Competition 2: Product Selection

Once a product enters consideration, current evidence becomes critical.

OpenAI explicitly identifies structured metadata as an input and notes that factors including price, reviews and availability can matter depending on the user’s request. (OpenAI Help Center)

For an ecommerce SKU, the current information layer should include:

• Exact product name
• Brand
• Model
• Variant
• Product description
• Structured attributes
• Technical specifications
• Compatibility
• Current price
• Availability
• Reviews
• Ratings
• Practical use cases
• Clear differentiators

This isn’t about adding more words to a product page.

It is about reducing ambiguity.

The product should be easy for a machine to identify, compare and validate.

Competition 3: Merchant Selection

Winning product selection does not automatically mean the brand’s own store wins the transaction opportunity.

OpenAI says that after a product is selected, ChatGPT may show a list of merchants selling it.

Merchant rankings can consider factors including:

• Availability
• Price
• Quality
• Whether the merchant is the maker or primary seller of the product. (OpenAI Help Center)

That creates a distinct merchant optimization layer.

An ecommerce brand could have its product recommended while another retailer becomes the seller displayed to the shopper.

Shopify Has a Direct Integration Advantage

OpenAI says Shopify merchant product data is already integrated into ChatGPT through Shopify Catalog, helping products appear more accurately and completely in relevant conversations. OpenAI also allows merchants to apply to provide a direct product feed so product information can remain more current. (OpenAI Help Center)

That does not guarantee selection.

But it clearly makes product-data freshness an operational issue ecommerce teams should care about.

Shopping Research Adds a Second Retrieval System

OpenAI’s separate Shopping Research documentation adds another layer.

Once ChatGPT clarifies the user’s intent through follow-up questions, Shopping Research can begin a multi-step product discovery process.

OpenAI says that process may use:

• Merchant product data supplied through the Agentic Commerce Protocol
• Publicly available product information
• Other relevant retail sources. (OpenAI Help Center)

Users can refine their criteria while that research continues.

The final buyer’s guide may contain:

• Top product recommendations
• Reasons each product matches the request
• Tradeoffs
• Price
• Features
• Reviews
• Side-by-side comparisons
• Additional matching products. (OpenAI Help Center)

This means ecommerce visibility should not be treated as one static ranking event.

It can involve repeated discovery, evaluation and refinement.

Machine Access Is Now an Eligibility Issue

Product information that cannot be reliably accessed may be excluded from part of the shopping discovery process.

One of the most actionable statements in OpenAI’s Shopping Research documentation concerns automated access.

OpenAI says Shopping Research reads information from retailers and other public sites in real time.

However, some retailers block automated access.

When that happens, OpenAI says Shopping Research may either skip those sources or rely on other websites selling similar products. (OpenAI Help Center)

That is important.

This is no longer a vague technical SEO concern.

If OpenAI’s shopping systems cannot access important product content, the retailer may lose an opportunity to supply first-party information during live product research.

For ecommerce sites, teams should review:

• Robots directives
• CDN rules
• WAF rules
• Rate limits
• Bot-management systems
• Authentication barriers
• JavaScript rendering dependencies
• Access logs showing OpenAI-associated traffic

The objective is straightforward:

Do not accidentally block the system you want to recommend your products.

The ChatGPT Ecommerce Selection Framework

For every strategically important SKU, build a clean product-information stack.

1. Identity

Make the exact product unmistakable.

Capture:

• Brand
• Exact product name
• Model
• SKU
• Variant
• GTIN/UPC where applicable

2. Attributes

Provide machine-readable information about:

• Size
• Material
• Color
• Capacity
• Performance
• Technical specifications
• Product category

3. Compatibility

Explicitly answer questions such as:

• What does this work with?
• What does it not work with?
• Which device, vehicle, platform or environment is it designed for?

4. Commercial Evidence

Maintain accurate:

• Price
• Promotions
• Inventory
• Shipping information
• Reviews
• Ratings

5. Semantic Use Cases

Explain real buying scenarios.

For example:

Rather than describing a boot only as “waterproof hiking footwear,” identify whether it is suitable for:

• Winter hiking
• Snowshoeing
• Backpacking
• Technical terrain
• Wet environments
• Multi-day use

Those use cases mirror the way consumers naturally ask AI systems for recommendations.

6. Merchant Evidence

Make merchant identity clear.

Establish:

• Manufacturer status
• Authorized seller status
• Availability
• Returns
• Warranty
• Pricing

7. Feed Freshness

Where supported, provide accurate product feeds.

OpenAI explicitly notes Shopify Catalog integration and the availability of direct product feeds for merchants seeking more current product information. (OpenAI Help Center)

8. Machine Accessibility

Confirm the important product pages can actually be accessed by AI shopping systems.

Do not assume crawler access.

Verify it.

Two Tests Ecommerce Teams Should Run

Test 1: Product Selection vs. Merchant Selection

Ecommerce AI visibility can be measured as consideration, product selection, and merchant selection rather than one final recommendation.

Ask tightly constrained product questions such as:

Best waterproof hiking boot under $250 for winter hiking and snowshoeing.

Then record:

• Which exact products appear?
• Does the client’s product enter consideration?
• Does it appear among the top recommendations?
• Which merchant is attached?
• Is the manufacturer shown?
• Is a third-party seller shown instead?

This separates product visibility from merchant visibility.

Test 2: Pre-Search Prior vs. Current Evidence

Choose:

• One well-established product
• One lesser-known but objectively competitive product

Then improve the lesser-known product’s current evidence:

• Structured attributes
• Product descriptions
• Pricing
• Inventory
• Reviews
• Compatibility language
• Merchant feed
• Page accessibility

Run the same controlled shopping queries repeatedly.

Do not claim that this reveals OpenAI’s internal ranking algorithm.

It doesn’t.

The purpose is to determine whether stronger current evidence improves candidate inclusion or recommendation frequency.

What Most Ecommerce Reporting Gets Wrong

Many teams are still measuring ChatGPT commerce visibility with one binary metric:

Recommended: yes/no

That is too crude.

A stronger model is:

This gives ecommerce teams diagnostic information instead of a vanity score.

What Actually Works Now

Based on OpenAI’s own documentation, the highest-confidence actions are not exotic.

They are operational:

• Make product identity unambiguous.
• Maintain high-quality structured metadata.
• Keep price and availability current.
• Build strong public product evidence.
• Maintain reviews and product comparisons.
• Provide accurate merchant information.
• Use product-feed integrations where appropriate.
• Keep critical retail content accessible to automated systems.
• Test product selection separately from seller selection.

This is far more defensible than publishing speculative lists of “25 ChatGPT ranking factors.”

Key Takeaway

OpenAI is giving ecommerce brands a clearer picture of how AI shopping works.

And it isn’t one ranking.

It is a sequence.

Did the product enter consideration?

Did current information validate it?

Did it win product selection?

And did the client’s store win the merchant competition?

That should become the new ecommerce AI visibility funnel.