How a DTC brand can make a new product discoverable and recommendable in ChatGPT

A new DTC product does not need a large review count to become discoverable in ChatGPT, but no brand can force a recommendation. Start with accurate, accessible product data: let OpenAI’s crawler reach the site, keep the product page complete and current, and use Shopify Catalog or an approved product feed where available. Then make the product easy to evaluate against a specific need, while earning independent evidence over time.

First separate discoverability from recommendation

Discoverability means ChatGPT can find and accurately identify the product. Recommendation means the product fits a shopper’s stated constraints well enough to be included among the options. The first is largely a data and access problem; the second also depends on relevance, credibility, availability, price, and the alternatives ChatGPT finds.

OpenAI says ChatGPT shopping can use merchant product data from the Agentic Commerce Protocol, public product information, and other retail sources. It may compare products using details such as price, features, and reviews. That means a new product can be eligible for consideration without pretending it has a review history it does not have.

Reviews are not a field you should manufacture. If the product has no reviews, say so through the product experience and supply stronger first-party detail instead. A clean catalog record can make a product visible; it cannot make an unsuitable product the best answer.

Make the product technically findable

The first practical task is to remove barriers between ChatGPT and the product’s canonical page. OpenAI says any website or merchant can appear in ChatGPT Search and advises merchants not to block OAI-SearchBot in robots.txt.

Use this launch checklist:

Keep the product detail page publicly accessible and return a successful response.
Give the product one canonical HTTPS URL and use that URL consistently in navigation, feeds, and product references.
Put the product name, description, price, currency, availability, variants, images, and key specifications on the page.
Make material facts visible in text, not only inside images, videos, or a client-rendered interface that a crawler may not receive.
Keep discontinued, out-of-stock, and variant information accurate.
Check that the page does not require a login, location selection, or a chat interaction before the basic facts appear.

OpenAI’s product-feed specification requires a product title and product-detail URL among the basic product data. It also defines an is_eligible_search flag that controls whether a product can be surfaced in ChatGPT search results. Treat the feed as a structured version of the same truth on the product page, not as a place to add claims the page cannot support.

Use the commerce path available to your brand

For Shopify merchants, OpenAI says product data is already integrated through Shopify Catalog, so individual merchants do not need additional work for that integration. The remaining work is catalog hygiene: accurate titles, variant data, images, prices, inventory, and product-page copy.

Brands with access to direct product feeds can provide OpenAI with structured catalog data. OpenAI’s developer documentation says onboarding product feeds is currently available to approved partners, and recommends applying through its merchant process. The documentation also recommends a complete feed at least daily, with updates during the day through the API, depending on the integration method.

If direct-feed access is not available, do not wait for it before fixing the site. Follow OpenAI’s crawler guidance, make the product page comprehensive, and use the commerce integration your platform already supports. Recheck the result after changes; a feed or catalog cannot correct a page that has conflicting prices, unclear variants, or missing availability.

Write for the buying question, not just the product name

A new product becomes more recommendable when its page answers the constraints a shopper uses to choose. “New insulated jacket” is weak context. “A packable rain jacket for wet, windy commutes, under a defined budget, in petite sizes” gives a recommendation system something to match.

Build a concise fact set around five areas:

Shopper constraint	What the page should make explicit
Fit	Dimensions, sizing method, intended fit, weight, and model measurements where relevant
Use case	Who it is for, where it works, and the conditions it is not designed for
Performance	Materials, capacity, compatibility, durability, care, testing method, and meaningful limits
Choice	Differences between variants, colors, bundles, and the closest alternative in your range
Purchase risk	Returns, warranty, delivery regions, stock status, and what is included

Do not hide the decisive qualifier in marketing language. “Built for adventure” is less useful than a specific explanation of temperature range, load limit, water resistance, ingredients, compatibility, or maintenance—whichever actually determines whether the product fits.

Create comparison content only where a real decision exists. A short page explaining “which version is right for a small apartment?” or “what changes between the standard and pro model?” can be more useful than several near-duplicate articles repeating the product name.

A product with no customer history can still publish verifiable evidence, but first-party claims and independent corroboration play different roles. Your own page can establish what the product is; outside sources can help establish whether the claims deserve confidence.

Use evidence that matches the claim:

Publish the test method, conditions, and result for performance claims.
Identify materials, ingredients, manufacturing details, or compatibility requirements precisely.
Provide documentation from qualified testers, labs, designers, or professional users when it exists.
Offer samples or early access to people who can assess the product honestly, without requiring positive coverage.
Seek independent editorial, expert, retailer, or community discussion that describes the actual product rather than repeating launch copy.
Keep a dated change log when a formula, material, or specification changes.

Do not buy reviews, create fake customer stories, or ask partners to disguise sponsored claims as independent recommendations. Thin evidence is a launch-stage limitation to manage, not a gap to fill with fabricated authority. Also avoid borrowing another product’s reputation: ChatGPT may encounter similar products, but similarity does not transfer their reviews or performance history to yours.

Turn early shopper questions into product clarity

Before enough orders accumulate for customer-data patterns, use the questions people ask before purchase as a faster source of learning. Questions about sizing, compatibility, performance, delivery, and alternatives reveal which facts are missing or difficult to compare.

Anagram’s Site Agent is designed for this on-site decision moment: its official site describes conversational support while shoppers compare options and decide what to buy. Its AI Visibility product describes monitoring how a brand appears in ChatGPT, comparison with competitors, and customer questions that show where to focus next. Those are useful complements to catalog distribution—not a guarantee that ChatGPT will rank or recommend a product.

A practical loop is:

List the product’s intended use cases and disqualifying constraints.
Ask the same natural-language buying questions a shopper would ask.
Record where the answer is uncertain, incomplete, or overly dependent on a sales representative.
Improve the product page, comparison content, feed fields, or policy information.
Test again across the product and its closest alternatives.

Anagram’s AI Visibility product can show how a brand appears in ChatGPT, how it compares with competitors, and what customer questions reveal about where to focus. Its homepage describes the connection between a branded Site Agent, shopper-question learning, and improving site and AI visibility.

Measure the right launch signals

Do not use recommendation volume as the only early success metric. A new product may be correctly surfaced yet lose because it is out of stock, poorly differentiated, outside the requested budget, or missing a decisive attribute.

Track four layers separately:

Access: Can the crawler reach the page, and does the page expose the current product facts?
Representation: Does ChatGPT describe the product, price, availability, and use case correctly?
Fit: Does it appear for the specific shopping questions the product was built to answer?
Business outcome: Do referred shoppers click, engage with product guidance, add to cart, and purchase?

Keep a test set of realistic questions with explicit constraints, such as use case, budget, size, location, and preferred trade-off. Record the date, product facts shown, cited sources, competitors mentioned, and whether the answer gets the product right. ChatGPT’s output can change as product data and public information change, so treat this as monitoring rather than a one-time launch certificate.

The shortest path is not “get reviews first.” It is: make the product accessible, make its structured data current, explain its fit and limits in concrete language, earn independent evidence honestly, and use early shopper questions to close the gaps. That gives ChatGPT something accurate to discover and a defensible reason to recommend.