Official manufacturer image of the model discussed in the guide.
A shopper asks ChatGPT whether a dock can run two monitors from their MacBook. Seeing the product recommended is a good start for a merchant, but it doesn't necessarily mean the answer gives the shopper everything they need. If it leaves out the chip, lid position or connection requirements, the shopper could still end up buying the wrong product after clicking through.
That's where generative engine optimization, or GEO, comes in. The goal is straightforward: make accurate product and store information available, then check what buyers actually learn from AI answers and what happens after they follow the links. Pay attention to the exact product being recommended, the conditions attached to that recommendation, the evidence supporting each claim and the variant the shopper can actually purchase.
The Shopify features and admin paths below were checked against documentation on November 4, 2026. The Satechi example draws on publicly available product and support pages. The implementation steps are intended for merchants applying the same approach to their own stores. The case study uses published specifications and suggested test queries.
The same Shopify merchant might appear as a product card in one answer and as a link to a compatibility guide in another. These are different ways of presenting information, so they need to be checked separately.
| Workstream | Material to maintain | Where to check it |
|---|---|---|
| Shopify product data and store facts | Catalog product and variant data, along with Knowledge Base facts such as delivery and returns | Sales channels > Agentic, plus the target channel’s sources, eligibility and settings |
| Public pages and search | Product pages, compatibility guides, policies, crawl and index settings, and structured data | Public pages, source or rendered HTML, Search Console, and the answer’s actual citations |
Shopify Catalog organizes product data for supported shopping surfaces. Agents can retrieve published Knowledge Base facts through Storefront MCP. Google AI Mode and Gemini also involve the Google & YouTube channel or Merchant Center. Relevant Meta shopping surfaces have their own Facebook and Instagram by Meta channel requirements. Each destination needs to be checked on its own. Shopify product discovery
AI answers may also draw on a manufacturer's help pages, other retailers or independent reviews. Being visible in search and being eligible for a shopping channel are two different things. Start with a group of commercially important products or items that regularly generate compatibility questions. Record the baseline before making changes to any data source.
Once the process works for this initial group, apply the same conventions to the rest of the catalog.
A product page might display a specification in an image while the description sent to Catalog leaves that same information out. Before changing anything, compare the different places where the team maintains product information.
Go to Sales channels > Agentic and review the channel and Sources > Shopify Catalog. If a product is marked Ineligible, open the status to see the reason. If no eligibility result is available, record that status and check again later or contact support.
The current ChatGPT channel requirements include selling to U.S. customers, having Catalog-eligible products, accepting the Agentic Storefronts supplemental terms and completing the terms of service, privacy and return policies under Settings > Policies. The merchant itself can be based outside the U.S. Other channels have their own requirements, so check them separately. For affected products, Shopify requires applicable legal and compliance disclosures to appear within the first 6,000 characters of the description. Shopify ChatGPT channel requirements
The store owner can access Agentic. Staff members need the appropriate App development > View apps developed by staff and collaborators or App and sales channel > Agentic permission. Shopify Agentic administration
Stores selling to both B2B and retail customers should also review the products and prices being shared with each audience. Shopify recognizes its native B2B controls, but a theme or third-party app that hides wholesale prices may behave differently. Check B2B catalogs, authenticated access and channel settings together.
Keep in mind that Unlisted affects product discovery, including storefront search, search engines and sitemaps. If prices are sensitive, use the appropriate access controls rather than relying on product visibility settings. Shopify discovery and B2B scope
Find the product in the Agentic search preview and expand Listing insights. The five groups of signals provide a useful starting point for identifying gaps in the product data.
| Signal | Action | Result to retain |
|---|---|---|
| Description completeness | Answer recurring questions about use, limits and package contents | A description that supports a purchase decision |
| Variant and option completeness | Check the names, inventory and actual differences. Expand unfamiliar abbreviations on first use | A clear identity and conditions for each purchasable option |
| Image coverage | Add useful product, port and use images and write the decisive specifications in page text | Visual explanation with searchable specifications |
| Product reviews | Identify the source of ratings and counts | A documented source and known coverage |
| Shop policy completeness | Align shipping, refund and return wording with actual service terms | Consistent policies in the admin and on public pages |
Shopify uses description length as one of its completeness signals and counts reviews verified by trusted sources. Focus on filling gaps that could affect a shopper's decision. These indicators should not be treated as a measure of channel ranking or recommendation performance, which need to be evaluated separately. Listing insights
Satechi’s Thunderbolt 4 Multimedia Pro Dock is a useful example of why product specifications and device compatibility need to be documented separately. The dock's available ports don't automatically tell a shopper how many independent displays their computer can drive.
| Public product fact | What to keep separate |
|---|---|
| The specifications list regional SKUs including ST-DT4PMM-US, ST-DT4PMM-EU, ST-DT4PMM-UK and ST-DT4PMM-AU | Each plug option should match its real SKU |
| The included power adapter is rated at 135W and host PD is up to 96W | Adapter rating and host charging output answer different questions | | Base M1 and M2 MacBooks are limited to one external display | Show the relevant host limitation near the compatibility claim | | Base M3 MacBook Air or Pro needs clamshell mode for two displays at 4K 60Hz | Keep the chip, display count and lid condition together |
Satechi technical specifications and compatibility · Satechi package and power documentation
For a product like this, maintain a record of the model, regional SKU, power figures, host requirements, display limits, information source and verification date. Use variants for the options customers can actually order. Put shared specifications in the description or defined product metafields. The full compatibility conditions should also appear in the product page body. If the page uses a selector or custom helper component, make sure it draws from the same underlying data.
When introducing a new metafield, define its name and type first, then enter the actual values. After that, check both the public product page and the Catalog output. Pay particular attention to whether a capability statement applies to every buyer or only to a particular device under specific conditions.
Go to Sales channels > Agentic > Sources. Open the dropdown beside Shopify Catalog and select Review product data mapping.
Keep the default mapping wherever it already provides the correct information. Under Product fields, check the source and coverage of the title and description. Open the coverage details and use Fix [number] products to address missing information. If the complete data lives in a metafield, point the relevant field to that metafield.
Title and description support sources, fixed text and Liquid. Product category belongs in the product record.
Preview three types of items: one with a populated value, one without a value and one with an unusual specification. When a mapped field is blank, Catalog falls back to the default field. Check that the resulting description still contains the information a buyer needs.
Record when the mapping is saved, since changes take time to process. The preview shows what Catalog receives as input. Each destination controls its own presentation.
Check List price separately. The available options are Price and Compare-at price. The latter is used when it exceeds the selling price and can be sent in UCP as list_price. Record the market, currency, variant, sale price and reference price together so they can be checked against one another. Shopify Catalog mapping
When Custom variant grouping is enabled, Catalog uses the selected grouping method and ignores the Combined Listings grouping configuration. The online store keeps its existing product structure.
Before changing the grouping method, save the current groups, representative products and option values so the original setup can be reviewed if needed.
Only group products that genuinely belong together and have distinguishable options. Choose a title separator, metafield or tag prefix, then inspect Group name and Products in group.
Review the products that have been grouped, left separate or marked need review. Identical option combinations across products can trigger a review. If that happens, correct the underlying product data before saving the grouping again.
For a product such as the Satechi dock, actual plug variants and host compatibility should remain separate pieces of information. Shopify Catalog grouping Product descriptions should answer questions about the item itself. Shipping, returns and customer support belong to store-level information.
Go to Sales channels > Agentic > Sources > Knowledge Base. Storefront MCP can make published facts available to agents. Each destination decides when and how to use those facts in its answers.
Facts generated from settings such as policies or contact details are published when created. Facts scanned from storefront pages need to be reviewed first.
When a Suggested update appears, compare it with the current page. Choose Approve & publish if the suggested content is correct, or reject it if it isn't, then save the changes.
Check the fact text, its source, publication status and reference URL together. If information is missing or outdated, edit the fact directly or use Request rescan.
Shopify says a rescan can take up to 48 hours. Also, a public page may still provide information even when the corresponding Knowledge Base fact is unpublished. That means an answer can be factually correct while the agent's retrieval route remains unclear. Knowledge Base publication and scanning
Use Additional facts > Add fact and select a type that keeps the important conditions intact.
Enter the title and content, add public references, hover over Unpublished, select Publish and click Save. The 1,000-fact limit includes both generated facts and manually added ones.
Custom storefronts and the Agentic plan currently support additional facts only. A password-protected storefront cannot be scanned and mainly displays information sourced from the admin.
The old Knowledge Base app is scheduled to shut down on January 15, 2027. For an old app containing no more than 1,000 facts, automatic migration and publication apply to manually overridden store facts and added facts. Review the migrated content item by item rather than assuming every entry is still accurate. Knowledge Base limits and migration
A field has one source but can have multiple references. To add references manually to an existing category field, switch its source to Manual, click the reference count and add the relevant public page under References.
This change removes the automatic references from the previous source, so the team becomes responsible for keeping the fact and its references up to date. Save the changes and check both the links and the wording again.
For example, link return-related facts to the current return policy and warranty claims to the warranty instructions. Updating a reference URL does not automatically update the fact's wording. These links help agents locate supporting information, but the actual answer still needs to be checked to see which sources it displays. Published facts are public, so write them in customer-facing language. Private quotations and customer records should stay out of these fields. Knowledge Base references
Choose a product or policy URL and check its response, robots.txt, page or HTTP indexing directives, canonical and discoverable internal links. Use URL Inspection to review whether the page has been discovered, when it was last crawled and its indexing status.
The different settings serve different purposes. Robots rules govern crawling, noindex concerns indexing, and the canonical indicates a preferred URL. Identify which setting is causing the problem before changing anything, and preserve access controls that are there for a reason.
Shopify manages network-level crawler security, while a theme can customize robots.txt.liquid. Catalog sharing has its own channel controls.
For localized stores, compare hreflang, the canonical, the regional URL, currency and product availability. Record the market used in each test so results from different regions aren't mixed together. Shopify crawling and sharing
Shopify provides /agents.md, /llms.txt and /llms-full.txt. By default, the latter two mirror the first. Open these files at the primary domain root and check the store name, domain, policies and discovery endpoints.
These files do not have copies in Markets language subdirectories.
For a documented custom requirement, go to Online Store > Themes > Edit code > Add a new template.
| Template | Scope |
|---|---|
| agents.md.liquid | Customizes the three discovery URLs by default. A dedicated template takes precedence for its URL |
| llms.txt.liquid | Overrides /llms.txt |
| llms-full.txt.liquid | Overrides /llms-full.txt |
These templates use the restricted agents and request contexts. Preserve UCP, MCP and sitemap discovery, and make sure someone is responsible for maintaining the files.
The repository instruction file AGENTS.md serves a different audience and should not be confused with the storefront discovery files.
Google's AI search guidance focuses on accessible, indexable pages and consistent visible information. Its AI features do not require a special AI file or schema. Shopify discovery templates · llms.txt precedence · Google AI features
The included 135W adapter, up to 96W host charging and base M3 lid condition should appear close to the claims they qualify.
A custom Shopify build can use approved fields to render specification tables, compatibility groups and messages tied to the selected option. Where practical, include the decisive facts in the server response, then inspect the initial HTML, rendered page and target crawler's view.
On a phone, select different versions and check whether the displayed conditions and selected SKU change accordingly. Where testing is authorized, add the product to the cart and confirm that the line item matches the selected option.
The same important specifications should also be visible in the source or rendered page inspection.
Start by reviewing the JSON-LD already generated by the theme and apps. Look for competing Product entries or inconsistencies in price and stock.
Check the product name, brand and identifiers, along with the corresponding Offer price, currency, availability and URL.
Depending on the page structure, ProductGroup, hasVariant, variesBy and productGroupID can be used to describe genuine variant relationships.
Keep the identifiers straight: SKU is the merchant's inventory identifier, GTIN identifies a trade item, and MPN is a manufacturer's part number. Populate the identifiers that actually exist for the product. Google product variant structured data · GTIN · MPN
Check the supported rich results and schema syntax, then compare the output with the selected variant and the information visible on the page.
Google stopped showing FAQ rich results on May 7, 2026. Keep answers that help buyers and use markup that reflects the actual meaning of the page. Google Search updates
Under Sales channels > Agentic, enter a product or purchase query in Enter a search query to see how you might rank.
You can also use Top search queries or Queries where your products appeared. For supported previews, open Preview this search.
Save the original query, returned products and Listing insights. After fixing verified data gaps, run exactly the same query again.
Catalog's original ordering and the final ranking on a destination may differ, so don't treat the preview as the final recommendation order. Search preview and reranking
A sample of 15–30 questions is a reasonable starting convention for a team. Build the questions from actual sales conversations, site searches and return disputes.
An unbranded question helps assess whether a product can be discovered. A question naming a specific product helps check whether the information returned about that product is accurate.
| Purpose | Suggested question | What to verify |
|---|---|---|
| Unbranded discovery | Which docks can drive two 4K 60Hz displays from a base M3 MacBook Air if I can close its lid? | Correct chip and lid assumptions, and evidence for the candidates |
| Compatibility | How does dual-display support differ between a base M1 MacBook and a base M3 MacBook with the Satechi Thunderbolt 4 Multimedia Pro Dock? | Separate the host limits |
| Power | Does the included 135W adapter mean the dock charges my laptop at 135W? | Distinguish adapter rating from up to 96W host PD |
| Product identity | What differs between ST-DT4PMM-US and ST-DT4PMM-EU? | Correct region and actual purchasable SKU |
| Policy | What are the current Satechi store return and warranty conditions? Give the official sources | Check the current full terms and the applicability of each cited condition |
Use these questions to plan a test. Keep the language, market, exact prompt and conversation setup consistent. Record the date, platform, model or mode, region, login status, memory and personalization settings. Preserve the complete answer, product link and citation URLs.
Keep repeat samples rather than relying on a single response. Starting a fresh conversation can reduce some context variation, but first-party and third-party sources, query context and personalization can still affect the result. OpenAI shopping with Search
Take an individual claim from the answer and open the cited page. Find the passage that supports it, then check the edition, market and model to make sure they match.
Save product card links separately from explanatory citations. For each test, retain the question and environment, original answer, claim, actual URL, official evidence, observed gap, working hypothesis, correction and retest time. If a claim has no citation, record that explicitly.
| If you observe | Evidence to collect | Action once verified |
|---|---|---|
| Missing compatibility condition | Full answer, cited passage, Catalog preview, page text and variant relation | Fix a mixed description or mapping. If the official page is accurate, make the supporting section easier to find and retest |
| Old returns language | Current policy, old cited page, Knowledge Base publication state and indexing | Repair the specific stale source. Verify the setting before attributing the error to indexing or the knowledge base |
| Different price | Time, currency, market, variant, stock, link, Offer, feed and Catalog preview | Correct the specific market, variant or synchronization issue |
| Preview includes the product and the real answer omits it | Comparable repeated samples, qualification and preview | Fix proven data gaps, then keep observing channel selection |
A citation shows what the answer presented as evidence. Any explanation of the model's internal retrieval path should be treated as a separate hypothesis, not as something established by the citation alone.
For a non-headless store, go to the analytics section under Sales channels > Agentic. Select a reporting period and channel, then review Sales, Orders, Online store sessions and Online store conversion.
The combined report is available under Explore > Understand your performance on agentic channels > Review.
Sales includes both referred and direct checkout sales. However, an individual citation needs its own attribution evidence. Headless storefronts currently require a different analytics setup.
Google includes traffic from AI search features in its overall Web search reporting. A change in Web clicks therefore needs additional evidence before it can be attributed specifically to AI features. Shopify Agentic analytics · Google AI reporting
Use Settings > Customer events to manage and test pixels.
Standard events include product_viewed, product_added_to_cart, checkout_started and checkout_completed. If a custom selector is used, recording the selected variant requires the relevant behavior to be defined and captured in the actual implementation.
Send events to the chosen analytics destination through the pixel setup, while respecting consent settings.
Check that the event, cart variant ID and resulting order agree. If an event is missing, review consent, browser restrictions and collection errors before assuming the tracking setup is incorrect. Standard events · Shopify pixel management
Also record returns and their reasons using order and support records. Separate compatibility issues, wrong-region purchases, delivery problems and product defects.
When comparing results, use comparable products, markets, time periods and sample sizes. Keep unidentified traffic and orders separate rather than assigning them to a specific AI source without supporting evidence.
WESWOO can organize the work around three practical deliverables.
Start in the admin when product fields or policies are incomplete. Update the page and selector when important specifications are buried in images or the selected option doesn't match the cart. Assign development work when apps or templates produce conflicting information.
For the next product launch or policy update, use the same register and rerun the questions affected by the change.