AI Recommendations: Find the Products Competing With Yours

To find which products compete with yours in AI recommendations, inspect answers to a real buying question, then separate the recommended product from the seller offering it. Record the manufacturer, model, variant and purchase destination before building your competitor shortlist. Another store selling your product is a retail competitor, not evidence that the assistant preferred another manufacturer.

This distinction matters for commerce teams deciding what to investigate next. A competing model, a different size and a different seller call for different follow-up work. You can make this distinction with an observation sheet, the original answer and verified product pages, without assuming you know why an assistant selected an option.

1. Choose the purchase decision you want to investigate

Start with a category question that describes the buyer’s task and essential constraints. Keep direct questions about your own brand separate: asking whether your product is suitable is not the same test as asking which products the assistant suggests without naming you.

For example, a constructed test question could be: “Which everyday backpack has a separate laptop compartment and can be delivered to Sweden?” This is an illustrative prompt, not evidence of search demand. In your own work, use customer wording where you have a permitted source, or label your test questions as constructed.

Before collecting answers, identify your reference product precisely. Note the manufacturer, model, relevant size or configuration, and the seller you represent. Decide whether your immediate question is about competing products, competing retailers, or both. That decision determines how you classify the results.

2. Check the exact AI surface and market

A text answer, a shopping product card and a Google Search AI response should not be treated as the same observation. Check that the experience you want to inspect is available for the market and account you are using.

• ChatGPT shopping: OpenAI’s merchant guidance, checked on 2 November 2026, describes products tailored using shopper preferences, context and product data. It states that shopping is live for users in the US. Do not use that statement as confirmation of Swedish shopping availability.

• Gemini Apps shopping: Google’s shopping help, checked on the same date, says Gemini uses Google’s Shopping Graph. Its supported-country list does not include Sweden. The guidance also specifies a personal Google Account and other eligibility requirements.

• Google Search AI Overviews and AI Mode: these are separate Search surfaces. Google’s website guidance explains their supporting links and eligibility conditions; it is not a specification for shopping inside Gemini Apps.

For a Sweden-in-English investigation, preserve that scope. If you inspect a US shopping experience separately, label it as a separate market observation. The shopping availability limits above do not establish that ordinary text answers cannot discuss products in Sweden.

If the required shopping feature is unavailable, mark that observation unavailable. You have not established that your product, or a competitor’s product, was absent from a valid result.

3. Capture the recommended item and its destination separately

For each available response, save the exact question, date, platform, surface, market and language. Include the model or version where shown, whether search was enabled, and relevant account or conversation context. Preserve the complete answer or screenshot, not just the names you recognized.

Then record two separate pieces of evidence:

• The product recommendation: the exact wording or card that presents an item as an option, including the stated suitability or limitation.

• The destination: the linked product page, buying option or supporting source, including the seller shown there.

Google’s Gemini shopping documentation says product cards can provide descriptions, reviews and buying options. OpenAI’s merchant documentation describes product discovery followed by purchases on merchant-owned websites or apps. These documented experiences make it useful to distinguish the item from the business selling it.

Do not automatically classify every link as an endorsement. A supporting citation can explain a specification without recommending its publisher. Likewise, a product named in prose may have no link to its manufacturer. Preserve mention, citation and recommendation observations independently.

4. Resolve manufacturer, model, variant and seller

Open the relevant product destinations and compare their identity with your reference item. Use the manufacturer’s documented specifications for the model and variant, and the merchant’s current page for its offer and delivery terms. If the sources disagree, keep the identity unresolved until you can verify it.

Use these classifications in your working record:

• Competing product: a verified different model or manufacturer offered for the same buyer task.

• Competing seller: the same verified product and variant offered by another merchant.

• Different variant: the same model family with a size, capacity, configuration or other difference that affects the comparison.

• Unresolved identity: the answer or destination does not establish which item is being offered.

Keep the assistant’s claimed fit separate from your verified finding. “Suitable for a laptop” is answer text. Whether the exact compartment fits the buyer’s device still requires checking the relevant product specification.

A filled example: three names do not mean three competing products

Fictional example: a retailer sells Manufacturer A’s Daily Pack, standard size. An invented answer to the backpack question presents the following options. The labels, offers and verification outcomes are illustrative, not collected AI results.

Item in the answer

Destination checked

Identity finding

Next verification

Manufacturer B, Commuter Pack

Merchant North’s page for that model

A different product, so retain it on the product competitor list

Check its compartment specification and delivery to Sweden before accepting the claimed fit

Manufacturer A, Daily Pack, standard size

Merchant South’s page for the same verified variant

Your reference product through another seller, so retain the merchant on the retail competitor list

Compare the actual offer and delivery terms, not manufacturer visibility

Manufacturer A, Daily Pack Compact

A category page that does not establish the selected variant

Unresolved variant, not a confirmed like-for-like alternative

Find the exact product destination and verify its dimensions before classifying suitability

The resulting shortlist distinguishes a competing model, another seller of the reference product, and an unresolved variant. A report containing only the displayed names would lose those distinctions.

If you represent the manufacturer, the second row is evidence that your product appeared, even though another merchant received the purchase destination. If you represent the retailer, that same row identifies a seller worth investigating. Neither interpretation proves why the destination was selected.

5. Choose the follow-up that matches the finding

For a confirmed competing product, inspect the evidence for the buyer constraint. Does the competing model actually meet it? Does your own product meet it? If your product is unsuitable, do not rewrite the page to imply otherwise. If it fits but your page leaves that fact unclear, improve the relevant existing passage with verified detail.

For a competing seller, compare product identity, availability, destination and applicable delivery terms. A new article about the manufacturer will not itself resolve an incorrect variant mapping or an outdated merchant offer.

For inaccurate catalog information, check the source and update route appropriate to the platform. OpenAI says product feeds can give merchants greater control over accurate, current product representation. This supports investigating feed quality; it does not establish that submitting or correcting a feed guarantees selection.

For a Google Search page issue, keep the technical diagnosis specific. Google says supporting links in AI Overviews and AI Mode require indexing and snippet eligibility, with no additional technical requirements. Its guidance also recommends visible textual information, structured data matching the page, and current Merchant Center information. Those conditions are not proof that a particular product will be recommended.

For an unresolved item, preserve the uncertainty. An inaccessible destination or ambiguous model name is not evidence of a product defect or a missing content page.

6. Turn the shortlist into a repeatable record

Your completed record should let another person trace each competitor classification back to the answer and the verified item. Alongside each row, keep the original response, exact destination, verification date, identity status and the reason for the chosen follow-up.

When collecting later observations, retain the same question and market scope, and record changes to the surface, account context or product offer. Report what appeared and what changed. Do not treat a later recommendation as proof that a catalog correction or page edit caused it.

The immediate outcome is a defensible product-and-seller shortlist, not a ranking formula. Start by resolving the identities in an answer you already have. Only then decide whether the next useful action belongs with product content, catalog operations, the retail offer or further observation.