AI Product Recommendations and SEO: What Changes for Organic Visibility

These are the “you may also like”, “frequently bought together” and “complete the look” blocks on product, category and cart pages. Modern engines use machine learning on browsing and purchase data to choose items per visitor or per product. Shopify offers built-in recommendations and a Search and Discovery app, and dedicated vendors such as Nosto, Klevu, Algolia Recommend and Bloomreach provide more advanced options. Magento, now Adobe Commerce, has its own Product Recommendations module.

Their primary job is conversion and order value. SEO is a side effect, but a significant one, because recommendations add links and text to thousands of pages at once.

Google discovers pages largely by following links. A product that is linked only from page 14 of a category may be crawled rarely. A recommendation block that links to it from related products gives Google another path in.

That only works if Google can see the links. Google renders JavaScript, but rendering can be delayed, and some widgets load content only after user interaction or depend on cookies that Googlebot does not keep. In those cases, the links simply do not exist for crawling purposes.

How to check what Google sees

• Use the URL Inspection tool in Search Console, run a live test and open “View tested page” to see the rendered HTML.
• Search the rendered HTML for the product links from the recommendation block.
• Compare with what a logged-out visitor sees in a private browser window.

If the links are missing in the rendered HTML, the block contributes nothing to discovery.

Personalised blocks show different products to different visitors. That is fine for users. For search, it means Googlebot sees one version, often the default, non-personalised one. This is not cloaking as long as you are not showing Google something deliberately different from what users get. Google’s guidance treats personalisation for users as acceptable.

The practical issue is consistency. If the default block is empty for new visitors, Googlebot sees an empty block. Configure a sensible fallback, such as best sellers in the same category, so every page has relevant links even with no history.

Recommendation tools can create SEO problems that are easy to miss because pages still look fine.

• Tracking parameters on links: URLs like product?rec=abc123 can create many duplicate URLs. Use canonical tags pointing to the clean product URL, or pass tracking via data attributes instead of the link.
• Linking to out-of-stock or discontinued items: wasted crawl and a poor experience. Filter them out.
• Heavy scripts: large widgets slow down Largest Contentful Paint and Interaction to Next Paint. Load them after the main content and check Core Web Vitals.
• Layout shift: blocks that push content down when they load hurt Cumulative Layout Shift. Reserve space for them.
• Too many links: a page with five recommendation carousels of twenty products each dilutes link signals and buries the main content.

They can. Product pages are often thin: a title, a few bullets and specs. A well-chosen related products block adds contextually relevant product names and links, which helps both shoppers and search engines understand how products relate.

Relevance depends on the logic. “Customers also viewed” based purely on behaviour can pair unrelated items, such as a phone case shown on a garden hose page because of one bulk shopper. Many engines let you set rules, for example same category or compatible accessories only. For SEO, rule-based or hybrid logic is often more useful than pure behaviour.

AI shopping features, from Google’s AI Mode to assistants like ChatGPT and Amazon Rufus, increasingly compare products and suggest alternatives themselves. Your on-site recommendations will not appear in those answers directly. What does transfer is the data: clear product attributes, compatibility information and structured data such as Product markup with price, availability and reviews.

The same groundwork serves both: complete attributes make your own engine smarter and make your products easier for external AI systems to understand.

Some engines create their own pages, such as “similar items” listings or personalised landing pages. These rarely deserve to be indexed. They duplicate category content, change constantly and give Google little stable value. Check whether your vendor creates such URLs, and if so, apply a noindex tag or block them from crawling, depending on whether they hold links you want followed. Keep the product and category pages as the indexable layer, and treat recommendation output as a navigation aid on top of them rather than as content in its own right.