The similar products block occupies a strategic place on a product page: it is seen by hesitating visitors, and it often decides whether the customer leaves or continues. Its quality depends entirely on the rule that fills it, and that rule is rarely examined.
The five methods
Manual association. You designate the linked products yourself. Maximum precision, maximum cost: on three thousand references, the exercise is not sustainable and it ages as soon as a product is withdrawn.
The same category. The block shows other products from the current category. It is the most widespread native behaviour, and it often produces absurd pairings: a €4 screwdriver next to a €300 drill.
Shared attributes. The matching is done on shared characteristics: same brand, same material, same power. More relevant, provided your attributes are correctly filled in.
Purchase behaviour. The products viewed or bought by the same customers. This method captures links you would not have imagined, and it requires a sufficient data volume to be reliable.
Semantic similarity. The matching is done on the meaning of the content, by comparing numerical representations of the descriptions. It works without behavioural data and without structured attributes, which makes it usable from day one.
What each method misses
It is by identifying the flaws that you choose correctly.
The category misses the price range and the product level. It mixes the accessory and the main equipment.
Attributes miss everything that is not structured. On a catalog where 40% of products have no characteristics filled in, the method only works on the remaining 60%.
Behaviour misses new arrivals, which have no history yet, and it suffers from a reinforcement effect: the products already visible become more so.
Semantic similarity misses the distinctions that the text does not carry. Two products described identically but of very different qualities will be judged close.
Manual association misses nothing but it does not scale.
AI Semantic Search — PrestaShopThe search that understands intent, not just keywords€149.00
The rule that works: the combination
No single method gives a good result on a real catalog. The effective configuration layers three levels.
An algorithmic base, semantic or behavioural similarity depending on your data.
Consistency filters, which discard absurd pairings: price range, availability, product universe.
Priority manual associations on your fifty main references, which override the calculation.
This organisation gives you complete catalog coverage with control where the stakes are high.
The consistency filters, in detail
They are what makes the difference between a credible block and a ridiculous one. Four rules.
The price range. Do not offer a €20 product next to a €400 product, nor the reverse. A range from 50% below to 150% above the current price is a reasonable benchmark.
Availability. An out-of-stock product has no place in a recommendation block. It is the most visible flaw and the simplest to fix.
Excluding the current product and its combinations. It seems obvious and it happens regularly.
The range level. On a catalog with marked ranges, an entry-level product and a professional product do not recommend each other, even if their prices are close.
Similar or complementary: two different blocks
A frequent confusion that costs revenue.
Similar products are alternatives: the customer hesitates between them and will buy only one. The block serves to retain a visitor who is not convinced by the product being viewed.
Complementary products add up: accessories, consumables, related parts. The block serves to increase the basket.
The two have neither the same rule, nor the same place, nor the same heading.
Recommended placement: the complementary near the buy button, where they capture a decision already made. The similar lower down, where they catch a hesitation.
Common mistake: displaying alternatives just below the add-to-cart button. You divert a customer who was going to buy.
How many products, and where
Four to six products per block. Below that, the block looks thin. Beyond, it becomes a catalog and the click-through rate per product collapses.
On mobile, a horizontal scroll with four products works better than a grid that lengthens the page.
An often neglected point: the block must be stable between two visits to the same page. A customer who returns to find a product they had seen in a recommendation and does not find it is lost.
Measuring, and knowing what to measure
Three indicators, and the third is the only one that really counts.
The click-through rate on the block, which measures its perceived relevance.
The add-to-cart rate from a product reached by recommendation, which measures the quality of the matching.
The incremental revenue, that is, what the block brings beyond what would have sold without it. It is the hardest to isolate, and it requires a comparison between pages with and without the block over an equivalent period.
Beware of a measurement trap: a block that displays your best sellers gets a good click-through rate because those products sell anyway. The click is not proof that the recommendation created value.
Where to start
Three steps, in this order.
Look at your current block on ten pages taken at random, including two entry-level and two high-end products. The inconsistencies appear immediately.
Add the consistency filters before changing algorithms. On many stores, a simple price and availability filter clearly improves the result without changing anything else.
Separate similar and complementary into two distinct blocks, with explicit headings.
These three steps are done in a day and they produce most of the available gain.
The Semantic Search and Similar Products module brings this layer to PrestaShop 8 and 9: matching by content similarity working without behavioural data or structured attributes, consistency filters on price and availability, and priority manual associations on chosen references.
