Filters are where a technical catalog either helps the shopper or gets in the way. A shopper who searched “bearing” and sees 300 results wants to narrow by bore, outer diameter and seal type. A shopper who searched “ball valve” wants thread size and material. One fixed list of filters can’t serve both.
Pick filters by product type
Start with your main product types and ask: what does a buyer need to know to pick the right one?
| Product type | Useful filters |
|---|---|
| Bearings | Bore, outer diameter, width, seal type, brand |
| Bolts and screws | Thread, length, standard, grade or material, head type |
| Fittings and valves | Thread size, thread type (BSP, NPT), material, angle |
| O-rings and seals | Inner diameter, cross-section, material (NBR, FKM) |
| Electrical | Rating, poles, voltage, lamp base, colour temperature |
| Replacement parts | Fits brand, fits model, part type |
Brand and availability are useful almost everywhere. Price is less useful than you’d think, because a technical buyer picks by fit first.
Show the filters that fit the results
A search for “bearing” should show bearing filters. A search for “1/2 ball valve” should show valve filters. The simplest way to get there is to build the filter list from the products that matched: count which attributes the results have, and show the ones that split them in a useful way.
That also avoids a common failure: a long sidebar of filters where most options lead to zero results.
Where filter values come from
Filters only work with clean values. 20 mm, 20mm and Ø20 must end up as one value, or the filter shows three options for the same thing.
Most technical catalogs have the data, but in the wrong place:
- In the title:
Deep groove ball bearing 6204-2RS 20x47x14 mmcontains bore, outer diameter, width and seal type. - In the description, as a sentence or a spec table.
- In metafields, for stores that imported structured data from a supplier.
- In tags, often inconsistent.
Moving all of that into fields by hand is a big job for a catalog of thousands of products. That is why many stores never get past brand and price filters.
Using AI to suggest values, carefully
Language models are good at reading a title like Hex bolt DIN 933 M8x25 grade A2-70, box of 100 and pulling out thread M8, length 25 mm, standard DIN 933, grade A2-70 and pack size 100. They are also capable of guessing things that aren’t there.
A safe setup has three rules:
- Only values that are written in the listing. The model must quote the words it used. No inferred compatibility, no guessed sizes.
- A person approves before anything goes live. Suggestions wait in a review list; approved values become filters.
- Run once per product, not per search. Reading a product again only when its text changes keeps cost low and results stable.
How Tolra handles it
Tolra builds the filters on the results page from the products that matched each search, so a bearing search shows bearing filters. On the Growth and Pro plans, AI reads each product’s text (title, description and product fields) and suggests values like size, thread, seal type, material or compatible model, with the words it found them in. You approve or reject them in batches; only approved values become filters. New and changed products are read again; price and stock changes don’t count. They can also be written to metafields on your products, so the data stays in your store.
See writing product titles for a technical catalog for titles that give AI and shoppers more to work with.
Questions
How many filters should a search results page show?
Only the ones that split the current results in a useful way. A filter where every product has the same value, or where each product has a different one, doesn't help.
Should price be a filter?
For technical products, size and fit usually matter more than price. Shoppers who already know the part rarely filter by price. Sort options cover it when they do.
Should filter values live in metafields?
Metafields are a good home for structured values, because the data stays in your store. On the Growth and Pro plans, Tolra can write approved filter values to metafields on your products, so they are not locked inside the app.
Examples use made-up brands and part numbers unless a real format is needed to show the point. Try them in the interactive demo.