WooCommerce chatbot product recommendations are a catalogue problem rather than a prompting one. If the attribute a shopper asks about is empty, no instruction makes the answer appear; if it is filled, the assistant narrows the range in one message. This guide covers which fields matter, how to structure them, what a good suggestion looks like, and where the assistant should stop.
The fields that decide the answer
| Field | What it enables | Common state |
|---|---|---|
| Categories | Narrowing by use: cycling, commuting, running | Often supplier-shaped rather than customer-shaped |
| Attributes | Filtering by the property asked about | Defined for variations only, empty for everything else |
| Price | Answering "under £120" honestly | Usually fine, unless tax display confuses it |
| Stock by variation | "in large" being a real constraint | Fine when stock management is on, absent when it is off |
| Short description | The one-line distinction between similar items | Marketing prose with no facts in it |
| Long description | Material, care, compatibility, what is included | Where the useful facts hide, unstructured |
The pattern across most WooCommerce stores: prices and stock are reliable, categories are inherited from a supplier feed, and attributes exist only where a variation needed them. That combination produces recommendations by keyword rather than by need.
Attributes versus variations
WooCommerce uses attributes for two jobs, and conflating them is the usual cause of thin product answers. Variation attributes such as size and colour exist because the product needs them to function. Descriptive attributes such as material, weight, waterproof rating or compatibility are optional, and most stores skip them because nothing in the checkout requires them. The assistant needs the second kind. Adding four descriptive attributes to your best-selling fifty products is the single highest-value hour in the whole setup, and it improves layered navigation and filtered search at the same time.
What a good recommendation looks like
- Respect every constraint given, including price and size, rather than the easiest one.
- Name the difference between the options, which is what the next question would ask anyway.
- State availability at variation level if stock is tracked.
- Stop at three. A longer list is a search results page, which the shopper could already see.
- Say when nothing fits. "We do not have a waterproof option under £80" earns the next question; substituting a showerproof jacket does not.
Where recommendations should stop
Three limits keep this honest. The assistant should not claim suitability it cannot support, so "will this keep me warm at minus ten" is answerable only if a temperature rating exists. It should not invent popularity: best-seller claims must come from your own sales data through the plugin, never from an impression. And it should not steer by price, because a shopper who feels sold to stops asking questions. The wider rules are in chatbot guardrails for online stores, and the conversational patterns in conversational product recommendations.
When the catalogue comes from a supplier feed
Many WooCommerce stores import products from a distributor, which brings a specific version of this problem: the data is complete in the supplier's terms and empty in yours. Feeds usually carry SKU, price, stock and a marketing blurb, and rarely carry the four properties your customers ask about. Two practical responses. Map whatever the feed does provide into real attributes rather than leaving it in a free-text blob, since a mapped field can be filtered and a blob cannot. Then enrich only the products you actually sell in volume, accepting that the long tail stays thin. Re-import discipline matters here too: if the feed overwrites descriptions nightly, your enrichment has to live in fields the import does not touch, or it disappears the first time the supplier updates a price.
Categories worth reshaping
Supplier categories describe how a warehouse thinks; customers describe what they want to do. A category tree built around brands and product types answers "show me Brand X jackets" and struggles with "something waterproof for commuting". You do not need to rebuild the tree: adding a few use-based categories or tags to your best sellers, and making sure the words customers use appear in the product text, closes most of the gap. This is also ordinary merchandising work that helps navigation and search, which is the argument Google makes about helpful content.
A practical afternoon
- Sort products by revenue and take the top fifty.
- Add four descriptive attributes to each, the ones your inbox asks about.
- Rewrite the short description as one factual line that distinguishes the product.
- Check stock management is on where you want availability answered.
- Add two or three use-based tags where the supplier categories do not match customer language.
- Ask the assistant five real shopping questions and read the replies as a customer.
The remaining products answer basic questions from their titles and prices, and get attributes when the unanswered-question list names them. The wider WooCommerce setup is in the WooCommerce setup guide, and the order-side work in WooCommerce chatbot order tracking.
Writing product text the assistant can use
Attributes carry the facts, and the description should carry the distinctions. Most WooCommerce descriptions open with a paragraph of atmosphere and never state anything comparable, which leaves the assistant unable to answer "what is the difference between these two". The fix is two sentences at the top of each description: what it is for, and what makes it different from the nearest alternative in your own range. "A lightweight commuting shell, fully waterproof with taped seams, that packs into its own pocket. Lighter than the Rain Lite but with a shorter cut." That is comparison-ready text, and it costs a minute per product. Keep the atmospheric paragraph underneath if it earns its place; nothing is lost by having both, and the shopper gets the answer in the first line either way.
Measuring whether it helps
Two checks, monthly. Read five product conversations and ask whether the assistant answered from real data or paraphrased a description. Then look at whether conversations mentioning a product precede orders from the same visitor, as a signal rather than attribution. The metric set worth keeping is in the six chatbot KPIs worth watching. What moves both numbers is catalogue completeness, not the assistant's settings, because the assistant composes answers from what it finds, the design described in Lewis et al., 2020. Vatdi syncs WooCommerce catalogues on every plan with product allowances per tier, shown on the pricing page as of September 2026.
Frequently asked questions
Can a chatbot recommend WooCommerce products?
Yes, by narrowing your synced catalogue against what the shopper asks for: category, price, attributes and stock. The quality depends on whether those fields contain values. An assistant cannot filter by a property that is blank on every product, however the question is phrased.
Which attributes should I fill first?
The four your inbox asks about most, on your fifty best-selling products. For clothing that is usually material, fit, weight and care; for electronics, compatibility, power, capacity and what is in the box. Cover the sellers first and extend when the unanswered list names a gap.
Why does it list products that do not match my budget?
Usually because the constraint was not treated as one, or because prices and variations are not syncing cleanly. Check that prices reflect your tax display settings and that variable products appear with their options. A good assistant respects every stated constraint or says nothing matches.
Can it tell shoppers what is popular?
Only from your own data, such as sales counts synced through the plugin. Claims about popularity that your store cannot support are small fabrications that undermine everything else the assistant says, so an honest "I do not have that" is the better answer.
Should it recommend on every conversation?
No. A customer chasing a parcel does not want a suggestion, and an instruction to always recommend applies everywhere including there. Recommendations belong in conversations where someone is choosing, which the assistant can tell from the question. Suggesting in a support conversation reads as opportunism and usually ends it.
Do I need a recommendation engine as well?
Not at this size. A grounded assistant narrowing a well-described catalogue covers the conversational case, and your store already has related-product features for the page-based case. Add a dedicated engine only when catalogue size makes manual merchandising impractical, and even then the attributes still have to exist.