Guide · Product data
AI product recommendations suggest other items to a shopper based on what they're looking at, what other people bought, or what they ask for. On a small store they only work when the product data underneath is clean: one record per product, real attributes like style, diet, size, or fit, and stock that matches the shelf. Start with hand-picked pairings for your best sellers, fix the data, then let AI fill in the rest.
Most recommendations on a store website come from one of three places.
1. Bought together. "Customers who bought this also bought." These are learned from your order history.
2. Similar products. "You might also like." These match products on their descriptions, attributes, or collections.
3. Conversational. A chat assistant asks what the shopper wants, then suggests products. "A gift for a cheese lover under $50" turns into a short list.
Each one reads different data, and each one breaks in its own way.
| Kind | What it reads | What breaks it |
|---|---|---|
| Bought together | Order history per product | Too few sales per product, duplicate records splitting the history |
| Similar products | Descriptions, attributes, collections | Empty or copied descriptions, missing attributes, messy collections |
| Conversational | Attributes, prices, live stock | Facts buried in descriptions, stale stock, no rule for "I don't know" |
Shopify is a good example, because its rules are published. According to Shopify's help page on product recommendations:
Two takeaways for a small store. If a product hasn't sold much, the bought-together signal is thin, so the system leans on your descriptions and collections. And if your stock counts are wrong, the rules that hide sold-out items can't help you.
A small store's advantage is that the staff already know what goes together. The wine that works with the cheese. The sauce that goes with the pasta. That knowledge is better than a guess from a few dozen orders.
So start there:
Then let automatic recommendations fill in around them.
The same cleanup helps every kind of recommendation, plus your search, your filters, and Google:
That's the work on our product data cleanup and enrichment page.
Wine and spirits. Pairings and "if you like this, try that" by style, region, and price. A recommendation for a bottle you don't have is worse than none. See wine and spirits stores.
Specialty food. Gifts by occasion and budget, and swaps that respect a diet. A "you might also like" that suggests a product with nuts next to a nut-free one needs allergen fields to get right. See specialty food stores.
This is an illustrative example with made-up details, not a client story.
Once a month, open your top 20 product pages and look:
If any of those fail, the fix is almost always in the data, not the recommender.
Hauslight is the AI chat assistant AgentHaus adds to websites. Here's what it does:
Like any recommender, it's only as good as the product data behind it. More on that in why your AI chatbot gives wrong answers.
Not sure what your recommendations are reading? Get a free store check. We'll run real customer searches on your site, check how your product data reads to Google and AI tools, and tell you the three fixes we'd make first.
They can, but the kind that learns from purchase history needs sales on each product, which a small catalog may not have yet. Hand-picked pairings and clean product attributes usually do more at first.
Clean product names, attributes customers care about (style, diet, size, fit), accurate categories, and live stock. Bought-together recommendations also need order history.
Usually because stock isn't synced, products are duplicated, or the attributes the recommender reads are missing or wrong. Fix the product data and the stock sync before tuning the recommendations.
Start by hand-picking for your best sellers, using what your staff already know goes together. Let automatic recommendations fill in around them once your data is clean.
Growth creates friction. We remove it.
