List
Most WordPress stores still upsell with a hand-picked “You may also like” block from 2019. Same four SKUs on every product. Same row on the cart. Same ignore from shoppers who already know those items are unrelated.
Recommendations only work when they track what someone actually browsed and bought. That is a behaviour problem, not a “write better product copy” problem.
Generatify includes AI product recommendations built from real browsing and order history, with placements on product, cart, and thank-you pages, and a best-seller fallback on cold start. This post is that surface only. Not the helpdesk agent, not comment toxicity, not alt-text automations.
If you still need the model-connection baseline, read how to integrate ChatGPT, Claude, and Gemini into WordPress. For the “chatbot that acts on the page” angle, see why a chatbot is not enough and you want a page agent.
The problem: static related products pretend to be personalization
Hand-curated related products fail in predictable ways:
- New catalogue items never get linked
- Seasonal products stay related to winter gear in July
- Cart upsells ignore what is already in the basket
- Thank-you pages waste the post-purchase attention window
A behaviour-based row fixes the boring part: show what is adjacent to what this shopper (or shoppers like them) already cared about. Generatify is built for WooCommerce-style catalogues where recommendations, chatbot training on the catalogue, and product-page content generation share one AI stack.
What Generatify actually ships for recommendations
Keep claims tight to the product page.
Built from real browsing and order history, not a static merchant list.
Three placements: product, cart, and post-purchase (thank you).
Cold-start fallback to best sellers when there is not enough signal yet.
Recommendations sit beside the rest of Generatify: admin assistant grounded in your content, helpdesk agent, page agent, smart compose, editorial automations, content generation in the editor, and comment moderation. You connect your own ChatGPT, Claude, Gemini, or Kimi key on Pro and pay the model provider directly. No Generatify AI markup in the membership price. Free has no model connected.
Where to place the row (and why order matters)
Product page. Under the fold near “Add to cart” or below the description. Goal: keep a browser moving to a second SKU without a new search.
Cart. Suggest complements, not duplicates of what is already in the cart. Goal: raise average order value before checkout anxiety kicks in.
Thank-you / post-purchase. Suggest the next logical product or refill while attention is highest. Goal: second purchase or account return, not a second payment wall in the same second.
Turn placements on one at a time. Measure add-to-cart from the recommendation row before you spam all three.
Cold start without looking broken
New stores and new visitors have thin history. Best-seller fallback exists so the module still renders something credible. Seed it by:
- Making sure bestsellers are actually your good products, not clearance junk
- Letting a week of real traffic accumulate before you judge quality
- Avoiding empty states that say “no recommendations” in production
If the catalogue is tiny (under a couple dozen SKUs), expectations should stay humble. Recommendations shine when there is enough adjacency to learn.
Pitfalls
- No API key on Pro. Recommendations that need the model stack will not magically run on free with no provider connected.
- Fighting another upsell plugin. Two recommendation engines on the same template look spammy and dilute click data. Pick one.
- Recommending out-of-stock items. Keep catalogue hygiene tight. A smart row that links to empty inventory destroys trust.
- Putting the cart row above trust signals. Shipping promise and coupon field still matter more than a carousel.
- Ignoring privacy expectations. Be ready to explain that suggestions come from on-site behaviour, not a creepy third-party ad graph, if customers ask.
How this fits the membership pitch
Generatify is not a separate AI subscription. It is one Pro plugin inside DaftPlug: Progressify, Lightify, Generatify, Rankify, Snapshotify on one licence. Plans differ by site count (pricing). Fourteen-day money-back. New Pro plugins join existing memberships without a new line item.
Connect the model before you expect magic
Pro recommendations sit on the same provider connection as the rest of Generatify. Connect ChatGPT, Claude, Gemini, or Kimi with your own API key. You pay the provider directly. Switch models when a cheaper one is enough for catalogue tasks and a stronger one is better for long drafts.
Without a connected key, you are looking at a free install with no model. Do not debug “empty recommendations” for an hour before checking that.
Measure the row like a merchandiser
Treat the recommendation module as a merchandising experiment:
- Note baseline add-to-cart rate on product pages for a week.
- Enable the product placement only.
- Watch clicks on the recommendation row and assisted add-to-carts.
- Then test cart, then thank-you.
If the product row underperforms a simple “bestsellers in this category” block, fix catalogue data and cold-start seeding before you blame the model. Garbage adjacency in the catalogue produces garbage suggestions.
Pair with page agent carefully
Generatify’s page agent can navigate, filter, and add to basket on a visitor’s request. That is a different surface from the recommendation row. Do not turn both on at full aggression on day one if your theme is fragile. Ship recommendations first. Add the page agent when you are ready to set hard limits on what it may do.
Membership talking point when it fits
One Pro licence unlocks Generatify plus Progressify, Lightify, Rankify, and Snapshotify. New Pro plugins join existing memberships. You are not stacking a separate AI SaaS on top of a separate upsell plugin and a separate SEO tool.
Takeaway
Stop hand-picking the same four upsells for every SKU. Use Generatify recommendations from browsing and order history on product, cart, and thank-you pages, keep best sellers as the cold-start safety net, and measure whether the row earns clicks before you call it “AI personalization.”