Smart embeddings — Standard plan
Your products speak more languages than your customers do.
AI Search Smart Embeddings uses an LLM to teach each product five ways customers might search for it — before the vector is even generated. “Cheap sofa” starts matching “affordable couch”. “Linen trousers” matches “beige pants”. Done at publish time, cached forever, zero latency at search.
Start 7-day free trial →Card required for the trial. Cancel anytime before day 7 at no cost.
Default embeddings index what the product says. Smart embeddings index what customers mean.
AI Search ships with semantic matching out of the box. Smart embeddings go a layer deeper — they expand your product descriptions with synonyms, descriptive terms, use-cases and common misspellings before the embedding is generated. The resulting vectors cluster in exactly the parts of semantic space where real customers search.
What enrichment actually looks like
Raw product
Hoodie Oversized Black. Heavyweight cotton pullover.
After Smart embeddings
Hoodie Oversized Black. Heavyweight cotton pullover. · chunky sweatshirt · baggy pullover · streetwear hoodie · relaxed-fit hooded top · black hooded sweater
All of those phrasings go into the same vector. The customer’s actual query lands near the matching one — even if they used a word your product title never had.
Three queries, three stores, three lost sales — recovered.
Fashion
“beige pants”
Standard search → 0 results. Smart embeddings → matches your “Linen Trousers, Natural” and “Chino Slim-Fit Taupe” products that never used the word “beige”.
Home
“cheap sofa”
Standard search → nothing (your SEO team banned the word “cheap” from titles). Smart embeddings → matches your “Affordable Couch Collection” and “Budget Loveseat Range”.
Beauty
“something for breakouts”
Standard search → zero. Smart embeddings → matches your acne cream, salicylic cleanser and spot-treatment serum — none of which contain the word “breakouts”.
How it works
- 1 You publish or save a product. The publish hook fires exactly as before. Nothing new in your admin flow.
- 2 Our LLM reads the product title + description + attributes. A small, fast model (gpt-4o-mini) generates 3-5 alternative phrasings customers might realistically type. No invented facts — only synonyms, descriptive terms and common misspellings.
- 3 The enriched text is embedded and stored. The resulting vector covers both the literal product text and every alternative phrasing. All of this happens once per product and is cached forever.
- 4 A customer searches — your store wins the match it was losing. No LLM call at search time. No latency added. The visitor sees the same instant results page as before — just with the previously-invisible products now surfacing.
Zero cost at the search path. Fixed cost at publish.
The LLM enrichment is paid once per product, at publish or sync time — never again. Your product vectors stay in cache forever. Your visitors’ searches cost exactly the same as they did before Smart embeddings were turned on. The only thing that changes is which products the vector matches.
Try Smart embeddings for 7 days, free.
Install the free plugin, open the setup wizard, hit “Start 7-day free trial” on the Smart embeddings card. Your whole catalogue syncs with enrichment on the very first run. If it’s not for you, cancel from inside the plugin before day 7 — no charge.
Install free from WordPress.org →After trial: €14.90/month on the Standard plan. Cancel anytime.