πŸ“‘ In this article

  1. What is visual search and why it's not "yet another search"
  2. How it works technically: CLIP and image embeddings
  3. Use cases: which categories work best
  4. Metrics: how much it improves conversion
  5. Challenges: where visual search doesn't work
  6. AI Search Visual Search: when and how much
  7. Waitlist: 20% launch discount
  8. FAQ

A buyer in a furniture store sees a sofa in a cafΓ©, photographs it, uploads to your site. A second later β€” 12 similar sofas from your catalog. That's visual search β€” finding products by image instead of text. In this article: how it works technically, who really needs it, and when it ships in AI Search.

What is visual search and why it's not "yet another search"

Buyers in 2026 photograph products in offline stores, on streets, on Pinterest and Instagram. Classic text search requires them to describe β€” "white leather sofa with armrests". If you have 5 white leather sofa variants β€” buyer still doesn't find the exact one they saw. Visual search solves this differently:

  • Buyer uploads photo (camera, gallery, URL)
  • Algorithm converts image to vector and finds similar vectors in your catalog
  • Results sorted by visual similarity, not textual
πŸ’‘
Key difference: text search answers "what do you call it". Visual search β€” "what do you see". For categories with strong visual component (apparel, furniture, accessories) it's a fundamentally different experience.

How it works technically: CLIP and image embeddings

At the core β€” CLIP (Contrastive Language-Image Pre-training) by OpenAI or one of its open-source alternatives (OpenCLIP, SigLIP). The model was trained on hundreds of millions of "image + text" pairs and learned to convert them into vectors in a shared space.

Indexing workflow

  1. For each product in your catalog, take the main image
  2. Run through CLIP β€” get vector of length 512 or 768
  3. Store vector alongside product_id in DB

Search workflow

  1. Buyer uploads photo on site
  2. Server runs photo through same CLIP model β€” gets query vector
  3. Cosine similarity between query and all product vectors
  4. Top-K nearest products returned

Use cases: which categories work best

πŸ‘—
Fashion
πŸͺ‘
Furniture
πŸ‘Ÿ
Shoes
🌿
Plants
🎨
Decor
πŸ’Ž
Jewelry

Other use cases: industrial parts ("my faucet broke β€” find replacement"), accessories (bags, watches), shoes with thousands of variants.

Metrics: how much it improves conversion

For visual-friendly categories β€” generally seeing +15-30% mobile conversion uplift, time-to-purchase shortens from 3-7 to 1-2 clicks, bounce drops by 15-30 p.p.

⚠️
Caveat: numbers from various e-commerce reports and our own experiments. Depend on category mix, mobile/desktop ratio, brand loyalty. On non-visual categories (books, software, services) β€” no impact.

Challenges: where visual search doesn't work

  1. Quality of catalog photos. If product photos are poorly lit, watermarked, in different angles for same product β€” embeddings will be noisy.
  2. Categories without visual component. Books, digital products, services, packaged food β€” CLIP sees only the box.
  3. Very similar products. If you have 50 white shirt variants differing only by texture β€” CLIP confuses them.
  4. Catalog size. <500 products: visual search works poorly. >5000: starts to shine.
  5. Mobile UX. Photo upload must be one-tap.

AI Search Visual Search: when and how much

Visual Search planned as add-on to AI Search v1.0.5, release Q3 2026.

What's included

  • Upload zone in search form (camera, gallery, URL)
  • Page /visual-search for full-screen mode
  • Auto-indexing of product images via CLIP on our GPUs
  • Faceted filters over visual results
  • API endpoint for custom integrations

Expected price

Add-on ~$17-20/mo on top of base AI Search plan ($29-49 total). Free plan β€” unlimited text search, but visual search limited to 100 queries/month.

Waitlist: 20% launch discount

If you have a store in a visual-friendly category β€” join the waitlist at /visual-search. First 100 subscribers get:

  • 20% launch discount (~$15/mo instead of $19)
  • Early-tester limited free demo on your own catalog
  • Queue priority during launch days

β†’ Join the waitlist (20% off for first 100)

🎁
What waitlist gives right now: 20% launch discount for first 100 subscribers, early-tester access, queue priority. Email only at launch β€” no spam before release.

FAQ

Can I test visual search in AI Search now?

No, not in public release yet. Internal alpha testing on 3 dev stores. Launch Q3 2026.

Will visual search be free for Free plan?

Free plan gets up to 100 visual queries/month for testing. Production volumes β€” separate add-on $17-49/mo.

Do I need to reindex products for visual search?

Yes. CLIP indexing runs separately from text. Time: ~1 second per product. For 10k SKU β€” ~3 hours on our servers.

What about products without images?

Visual search works only on products with photos. For no-photo products β€” fallback to standard AI Search.

Can I query by image URL instead of upload?

Yes. API will support both file-upload and URL β€” useful for marketplace integrations.