Chat EPDocs
ProductsAI Search

AI Search that keeps every query moving

Empathy’s AI Search gives you a collection of state of the art AI features that strengthen keyword search by guiding shoppers, growing baskets, and recovering intent without replacing the search experience shoppers are familiar with.

Related Tags

Related Tags close the gap between what a shopper searches and what they actually meant, turning a broad first guess into a precise one in just one click, without the shopper ever needing to find the right word themselves.

Related Tags refining a search with one-click suggestions
After the first search

Refine without retyping

Shoppers narrow a broad search down to exactly what they meant — without retyping, and without needing to know the right word before they searched.

Most shoppers don’t start with the exact term for what they want — they start broad, and only recognize the specific version once they see it in front of them. Related Tags appear alongside the first results, complementing the original query with more precise terms — pulled from your catalogue and how other shoppers have refined the same search. One click launches a new, sharper search, so the shopper never has to find the words themselves.

Example

A shopper searches “washing machine” and lands on hundreds of results — different sizes, brands, loading style, all mixed together. Related Tags surface the terms that actually split that list apart: front-loading, LG, Bosch. The shopper doesn’t need to know which machine they want before they search, they just need to recognize it once it’s in front of them.

Next Queries

Next Queries keep momentum going at both ends of the journey: as shoppers type, by surfacing what to search for next before they’ve even hit enter; and after an add to cart, by picking up where the add to cart left off, so discovery and basket-building feel like a natural progression.

Next Queries suggesting follow-up searches
01As they type

In the search box

Shoppers see what to look for next as they type.

Next Queries in the search box turn a shopper’s first few keystrokes into a shortlist of likely next searches, built from your catalogue and how other shoppers have searched before. They lower the cost of the hardest part of any search — deciding what to type — and keep momentum high before the shopper has committed to anything.

02After add-to-cart

In the results

After an add-to-cart, shoppers get the next mission.

When a shopper adds something to cart, Next Queries appear in context to complete what they came to do. This is where basket-building happens: not a generic cross-sell, but the next step in the same mission — grown from what shoppers who bought this also went on to search for, and shaped by your merchandising strategy where it matters.

Example

A shopper adds an ice cooler to their cart. Next Queries surface the searches that typically follow it, matching camping stoves, drinks and snacks in the same line. The next mission is already in front of the shopper before they’ve had to think of it, so the basket keeps growing without the need to search again.

Related Prompts

Related Prompts turn catalogue navigation into a conversation: instead of asking shoppers to decode filters or guess at a more specific search term, the system asks a simple question — and each answer either narrows what’s in front of them or opens the next products worth exploring.

Related Prompts showing conversational suggestions on search results
01When results feel too broad

Ask, don’t make them guess

Shoppers narrow hundreds of results down with one answer — no filters to decode, no better search term to think of.

Related Prompts read shopper intent and ask the question a good shop assistant would ask — “gold or silver earrings?” — instead of leaving shoppers to translate what they want into filter checkboxes or more precise keywords. One answer narrows the results to exactly that.

02Once they’ve found something

Keep the conversation going

After narrowing down, shoppers get asked what’s next — not left staring at a static grid.

The same conversational format that narrows a search can reopen it. “Want a matching ring?” isn’t a filter — it’s the next question in the same conversation, continuing discovery instead of ending it the moment the shopper clicks once.

Example

A shopper searches for the book “Atomic Habits”. As they explore results, they’re given the opportunity to explore more of the catalogue around related products. When they click on a related prompt it opens a standalone results page within the current search, allowing them to explore more of the catalogue and add products to their cart. Same conversational thread, but now it’s opening the catalogue back up instead of narrowing it down, with no new search typed and no filters touched, just a question answered.

AI Spell Check

Search can fail two ways: it can return nothing, or it can confidently return the wrong thing. AI Spell Check is built for both — recovering typos that would otherwise dead-end, and catching corrections that would guess right on spelling but wrong on product.

AI Spell Check runs two checks, not one. The lexical layer proposes a correction. The semantic layer verifies it against what the shopper actually meant — stepping in either when the first check finds nothing, or when it’s about to confidently land on the wrong product.

AI Spell Check correcting a misspelled search query
01The first check

Lexical

Proposes the correction

Matches against your own catalog’s embeddings, not a generic dictionary, so “blu shrt” still resolves to blue shirts — including typos on a word’s very first letters, where most spell checkers give up.

02The second check

Semantic

Verifies the intent

A spelling match isn’t the same as the right match. When two products share an almost-identical name, this layer reads intent rather than characters, confirming the first check’s correction actually means what the shopper meant — not just what looks closest on the page.

Example

Two pharmacy products can share almost identical names and do completely different things. The first check might “correct” the typo and confidently land on the wrong one — and the shopper would have no reason to doubt it. The second check catches that before it happens, protecting the shopper from a bad match and the retailer from the return.

AI Spell Check correcting Haspirina to Aspirina on a Farmacias del Ahorro search

AI Carousels

A zero-results page doesn’t mean you don’t have a product the shopper wants, and shoppers should be guided like they would by a shop assistant to alternatives they’re interested in. AI Carousels catch that gap: when a search comes back empty, Empathy’s AI Search still surfaces related products so the discovery doesn’t end.

AI Carousels showing product rows on a zero-results page
When a search returns nothing

Keep discovery moving even when no products match the search

Shoppers who hit zero results still find something worth looking at — no retyping, no rephrasing, no leaving.

When a search returns no matching products, AI Carousels activate automatically — no setup required. Rather than showing a blank page, Empathy Platform surfaces horizontally scrollable carousels of products that match what the shopper meant, even when the exact phrasing doesn’t exist anywhere in your catalog. Each carousel is labeled with the intent it’s answering, so the connection is obvious rather than mysterious.

Example

A shopper searches “lego puzzle for 6 to 8 year olds” on a toy store that doesn’t list any products by that name. Nothing matches, but that doesn’t mean the store doesn’t have any relevant products. AI Carousels surface what does exist under different names: lego duplo, lego creator, lego ninjago — labeled clearly by the intent they answer, so the shopper sees why they’re there. The search returned no exact matches from the catalogue, but the discovery didn’t end.

Empathy's approach to AI

Empathy AI runs on self-hosted GPUs and open-source LLMs on proprietary infrastructure. No AWS, no OpenAI tax, no data leaving the building.

Compute

Dedicated GPU cluster

Powered by the solar canopy on the Empathy AI building. No public cloud, no shared tenancy.

Models

Open-source LLMs only

Built on open-weight models Empathy uses and contributes to — auditable, replaceable, yours.

Data

Privacy-first

Your documents and prompts are processed locally. Nothing is sent to third-party AI providers.

Ops

Continuous oversight

The stack is monitored, evaluated and adapted continuously — transparency by design, not by audit.

Empathy AI building

Your shoppers' intent and preferences are yours. Keep them that way.

Whatever your industry, the search bar is the most honest expression of intent and preference a shopper ever gives you. What they actually want, in their own words, at the moment they want it. Run that through Big Tech's AI and it doesn't just answer your shopper, it trains someone else's model too. Empathy Platform keeps it inside your walls: your infrastructure, your rules and your data never leaving your control.

The intent

Your shoppers are telling you everything

Millions of queries reveal exactly what your shoppers want and prefer, query after query. Keep that intelligence on your side of the table, not inside someone else’s model.

The logic

Your rules run the ranking

Promotions, margins, seasonal strategy, brand agreements — commercial decisions applied to every result, configured by your teams, not a generic algorithm.

The edge

What your data knows, no one else does

Years of shopper behaviour and merchandising decisions are competitive advantage, not just data. Run them through a generic model and that advantage stops being exclusively yours.

FAQ

Questions about AI Search

Next step

See AI Search on your catalogue.

We’ll walk through your real queries and show how Related Tags, Next Queries, Related Prompts, AI Spell Check and AI Carousels create better routes to increased baskets and conversion.

Request a demo