Conversational Product Search vs Keyword Search: What Works

Your store's search box is very good at one thing: matching the words a shopper types to the words in your product data. That works well when the shopper knows what they want and what it is called. It works badly when they do not, and a lot of shoppers do not.

This article looks at where keyword search and filters fall short, what conversational product search does differently, and how to use both on the same store. They are not rivals. Each one serves a different kind of shopper.

Why keyword search and filters fail many shoppers

Shoppers describe needs, not product names

A shopper types "cream for tight skin after washing". Your product is called "Barrier Repair Cream". Unless its description happens to contain "tight" and "washing", keyword search either finds nothing or returns everything with the word "cream" in it. Neither helps.

People think in problems and situations: "headphones for running", "a gift for a dad who cooks", "shoes for standing all day". Catalogs are written in product terms: model names, materials, feature lists. Keyword search sits right in the gap between the two.

Several conditions at once

Real needs rarely come in one piece. "Dry, sensitive skin, under 30 euros, no fragrance" is four conditions. A search box handles one or two words well. Filters can handle more, but only if your store has a filter for each condition and the shopper finds all of them.

Filters assume the shopper knows what matters

Filters are a great tool for someone who already knows that "ceramides" or "water resistance" is the attribute to look for. A newcomer does not. On a phone, a long filter panel is also tiring, and it is easy to end up with zero results without understanding why.

Search cannot answer the next question

Even when search finds the right page, the shopper often has a question it cannot answer. Is this one fine for sensitive skin? What is the difference between these two? Do I need anything else with it? That is usually the moment they leave to compare elsewhere, or send a support email that gets answered hours later.

What conversational product search does

Conversational product search lets the shopper describe the need in a sentence and get a short list of suitable products, with a reason for each. Here is how AskMerra handles the common cases.

Needs instead of names

Take "dry, sensitive skin, under 30 euros". AskMerra picks out the concern (dry skin), the extra condition (sensitive) and the budget, then searches your catalog. The budget becomes a hard filter, so nothing over 30 euros slips in. Price phrases are understood in all six supported languages, from "under €30" to "sub 100 lei".

This works because every product was enriched when you imported it. AI read each description and recorded use cases, skin or hair types, concerns, key ingredients or specs. A product can match "tight skin after washing" even if those exact words never appear on its page.

Follow-up questions

A conversation keeps its context. After the first suggestions, the shopper can write "anything cheaper?" or "which one has no fragrance?" without starting over. That back and forth is what a good sales assistant does in person, and it is exactly what a search results page cannot do.

Comparisons

"What is the difference between the serum and the cream?" The assistant can look up the details of each product in your catalog and explain the difference in plain words, based on your own product data.

Products that go together

Many purchases are part of a routine or a set. AskMerra works out which products pair well, for example products used at different steps of a routine that address the same concern. A shopper who asks for a serum can learn that a matching cream exists, which is the natural cross-sell a good sales assistant would make.

Questions on the product page

On a product page, AskMerra knows which product the visitor is looking at. "Is this good for oily skin?" works without naming the product.

An example

Shopper: I need something for dry, sensitive skin, under 30 euros.

Assistant: Here are two fragrance-free options within your budget. The serum adds hydration and goes on first, and the cream helps repair the skin barrier. (Two product cards with live prices and stock.)

Shopper: If I only buy one, which should it be?

Assistant: For dry skin, start with the cream, since it covers both moisture and barrier support. You can add the serum later.

Shopper: Do you ship to Austria?

Assistant: (Answers from your shipping policy, with your real costs and delivery times.)

Discovery, a comparison and a policy question in one thread. With a search box, that would have been three separate trips around the site.

Conversational search still uses keywords

Conversational search does not throw keyword matching away. AskMerra combines three sources of candidates for every search:

• semantic search over enriched products, which matches meaning
• keyword search over names, brands and SKUs, which catches exact terms
• tags for concerns and attributes extracted during enrichment

It then applies hard filters such as price ranges and ranks the results by relevance and stock. A shopper who types an exact product name or SKU into the chat still gets that product. A shopper who describes a problem gets products that solve it.

When each one works best

Keep your search box and your filters. Add conversation for the shoppers they do not serve well.

Keyword search and filters work best when:

• the shopper knows the brand, model or SKU
• they are reordering something they bought before
• they want to browse a whole category and scan it visually
• the deciding attribute is simple, such as size or color

Conversational search works best when:

• the shopper knows the problem but not the product
• the need has several conditions at once
• they want advice, a comparison or a full routine
• they have a question the product page does not answer
• they would rather ask in their own language (more on that in multilingual e-commerce customer support)

Practical tips for using both

• Make your catalog rich. Detailed descriptions and attributes (ingredients, materials, compatibility, sizes) improve both keyword and conversational results. When you map your columns in AskMerra, you can turn columns such as skin type or material into searchable attributes.
• Use suggested questions. You can show up to six question chips per language before the first message. Pick the needs your customers mention most, phrased the way they say them.
• Put the assistant where people get stuck. With the JavaScript API you can open the assistant from any button on your site, for example next to your filters or on a search page that found nothing.
• Keep checkout calm. Hiding the widget on checkout pages keeps the purchase flow free of distractions, and it takes a single path rule in the widget settings.
• Read the questions. AskMerra's analytics show the questions shoppers ask most. If many people describe the same need in words your category names never use, that is a useful hint for your navigation too.

Try it with your own products

The difference is easiest to see with your own catalog and your own customers' wording. Create your account, import your products and ask the questions your search box struggles with. To see how the answers stay accurate, read how AskMerra keeps answers grounded in your catalog, or go to the docs for setup details.