Chat EPDocs

ChatGPT isn’t replacing onsite search. Shoppers are just arriving more prepared to buy

Onsite search use grew 30% in a year and long queries grew 71%

Empathy Teamby Empathy Team

Onsite search use grew 30% year on year, and queries of four or more terms grew 71%, according to Empathy.co's search-pattern study across retailers, September 2025 to September 2026. Shoppers may be researching outside of your store, but they're arriving more prepared than ever and go straight to the search box.

+30%

use of onsite search

+71%

queries of four or more terms

Source: Empathy.co search-pattern study, September 2025 to September 2026.

Does onsite search still matter when shoppers use ChatGPT?

Yes, more than ever. Many expected ChatGPT, Claude and Perplexity to take over shopping, with people researching and buying inside the assistant. Empathy.co's search-pattern study shows a different trend: onsite search use grew 30% year on year (September 2025 to September 2026). Shoppers do their research in AI assistants, then come to the store to buy.

Where do shoppers research, and where do they buy?

Shoppers research in ChatGPT, Claude, Perplexity and Google's AI answers. They ask what to buy, compare options and leave with a specific product in mind. Then they come to the store to make the purchase. The research happens in the AI assistant. The purchase happens in your store.

So the shopper who arrives has the research done and a high intent to buy. They know what they came for, and the first place they go is the search box. That makes the search box more important than ever.

How much has onsite search use grown?

Onsite search use grew 30% between September 2025 and September 2026, across the retailers in Empathy.co's search-pattern study. A shopper who arrives with a product in mind goes straight to the search box and types.

Why are search queries getting longer?

Search queries are getting longer because shoppers have learned to write prompts. A year of asking AI assistants full questions has changed what people type everywhere. Queries of four or more terms grew 71% in the same period (Empathy.co search-pattern study, September 2025 to September 2026).

What does a 2026 query look like?

A 2026 query carries several intent signals in one line. The 2025 version of the same need carried one.

20252026
Example queryantiage creamanti-wrinkle serum with vitamin C and hyaluronic acid
Terms27
Intent signals14: product type, concern, two ingredients
Where the shopper did the researchIn the store, by browsingBefore arriving, in an AI assistant
Where the shopper buysIn the storeIn the store, starting at the search box
What they expectA list to browseThe product they came for

What happens when a long query hits a keyword engine?

When a long query hits a keyword engine, the engine tries to match every word. It finds few products with all of them, so it returns nothing, or drops words and returns a loose list. The shopper gave four signals and got a response that used one. The better the shopper's research, the worse the result.

That is the gap. Shoppers changed how they search in twelve months. Most search boxes were built for two-word queries.

What should retailers change?

Retailers should change what happens before the results page is built. A long query needs to be read as intent: a product type, a concern, attributes. A question needs an answer. That means putting the AI in front of the results page, where it shapes the whole response.

AI-First means the AI comes before the results page.

What does the AI in your search box need to get right?

A prepared shopper makes one first search, and that search decides whether they find what they came for. If an AI model reads every query before the results page is built, it has to be there for every shopper, understand them the same way each time and keep what they tell you private. Where the model runs decides all three.

It has to stay on when demand peaks. The shoppers with the clearest intent arrive in the busiest weeks. Search use is up 30% and queries are getting longer, and most AI APIs charge by the word, some more at peak hours. That pushes retailers to switch AI features off on the days they matter most. Empathy.co runs its own models at a largely fixed cost, so a query on Black Friday gets the same treatment as one in February.

It has to understand the same request the same way. A shopper who knows what they want expects the store to recognise it, this week and next quarter. When an AI provider updates or retires a model, the answers change with it, and merchandising rules built around them can stop working. Empathy.co fixes the model version and only changes it on a published schedule, after testing.

It has to keep what shoppers tell you. A seven-word query reveals the concern, the ingredients and sometimes the budget. That is personal data, and some of the most valuable demand data a retailer has. Empathy.co processes it on its own infrastructure in the EU, under a single data processing agreement, with no outside AI provider in between.

AI-First Commerce Search was pioneered by Empathy.co, the commerce search company behind search for over 3,000 brands. Read how it compares in keyword vs semantic vs AI-first search.

Three checks to run on your own search logs this week:

  1. The share of queries with four or more terms, this year against last.
  2. The zero-results rate on those queries.
  3. The share of queries phrased as questions.

And three questions for whoever runs your search: where does the AI run, who decides when it changes, and does it stay on at peak?

Frequently asked questions

No. Shoppers research in AI assistants and buy in the store. Onsite search use grew 30% year on year in Empathy.co's study, as shoppers arrive ready to find a specific product.

Are shoppers buying in ChatGPT instead of online stores?

Empathy.co's study shows a different trend. Shoppers use ChatGPT, Claude and Perplexity to research, then come to the store to buy, and onsite search use grew 30% year on year.

What is a long-tail search query?

A long-tail search query is a specific query with several terms, such as a product type with attributes. In Empathy.co's study, queries of four or more terms grew 71% in a year.

How does ChatGPT shopping affect retailers?

Shoppers arrive later in their decision and with a more specific request. The store's search has to understand that request the first time.

What is zero-click search in ecommerce?

Zero-click search is when a shopper gets their answer from a search engine or AI assistant without visiting a site. For retailers it means more of the research happens off-site, and the shopper who arrives has a high intent to buy.

It can, if the AI is billed by the word: more searches and longer queries both add to the bill. With self-hosted models the cost stays largely fixed, so AI features stay on at peak.

With Empathy.co, queries stay on Empathy.co's own infrastructure in the EU, under a single data processing agreement.

Free Search Audit

See how your Search handles the queries shoppers type now

We run your store's real queries, long ones and questions included, and show you where the results page falls short.