70.22 %
Around 70 percent of online shopping carts are abandoned
average across 50 separate cart abandonment studies collected by the Baymard Institute
In conversation, XICBOT shows real product cards with image, title, price, rating and delivery time — and a button that leads into your shop's cart or to the product page.
The assistant shows cards for your products in the chat. Each card has an image, title, price, rating and delivery time. A button leads to the cart or to the product page. In this way a question turns into a tip or a comparison. Each card comes from your catalog. What is not in it, the assistant does not name.
Request a free demo To the shop assistant
A question turns into a recommendation, a row, a comparison or a bundle. Every card is built from your catalog: what is not in it, the assistant does not name.
Visitor:I am looking for a rain jacket for hiking, size M.
Assistant:Three models match and are available in M.
Visitor:Show me the lightest one.
Assistant:Here it is. The button leads to the cart address stored for this item in the catalog.
Visitor:How does it hold up against the brand I know from the high street?
Assistant:That one is not in your catalog, so I will not contrast it. I am happy to compare the two lightest models that are stored here.
Ask it yourself, Assistant of this website
Automated assistantA classic product search forces visitors to click through filters, lists and categories until they find the right product — or give up first. XICBOT turns that around: whoever describes what they are looking for in the chat receives not a text answer with links, but real product cards directly in the conversation. Each card shows image, title, price, rating and delivery time and depending on the setting, carries a button into your shop's cart or to the product page; if no address is stored for it, the card has no button. An item the catalog lists as unavailable does not appear at all. That turns a passing question into a concrete purchase impulse without the visitor having to leave the page. Product cards are the most visible building block of the shop assistant — where advice turns into a sale.
Which display appears is chosen not by the wording of the answer but by the result of the query. The assistant has four queries at hand — search, compare, bundle, complement — and from them the chat window draws five different blocks. All five are built from the same catalog you supply; with a connected shop based on Shopware Community Edition, the assistant asks it for the price and availability of the items shown at answering time. Around 70 percent (Baymard Institute) of online carts are abandoned — the shorter the path from question to cart, the fewer purchase intentions are lost along the way.
One concrete recommendation with image, title, price, rating and delivery time. It appears when the query has exactly one hit or when a particular item is meant.
Several matching items side by side when the query has more than one hit. The row sits below the answer rather than inside it, so the text stays short.
Two items in a table, both cards below it. The buttons deliberately sit under the table and not in its cells — a table with controls inside data cells is hard for screen reading software to interpret.
The item in question plus the accessories stored against it in the catalog. The anchor card is marked, because a bundle without a recognisable main item would be just a row.
Complementary items for the current piece, again with an anchor card. The references are stored on the item in the catalog — they are not guessed from the conversation.
Any of the five displays can carry a crossed-out price and a saving. They appear only when the catalog provides both and the old price really is above the current one.
If a condition is missing, the display falls back rather than the answer falling away: a comparison without a second item becomes a single card, a bundle without stored accessories becomes the anchor card alone. On top of that there is a ceiling — at most five tool calls per answer, and only one when the monthly allowance runs low. Anyone asking for six comparisons in a single question will not get them all in one answer.
70.22 %
Around 70 percent of online shopping carts are abandoned
average across 50 separate cart abandonment studies collected by the Baymard Institute
Every product card bundles exactly the information that matters for a purchase decision and leaves out everything that distracts. Instead of a long product page with menu, footer and advertising, the visitor sees only the essentials in the chat — compact, readable on mobile and with one clear action. Which details a card carries depends on what the catalog holds for that item: a line without a value does not appear as an empty line, it does not appear at all.
The price leaves the catalog twice: as a number and as ready-set text. Both the calculation and the formatting happen on the server, so the visitor's browser does not show a different number from the one that was calculated. The card itself is a link or a plain frame, never a button — a button with a link inside it breaks in Safari and Firefox. The action button therefore sits beside it and names the item title when read aloud; otherwise six cards would mean hearing the same sentence six times.
On a small screen, clicking through filters and lists is the most tedious, and that is exactly where the card behaves differently from a product page: below 560 pixels of window width the chat window switches to full screen, the recommendation row scrolls horizontally and carries a keyboard marker so it is not reachable by finger alone. The loader reads the width class from the window width of your page, not from the browser identification — a tablet held upright therefore counts as what it currently is.
The words the chat window contributes itself are held in German and English and follow the language of the conversation: Add to cart, View product, Recommendations, Total. What comes from the catalog — title, description, features — stands in the language it was supplied in. A catalog with German titles shows German titles in an English conversation too; whoever needs both supplies both. The rating goes out as a bare number and only becomes a sentence in the chat window, because 4.5 and 4,5 are the same rating in two notations and only the visitor's browser knows which one applies.
Short Paths Sell Better The more steps lie between question and purchase, the more purchase intentions are lost along the way. The card cuts the path down to one button: it opens your shop's cart or the product page in a new tab, and the conversation stays where it was — the visitor can keep asking without starting over.
A product card comes about in three steps: the catalog is supplied, a query builds cards from it, and everything beyond that waits for the visitor's approval. The separation is deliberate — reading costs nobody anything, acting does.
The assistant's catalog is a store of its own per tenant. Your range is supplied into it — title, image, price, crossed-out price, features, rating, delivery time and availability — keyed on your shop's item reference, so a second run of the same file overwrites instead of duplicating. The connection to a shop based on Shopware Community Edition is set up in the project. What is not in this store, the assistant cannot name, and that is exactly its purpose — more on the knowledge base beside it under shop assistant.
The button on the card carries the address stored for that item in the catalog and opens it in a new tab. The second path is the cart tool: it assembles the basket inside the conversation — set, add, update, remove, clear, voucher — and the basket is held by XICBOT, not by your shop. “Continue” leads to your shop's address with a random pickup token; through a read-only pickup address at XICBOT your shop can fetch the basket with that token, provided it has connected this pickup through an integration of its own. The assistant keeps no price or stock truth of its own: the total is calculated with the live price from your shop, or else with the last sync and its date. More on this under cart assistant.
The search in the catalog only reads and runs without asking back. Everything beyond that — a change to the basket — first appears as a proposal with two buttons, Confirm and Reject. The text the visitor reads there is stored when the proposal is made and not rebuilt later: what was promised still stands even if the catalog has changed in the meantime.
Every call then stands in the log, with input, result, duration and the path on which it was approved. In the conversation itself the visitor sees a shortened version: duration and approval path are dropped there, and a technical error text is replaced by a general sentence. If a tool reaches outside — into an inventory system in the Custom package, say — the same route applies, with target checking, a deadline of eight seconds (adjustable between one and twenty) and a cap of 64 KiB; anything above that is discarded rather than used in truncated form.
8 of 8
questions without a basis declined instead of guessed
eight questions deliberately outside the material; the wording of every answer is in the measurement file
Six steps lie between the question being sent and the card in the transcript. They are set out here not because the path is complicated but because each step carries a decision that can be read up later — including when no card appeared at all.
Only what the conversation's role may call and what is switched on is offered. If the list is empty, no tool field goes out to the model at all; a muted conversation receives none in any case.
Input and time are recorded before anything happens. A call that fails is therefore just as findable as one that succeeds — otherwise only success would be noticed.
The search runs strictly within your tenant's catalog. A stated item reference beats the search term; without either, the tool does not search but returns a note. If it finds nothing, the message back to the model says explicitly not to invent a product or a price.
To the model as data for the sentence of the answer, to the chat window as cards together with their form. That is why text and card stand side by side instead of the same list appearing twice.
The form turns into the display: single card, row, comparison table or bundle with an anchor card. If no card is present, the block is dropped entirely — an empty frame would be worse than none.
The search only reads and runs through. Handing over a basket stands in the transcript as a proposal until the visitor confirms or rejects — and stays there if they do neither.
Two steps carry the whole construction. The call sits in the log before it runs — which makes a failed request as findable as a successful one. And the result goes out twice: to the model as data, to the chat window as a card. That is why the sentence next to the card has the same data in front of it, and prices, quantities or deadlines in it that appear in no source are replaced with a placeholder by the check.
If no card is available, the block is omitted entirely. The assistant does not build an empty shell that something might fill later — a card without an article would be a claim about your catalogue.
Visitor:Take two of those and the matching cleaner with them.
Assistant:The cleaner is stored as an accessory on the device, so I have added it.
Visitor:And my voucher from the newsletter?
Assistant:I pass it on with the basket. It is not applied here — whether it holds is decided by your shop at redemption.
Visitor:Then hand the basket over.
Assistant:Please confirm once. “Continue” then takes you to the shop in a new tab.
Every question appears in the conversation list with channel and time.
The switch for answers without cover shows where the product data fell short.
The filtered selection can be handed on as CSV.

Product cards in the chat are not an off-the-shelf standard feature but are tailored to your shop, your assortment and your goals. The path there is manageable and begins with a free initial consultation and, on request, a demo with a test assistant for your own website.
In the free initial consultation we clarify which products should appear in the chat and which actions make sense — from the single recommendation to the bundle. On request you receive a free demo with a test assistant for your own website so you can experience the product cards before you decide.
Your range moves into the assistant's catalog: title, image, price, crossed-out price, features, rating, delivery time and availability, keyed on your shop's item reference. At most 500 items go in per call, and the same extract supplied twice overwrites instead of duplicating. For a shop based on Shopware Community Edition the catalogue is taken from the shop and synced on a schedule, and the assistant queries price and availability of a named item when answering; with other shop systems, how often the import runs decides how fresh the card is.
Without an address on the item there is no button on the card. Using an extract of your range we check whether address, image and price are filled in per article before the first card appears in the chat.
Together we define which displays appear — recommendation, comparison, bundle, complementary items or sale card — and which actions the assistant may carry out with approval, such as putting together a basket. That is also where it is decided whether a card's button leads into the cart or to the product page: both are stored per item, and without an address the card stays without a button.
We put the questions you really receive to the assistant and look at which presentation it picks. If no card is available the block is omitted entirely — and that is exactly what a trial run makes visible, before a visitor does.
After launch we take over hosting in Germany, operation, care and updates. Through conversation analytics you see which products are asked about, which questions the catalog did not cover and where conversations end — and that, rather than guesswork, shapes the next import.
The order cannot be shortened: without a catalog there is no card, and without an address stored on the item there is no button on the card. We check both before launch on an extract of your range — ten real items show which field is missing from the catalog faster than a list of field names does.
Product cards are not a switch you flip. They need a catalog, an import that can be repeated, and the decision which card carries which button. That is why a one-off setup stands beside the monthly operation.
from 149 € per month net
Minimum term: none, cancellable monthly
All prices net, plus VAT Source: XICBOT price list, as of 11 September 2026
Product cards, cart in the chat and shop connection are part of the Pro package (operation from 149 € per month net, setup from 2,900 € net). For tool and API control, custom actions and individual cases there is the Custom package (setup from 4,900 € net, operation by scope). Every project begins with a free initial consultation and, on request, a free demo. Ongoing services are booked individually and can be cancelled monthly; there is no minimum term.
We load an excerpt of your range into a test assistant and show how a question turns into a product card with image, price and an add-to-cart button.
The difference shows in everyday use. A plain text answer with links shifts the work to the visitor: search, click, switch to the product page, back to the chat. Product cards bring the decision to where the question arises.
Both columns share one condition: without an entry in the catalog there is no card. An item that was never supplied does not appear in the chat — not even when it is in the shop. That is the flip side of the very rule that keeps the assistant from naming a product that does not exist.
The following journeys are anonymised and illustrative. They show how product cards in the chat take on concrete tasks in the shop — from comparing two favourites to matching accessories.
More on Use in RetailHow product cards work together with advice and the cart in the same conversation is shown on the page for the assistant for online shops. How the cards look with your own items we show on request in a free demo, without sales pressure.
More on this in the guide: product cards and cart in chat describes the structure and the limits of the cards in detail, and guided selling in chat explains how the assistant derives a fitting suggestion from just a few follow-up questions. Why machine-readable product data is becoming even more important is set out in agentic commerce.
Two questions come up in almost every initial consultation: where the data in the card comes from and what the assistant may do without asking back. Both are covered at length above; here are the short answers, each with a path to the long one.
A product card is a clickable product tile that XICBOT shows directly in the conversation: with image, title, price, list price and saving, rating including the number of votes, delivery time in days and features — plus a button leading to the address your catalogue stores for the card — your shop's cart or the product page. Instead of a text answer with links, the visitor sees a concrete, purchasable product right in the chat. Depending on the question, the assistant shows a single recommendation, a row of suggestions, a comparison of two items or a bundle. At most six hits are shown, and only what is in your own catalogue is found.
XICBOT answers from the product catalogue — title, image, price, variants, availability, delivery time, rating and promotional prices. What is not in the catalogue the assistant does not show as a card, and it is instructed to invent neither product nor price. For a shop based on Shopware in the Community Edition the catalogue is taken from the shop and synced on a schedule; the assistant queries price and availability of a named item in the shop at the moment of the answer, and if the shop cannot be reached, it names the last synced state as such. With other shop systems you deliver the catalogue, and how current a card is depends on how often that happens. If someone puts together a cart in the chat, the price is calculated in exactly one place, so the chat cannot show a discount the handover does not know about; a voucher is redeemed in the shop. More on the shop assistant page.
Into a cart in the chat that sits with XICBOT, not in your shop. The cart tool type puts it together: it adds line items, changes the quantity, removes a position, clears the cart and applies a coupon — six kinds of change, at most 99 units per position. In the chat the result appears as a card with line items, total and a shipping line. Each of these changes writes and therefore requires confirmation: the visitor reads the proposal in plain text, only their click executes it, and a one-time key made of conversation, tool and sorted input prevents a second click from changing anything twice. The total is calculated with the current price from your shop or, if it cannot be reached, with the last synced price and its date; shipping costs and the free-shipping threshold come from the rules you store with the assistant. For the move into the shop, “Continue” opens your shop's stored address with a one-time pickup key; with it your shop can fetch the cart from XICBOT through a read-only pickup address, but it has to integrate that pickup itself. No order, customer account or checkout is created in the shop by this. Details under cart assistant.
Yes. Besides the single recommendation, the assistant can contrast two products directly — price, key features and rating side by side — and suggest matching accessories as a bundle or complementary items in the sense of customers also bought. A comparison needs two items: if only one remains, it is shown on its own instead of faking a comparison. At most six hits are shown per answer, and only what is in your own catalogue is found — not even via a reference from another card. Which card types and suggestions appear is defined together for your assortment.
The room for that is drawn tight. For every question at most five search hits from your knowledge base, plus your text blocks, go to the model, numbered, and sourced answers name the excerpts they rely on. Figures, prices and deadlines that are not in your sources are replaced with a notice. If the search finds nothing or the model declines, the assistant says that nothing on this is held in the knowledge base — and the case is logged as a knowledge gap, as is an answer without a used excerpt. Its rules forbid it to invent contact details, file numbers or appointments, or to pose as a human being. An error cannot be ruled out entirely with any language model; hence the binding to sources, the visible citation and the handover to a person.
Yes. Product advice, product cards and the cart work across languages. Foreign-language customers also get recommendations and are guided through the chat into the shop, in a shop based on Shopware in the Community Edition. The assistant detects the language in its own program code and replies in the language of the most recent question; your catalogue remains the only source, because the knowledge base is not filtered by language. So you maintain product texts once and need no translated second version. Details are available with the shop assistant.
Product cards and the cart in the chat are part of the Pro package: operation from 149 € per month net, setup from 2,900 € net. For tool and API control, custom actions and individual workflows there is the Custom package from 4,900 € net setup, with operation there by scope. Which package fits is decided not by the single function but by the scope: Pro allows five assistants, 50 knowledge sources and 2,000 documents, Custom 25 assistants, 250 knowledge sources and 20,000 documents. We name a binding price after a free initial consultation. All packages are listed under pricing. All services can be cancelled monthly; there is no minimum term.
We work with Shopware in the Community Edition, an open-source system. On this basis XICBOT takes the catalogue from the shop, syncs it on a schedule, queries price and availability of a named item when answering and builds the product cards. Further accesses run through shapes with a stored address once your inventory system is connected in the Custom package: order status and stock enquiry read, a price lookup reads as well, an order change writes and therefore asks before it is carried out. Every call leaves a log row that is written before execution, and knows seven states from proposed to deadline exceeded. If you have a special setup or your own systems, the connection can be built over defined interfaces — more on this under custom functions.
If your question is missing, ask it in the free initial consultation. When we do not know something, we say so — it is the same rule the assistant works by.
From Your Catalogue
A card shows only what is maintained in the catalogue for the item. For your request it therefore helps to name a few typical items from your range and which fields are maintained there, such as delivery time, features or ratings. From that we can see how the cards would look in your shop and where details are still missing in the catalogue.
Tell us briefly about your shop and your assortment. You receive a clear assessment and, on request, a free demo with a test assistant for your shop. You can reach us by phone at +49 5123 9579000.