Guide: AI in business › AI in sales and business departments
AI in e-commerce earns its keep most reliably in three places: product content, customer service, and whether the shop appears at all in the answers of AI assistants, where customers are starting their product searches. Since May 2026, Allegro has been available as an app in ChatGPT. Shopping through a conversation with AI is therefore no longer theory. Below we explain where AI genuinely helps an online shop, what it costs, how to prepare your shop and what the law says.
In 60 seconds Five things for shop owners
- In one sentence: AI in an online shop covers everything from product descriptions and customer chat to the visibility of your offer in ChatGPT and other assistants.
- The biggest winners: shops with a large catalogue, frequent customer questions, or selling in several markets or through a B2B channel.
- Costs: some tools are included in the price of e-commerce platforms, and AI chat is often billed per resolved case. Most of the work goes into tidying up product data.
- Starting point: product data – complete names, attributes, photos and descriptions. Every AI tool works on this data.
- What to watch out for: the chat must say that it is a program, and the shop is responsible for what AI has written about the product and the terms of purchase.
What is AI in e-commerce?
Artificial intelligence in e-commerce covers all the uses of AI that help a shop sell: generating and translating product descriptions, a chat that answers customers, recommendations, a site search that understands questions asked in the customer's own words, review analysis and, increasingly, the presence of the offer in AI assistants such as ChatGPT. The shop is only one of the areas in which AI is changing how a company works; the others are described in guide to AI in business.
This last change is the newest. Customers ask a language model (LLM, i.e. a program that understands and produces text; hence the term ecommerce LLM): “which trail running shoes under PLN 400?”, and get ready-made suggestions. If your shop is not among the sources the assistant uses, you are not part of that conversation. We describe the mechanism in more detail in the article on ranking in ChatGPT.
How many e-commerce companies use AI? There is no separate data for online retail alone in Poland. According to Eurostat, in 2025 8.4% of all companies in Poland with at least 10 employees used AI, against an EU average of 20%, while on the customer side it is the other way round: 53% of internet users surveyed by Gemius say they use chatbots or AI tools. Customers are moving faster than shops.
Sztuczna inteligencja w sklepie internetowym nie musi znaczyć dużego projektu. Hasło „sklep internetowy AI” w praktyce oznacza zwykły sklep, w którym kilka powtarzalnych zadań przejął program, a człowiek sprawdza wynik. Narzędzia z kategorii AI e-commerce różnią się głównie tym, gdzie pracują: w platformie sklepu, w obsłudze klienta albo poza sklepem, w odpowiedziach asystentów.
Where does AI really make money in an online shop?
Rozwiązania AI dla e-commerce oceniaj według dwóch pytań. Jak blisko są pieniędzy? Jak duże niosą ryzyko? Najczęstsze zastosowania zebraliśmy w tabeli. Rekomendacje i czat działają lepiej, gdy wiesz, do kogo mówisz, więc zanim je włączysz, przejrzyj swoje segmenty klientów w e-commerce i sprawdź, które grupy przynoszą największą część obrotu.
| Area | What AI does | Impact on sales | Risk |
|---|---|---|---|
| Product descriptions and specifications | writes, completes and standardises descriptions, translates into other languages | better visibility in search engines and AI assistants | incorrect specifications; requires checking |
| Customer chat (AI assistant in the shop) | answers questions about delivery, returns and product selection | fewer abandoned baskets, faster service | wrong promises for which the shop is liable |
| Shop search | understands questions asked as full sentences, not just keywords | the customer finds the product faster | low |
| Recommendations | suggests complementary products | higher basket value | low, if the data is good |
| Review and returns analysis | groups the reasons for dissatisfaction | better product range decisions | low |
| Photos and graphics | backgrounds, variants, campaign graphics | lower cost of materials | a photo that does not match the product is grounds for a return |
Shopping in ChatGPT: Allegro, AI agents and a shop ready for assistants
On 12 May 2026, Allegro launched an app in ChatGPT. The customer describes what they are looking for, the assistant browses the offers and shows product cards (price, photo, ratings), and a “Buy on Allegro” button takes them to the order on the platform. The search terms “ChatGPT Allegro” and “Allegro AI assistant” appeared in search engines precisely after this launch.
The takeaway for sellers is practical. In this model, the choice depends on how well the offer is described: attributes, photos, reviews, availability. The assistant will not read a “unique, top-quality product”, but it will read the size, material, weight and delivery time. The next step is shops with AI agents – programs that not only advise but place the order on the customer's behalf; we explain how such agents work in the article on agentic AI and AI agents.
How do you optimise a shop for AI? Five things make the biggest difference. Complete attributes in the product data. Structured data on product pages, i.e. a description in the page code that machines can read without guessing. An up-to-date product feed for price comparison sites and platforms. Genuine customer reviews. And finally, mentions of the shop in independent media and guides from which models draw their knowledge.
Who is AI in e-commerce for, and when does it pay off?
Shops with a large catalogue (hundreds or thousands of products) gain the most, because writing and correcting descriptions by hand simply does not scale. The second group is shops selling in several markets, where AI-assisted translation checked by a human is much cheaper than translating from scratch. We describe how to do this without slip-ups in the article on AI translation in business.
A separate case is B2B e-commerce with AI, i.e. ordering platforms for businesses. There, AI helps customers find the right product in an extensive technical catalogue, prepare an order based on previous ones and answer questions about specifications. The time saved for sales reps can be greater here than in retail, and if there is a sales team behind the B2B shop, connect it to the system we describe in the article on CRM with AI.
A small shop with a few dozen products and a small number of enquiries will gain the least. There, start with the tools built into the e-commerce platform rather than with separate implementations.
Jak Commwork®AI sprawdza, czy AI w sklepie naprawdę sprzedaje
W sklepie dane o sprzedaży leżą w kilku programach, a o tym, czy asystent AI poleci Twój produkt, decydują dane, których nikt nie mierzy. Commwork®AI łączy te źródła w jeden rachunek i sprawdza, gdzie sklep jest widoczny w odpowiedziach AI. Plan poprawek przygotowuje zespół Commplace®, a decyzję zawsze podejmuje człowiek.
- Wyszukiwarka Google 41%
- Reklamy płatne 27%
- Marketplace i porównywarki 18%
- Asystenci AI 6%
- Newsletter i wejścia bezpośrednie 8%
- 12.05.2026Allegro uruchamia aplikację w ChatGPT
- 2.08.2026Czat musi informować klienta, że rozmawia z programem
- Week 3Katalog uporządkowany, punkt wyjścia zmierzony
- Tydzień 5Nowe opisy bestsellerów w sklepie, kontrola parametrów
- Week 9Czat odpowiada o dostawie i zwrotach, z wyjściem do człowieka
- Week 12Porównanie z grupą kontrolną i decyzja o skali
- 01Przegląd katalogu
- 02Measurement in 8 AI engines
- 03Plan 90 dni
- 04Implementation
- 05Pomiar z grupą kontrolną
- Łączy platformę sklepu, analitykę, reklamy i system klientów w jeden rachunek od kliknięcia do zamówienia, a każdy wskaźnik ma źródło, datę, próg i właściciela.
- Zadaje pytania Twoich klientów w ośmiu silnikach AI i pokazuje, gdzie asystent poleca konkurencję zamiast Twojego sklepu. Jak wygląda taki pomiar, opisaliśmy przy AI visibility audits.
- Odejmuje sezon i sprawdza efekt grupą porównawczą, zanim wzrost po wdrożeniu czatu trafi do raportu jako sukces; o budżecie decyduje człowiek.
The view reproduced is modeled after the dashboards provided to customers. The names, numbers, and terms are examples and show the way the system works, not the results of a specific company.
How to prepare an online shop for AI implementation step by step
The question “how to use AI in e-commerce” usually ends with buying a tool, but it is better to start from the other end – the data the tool will work with, because that is what decides whether the implementation has any chance of success. Five steps for the first quarter:
Najczęściej zawodzi pomiar, bo dane o sprzedaży leżą w kilku programach naraz: w platformie sklepu, w analityce, w systemie reklamowym. Dlatego w Commwork®AI system łączymy je w jeden rachunek od kliknięcia do zamówienia i odejmujemy sezon, zanim ktoś uzna wzrost po wdrożeniu czatu za sukces.
- STEP 01Review your product data
How many products have complete attributes, consistent names, good photos and unique descriptions? Gaps here undermine everything downstream: the chat, the site search, recommendations, visibility in AI assistants.
- STEP 02Choose one area to start with
Most often product descriptions in one category, or a chat answering questions about delivery and returns. One area, a clear goal, a measurable result.
- STEP 03Measure your starting point
Conversion in the chosen category, the number of customer service queries, the basket abandonment rate, the time needed to prepare a description. Without this, you will have nothing to compare with after implementation.
- STEP 04Set control rules
Who checks descriptions before publication? What promises must the chat not make (discounts, delivery dates, exceptions to the terms and conditions)? When is the conversation handed over to a human?
- STEP 05Compare with a control group
Roll out changes in some categories and compare them with the rest or with the same weeks of the previous year. Seasonality and promotions can have a bigger effect than AI itself.
How much does AI in e-commerce cost and what drives the budget?
Some tools are already included in the price of the e-commerce platform, some are billed per use, but the biggest cost is usually the work on product data, which does not appear in any price list because it is done by your team or by someone you outsource it to. Prices below are as of September 2026.
Skąd wiesz, że wydatek się zwraca? Ustal próg przed startem, a nie po nim. Przykładowy próg: czat ma zamykać bez udziału człowieka co najmniej co trzecią sprawę o dostawę i zwroty, a zwroty z powodu niezgodności opisu nie mogą rosnąć w grupie produktów z nowymi opisami. Gdy jeden z tych warunków nie jest spełniony, najpierw poprawiasz bazę wiedzy i kontrolę treści, dopiero potem zwiększasz zakres.
Zanim podpiszesz umowę z dostawcą czatu albo narzędzia do opisów, zadaj mu kilka prostych pytań. Odpowiedzi szybko pokazują, czy rozmawiasz o sprzedaży, czy o samej technologii.
Jeśli chcesz zacząć od sprawdzenia, czy asystenci AI w ogóle polecają Twój sklep, zamów najpierw pomiar w ramach usługi pozycjonowanie w AI dla sklepów i marek, a dopiero potem wybieraj narzędzia.
- Za co dokładnie płacę: za rozmowę, za rozwiązaną sprawę czy za użytkownika?
- Skąd czat bierze odpowiedzi i kto aktualizuje bazę wiedzy po zmianie regulaminu?
- Jak klient przechodzi do człowieka i ile to trwa w godzinach pracy sklepu?
- Czy mogę pobrać zapis rozmów i sprawdzić, czego klienci szukali bez skutku?
- Gdzie są przechowywane dane klientów i czy podpiszesz umowę powierzenia?
- Jak zmierzymy efekt, żeby nie pomylić go z sezonem?
| Item | What it depends on | Example or note (as of 09.2026) |
|---|---|---|
| AI tools built into the shop platform | platform, plan | Shopify Magic: free on all Shopify plans (product descriptions, customer replies) |
| AI chat for customers | number of conversations resolved without a human | Intercom Fin: from USD 0.99 per resolved case |
| General tool for the team (descriptions, translations, analyses) | number of people | ChatGPT Business: USD 20 per person per month, billed annually |
| Organising product data | number of products, state of specifications and photos | usually the biggest work item |
| Integrations (shop, warehouse, customer service system, marketplace) | number of systems | individual quote |
| Content quality control | number of descriptions published | fixed team time |
Can I use AI in my shop and what does the law say?
Yes, you can use AI in e-commerce, but with a few obligations. Legal position as of 24 September 2026:
- Customer chat. From 2 August 2026, Article 50 of the AI Act requires that customers know they are talking to a program, unless this is obvious.
- AI-generated content. The AI Act provides for labelling certain synthetic content, especially realistic images and recordings. We explain what this means for marketing in our piece on labelling AI content from 2 August 2026.
- Description matching the product. The seller is responsible for the description and photo, regardless of who prepared them. A generated attribute the product does not have is a straight road to a complaint.
- Chat replies. The shop is responsible for what its chatbot has promised. We describe how to set the boundaries of such a conversation in the article on AI in customer service and chatbots.
- Personal data. Chat conversations and data for recommendations are customer data, so you need a data processing agreement with the vendor and information in your privacy policy.
Typowy sklep w 90 dni, czyli jak może wyglądać pierwszy kwartał
Poniżej przykład poglądowy, a nie historia konkretnego klienta. Załóżmy sklep z około 2 000 produktów, trzema osobami od treści i obsługi oraz jednym szczytem sprzedaży w roku. Taki sklep nie potrzebuje wielkiego wdrożenia. Potrzebuje kolejności.
Miesiąc pierwszy. Zespół przegląda katalog i zapisuje, których parametrów brakuje w najlepiej sprzedających się produktach. Równolegle mierzy punkt wyjścia: konwersję kart produktów, pytania do obsługi w tygodniu, zwroty z powodu niezgodności opisu i to, czy sklep pojawia się w odpowiedziach asystentów na dwanaście typowych pytań klientów. Tak wygląda początek każdego rzetelnego AI visibility audit.
Miesiąc drugi. AI pisze nowe opisy dla 300 bestsellerów, a człowiek sprawdza każdy parametr z kartą producenta. Na kartach pojawiają się dane strukturalne. Czat dostaje wąski zakres: dostawa, zwroty, status zamówienia. Wszystko, co wykracza poza ten zakres, trafia do człowieka.
Miesiąc trzeci. Zespół porównuje produkty z nowymi opisami z podobną grupą, której nie ruszano, i czyta zapisy rozmów czatu. Dopiero wtedy zapada decyzja: rozszerzyć, poprawić czy zatrzymać. Opinie klientów zbierane po zakupie pomagają ocenić, czy czat pomaga, czy irytuje; prostą metodę opisuje tekst o ankiecie zadowolenia klienta.
Ile to kosztuje w narzędziach? Przy cenach z tabeli wyżej: trzy licencje ChatGPT Business to 60 USD miesięcznie przy płatności rocznej, a czat rozliczany za rozwiązaną sprawę przy 500 takich sprawach w miesiącu i najniższej stawce z tabeli kosztuje około 495 USD. Najwięcej i tak pochłonie czas zespołu na sprawdzanie treści. Obie kwoty to rachunek na przykładzie, nie wycena.
- Czy znam 300 produktów, które przynoszą największą część obrotu?
- Czy każdy z nich ma komplet parametrów, zdjęć i aktualną dostępność?
- Czy zmierzyłem punkt wyjścia, zanim cokolwiek zmieniłem?
- Czy czat ma zapisany zakres i drogę do człowieka?
- Czy wiem, kto sprawdza opisy przed publikacją i ile ma na to czasu?
- Czy mam grupę produktów, której nie zmieniam, żeby porównać wynik?
- Jeśli sprzedajesz na Allegro: czy wiem, jak moje oferty wypadają, gdy klient szuka przez Allegro ChatGPT?
When should you implement AI in your shop, and when is it better to wait?
The best moment is two or three months before the peak season, which for most shops means before the pre-Christmas period. There is then time for testing and fixes, followed by calm measurement, so the effect arrives just when traffic is at its highest. Rolling out a chat in the week of a big promotion is asking for trouble.
It is better to wait while the e-commerce platform is being migrated, when product data is in disarray or when nobody has time to review content. In that case, tidy up the catalogue first. Also look at the whole customer journey rather than just one element; for help with this, see our earlier article on how to design the purchasing process. Miejsca, w których klienci rezygnują z koszyka, pokaże dobrze zbudowany conversion funnel; AI warto najpierw kierować właśnie tam.
The future of AI in e-commerce and what will change for shops
How will AI affect the future of e-commerce? Two directions are certain because they are already visible. First, some customers start shopping not in a search engine or on a marketplace but in a conversation with an AI assistant – the Polish example being the Allegro app in ChatGPT. Second, AI agents are emerging that compare offers and place orders themselves. In both cases, the winner is the shop whose data is complete and trustworthy.
Sklep z agentami AI to na razie raczej kierunek niż codzienność, ale przygotowanie do niego jest to samo co do zwykłej sprzedaży: aktualne ceny, stany i warunki dostawy w danych, które program odczyta bez zgadywania. Agent, który złoży zamówienie na podstawie błędnego parametru, wygeneruje zwrot tak samo jak klient. Skąd biorą się pewne siebie, ale fałszywe odpowiedzi modeli, wyjaśniamy w tekście o AI hallucinations.
The rest is, for now, forecasts. A sensible strategy is not about rebuilding the shop around one vision of the future, but about doing things that help regardless of how fast things change: tidy data, good descriptions, genuine reviews, independent mentions. Can AI build an e-commerce shop? It will help with the template, descriptions and graphics, but decisions on the product range and prices are made by people, as are decisions on how to serve customers.
Want to check how your shop performs in AI answers and what to do to make it appear more often? Visit the page AI search optimisation. We write about how ChatGPT is also becoming an advertising channel in the article on ads in ChatGPT.
Mistakes that cost a shop customers and margin
A generated attribute the product does not have ends in a return, and often also in a complaint and a bad review. The cost of a single incorrect description for a popular product can exceed the annual cost of quality control.
A customer with an order problem, stuck in a conversation with a bot, abandons the basket or complains publicly, where future customers will read it. A “talk to an agent” button is cheaper than a lost customer.
Descriptions copied from manufacturer materials or generated from the same prompt do not make the shop stand out, either in search engines or in AI assistants. Add information nobody else has: your own tests, photos, answers to customer questions.
A rise in sales in December after implementing AI in November is correlation, not proof. Without comparison with a control group or with the previous season, it is easy to overpay for a tool that achieved little.
A shop that does not look after its product data and independent mentions does not appear in ChatGPT's answers. Competitors who do quietly take over those customers.
Shop owners' questions about artificial intelligence
What is AI in e-commerce?
It is the use of artificial intelligence in an online shop: for writing and translating descriptions, customer service via chat, site search, recommendations, review analysis and also for making the offer appear in the answers of AI assistants such as ChatGPT.
Can I use AI in e-commerce?
Yes. Remember, however, that from 2 August 2026 the chat must tell customers that it is a program (Article 50 of the AI Act), and the shop is responsible for the content of descriptions and chat replies just as if a person had written them.
How does Allegro work in ChatGPT?
Since 12 May 2026, Allegro has been available as an app in ChatGPT. The customer describes in the conversation what they are looking for, the assistant shows offers (price, photo, ratings), and the purchase takes place on Allegro after clicking a button that leads to the order.
How do you optimise a shop for AI?
Complete your product attributes, add structured data to product pages, keep your product feed up to date, collect genuine reviews and seek mentions in independent media. AI assistants build their recommendations from exactly these sources.
How much does AI in an online shop cost?
It depends on the scope. Some tools are free on the platforms (for example Shopify Magic on all Shopify plans), and AI chat is often billed per resolved case (for example Intercom Fin from USD 0.99, as of September 2026). The most expensive part is the work on product data.
Can AI create an e-commerce shop?
AI will help prepare the template, descriptions, graphics and copy. It will not, however, replace decisions on product range, pricing, logistics or customer service, and a shop built entirely by AI usually has nothing that sets it apart.
Where should you start with AI in e-commerce?
Start by reviewing your product data and choosing one area, for example descriptions in one category or a chat about delivery and returns. Then take a measurement before implementation, run a pilot, and finally compare with a control group or with the same period of the previous year.
Sources
- SOURCEAllegro: Allegro uruchamia apkę na ChatGPT · 12 May 2026
- SOURCEGemius: report “E-commerce w Polsce 2025” · 24 September 2025
- SOURCEEurostat: 20% of EU enterprises use AI technologies · 11 December 2025
- SOURCEShopify Help Center: Shopify Magic · as of 24 September 2026
- SOURCEIntercom: pricing (including Fin AI Agent) · as of 24 September 2026
- SOURCEOpenAI Help Center: ChatGPT Business · as of 24 September 2026
- SOURCEEuropean Commission: transparency obligations under Article 50 of the AI Act · as of 24 September 2026
Read next
Sprawdzamy w ośmiu silnikach AI, czy asystenci polecają Twój sklep, i pokazujemy, czego brakuje w danych produktowych, treściach i wzmiankach. Pierwszy krok to pomiar na pytaniach Twoich klientów, potem plan na 90 dni.
Sebastian Kopiej, President of Commplace®. In public relations since 1996. Text prepared with the help of AI tools and fact-checked by the author. Data current as of September 29, 2026.