How to generate typical questions and answers per product via AI and show them as a chat or FAQ on the product detail page.
Shopware 6.7, an API key with one of the supported AI providers and a valid licence token. The plugin is part of the Stone & Water plugin suite.
The direct edition installs through an upload in the admin, no SSH access needed. After activation the menu entry appears under Extensions, Stone & Water - Plugin Suite.
Until activation the admin shows a licence screen with an input field. The check is bound to the shop domain and works fail-closed: without a valid licence the widget is not rendered in the storefront. If the licence server is unreachable, a three day tolerance applies.
In the admin under Extensions, Stone & Water - Plugin Suite, AI Product Questions. That is where provider settings, bulk generation and single generation live.
Four providers are supported: OpenAI, Anthropic, Mistral and Google. For each one you store an API key and pick a model.
The model list is maintained over time. Older models stay at the end of the list marked as older so existing configurations keep working after an update. Mistral entries point at -latest aliases and therefore always resolve to the newest version.
The prompt instruction permanently shapes content and tone of the generated texts - concise and technical, focused on use cases, or in your shop's own voice.
Content is produced per shop language. For the run, product data is loaded in the target language, the AI writes in that language, and the entries are stored under the matching language. The storefront then automatically serves the texts of the active shop language.
Content language is a multi-select with chips. The run is executed once per selected language. Without a selection the system language applies.
Text language defines where the writing language comes from: automatically from the locale of the shop language, or fixed. The fixed choice is useful when a shop language should deliberately be filled with texts in another language.
Sales channels are a multi-select with chips too, directly above the content language. The run is executed once per combination of channel and language, using the settings of each channel. Without a selection the general configuration applies.
With a channel selected, the run only covers products assigned to that channel - the article count in the overview follows that selection as well. An explicit product selection still takes precedence.
The run opens a window that states the planned scope up front: all selected channels and languages. During the run it shows the current pass as "x of y" with the current channel and language, plus a bar and counters for products, processed and skipped articles.
Error messages stay in plain words until you close the window. Instead of "no content generated", the message names the actual cause: error text from the AI service, HTTP status, truncation by length or safety filter, refusal, connection error, or missing product master data in the chosen language.
The bulk run generates questions, answers and pros and cons for many products in one go. Processing happens in batches whose size you control through the parallel processing setting. At the end you get a summary of generated, skipped and failed products.
For targeted fixes there is a separate area with product search by name or article number. Several products can be picked one after another; they appear as chips below the search field and can be removed with an X. Products already picked are hidden from the search results.
Optionally you can add an instruction that applies to this run only - a content focus or a different tone, for example. It is passed in addition to the configured prompt instruction.
Targeted generation lives in its own "Individual products" tab between Generation and Settings; the Generation tab now contains only the bulk run.
The search finds products by name and by article number, variants included. A chosen variant is resolved to its parent product and that number is shown; if a variant has no name of its own, the article number serves as the name.
The product detail page shows a chat area with a headline, an input field and a send button. The bundled default headline is „Ask me anything about this product", with a note below explaining that the digital advisor, like a human, can occasionally get things wrong.
If the AI has no basis for an answer, it says so clearly instead of guessing. On technical problems an error message asks the customer to try again later.
Above the chat and FAQ you can show a row of compact chips with the key details of the article: product name, article number, manufacturer, availability and price. Each entry can be switched on individually, per sales channel.
If you select none of them, the area does not appear at all - the chips are entirely optional.
When a customer picks a variant on the product page - colour or size, for example - the chosen options are passed to the AI as context. Answers then refer to the version actually being viewed rather than to the product in general.
If a variant has no description of its own, the plugin falls back to the one from the main product. Answers stay useful even when variants are only thinly maintained.
Alongside questions and answers the plugin generates a list of pros and cons per product, rendered as its own block and marked as coming from the AI product advisor.
Four settings affect visibility in search engines. All are off by default, so existing installations do not change by themselves.
Own URL per question: every question gets a shareable address and opens automatically when called. Optionally a chain icon appears next to it for copying directly.FAQ markup: the questions are marked up as an FAQPage so search engines can present them in the results. The addresses of individual questions are included where enabled.SEO question style: questions are generated as fully worded search queries, with answers of 40 to 60 words. Takes effect from the next generation.Internal linking: products mentioned in the answers are linked to their product page, at most three per answer. Linked are cross-selling partners and articles from the same manufacturer.With this setting all questions for an article also appear as a standalone, indexable page. It carries the questions and answers, the pros and cons, its own page title with description and a link back to the product. The widget links to it below the footnote.
"URL structure of the FAQ subpage" defines the address: product ID, article name, article number, or article name and number together. Umlauts and special characters are transliterated into URL-safe characters.
The address of the subpage uses the full article name, truncated at a word boundary at 255 characters - exactly as Shopware handles SEO addresses. Earlier truncations at 80 characters remain resolvable through fallbacks, so links already shared keep working.
In bulk generation the "Scope" selection decides what is regenerated. With "Pros and cons only" the existing questions stay untouched - useful when the questions are already indexed and only the assessment needs reworking.
For individual products there is a dedicated button for this. Both save cost with the AI provider because only the desired part is regenerated.
The channel selection is also available in the Generation tab. Bulk and single generation then use the settings of the chosen channel: AI provider, language, number of questions, question style and the selected product details.
Administrative access to the plugin is bound to dedicated permissions, separated into viewing and editing. Anyone with read-only rights in the Shopware admin can therefore see statistics but not generate or delete.
Free-text input in the shop - the option for visitors to ask their own question - is explicitly switched off until you enable it. When active, requests are rate limited per sender: at most eight per minute, after which the shop rejects further requests.
The texts live in a dedicated table called staw_product_question_direct. The direct and store editions deliberately use separate tables so they can run side by side. When updating from an older version an existing old table is renamed automatically and the data is preserved.
Without a valid licence it is deliberately not rendered. Also check whether texts have been generated for that product at all.
Usually a valid API key is missing or the chosen model is not enabled with the provider. The summary at the end of the run states the number of errors.
That is expected with „new products only" enabled, or when texts already exist and „overwrite" is off.
The check uses the host of the request. Opening the admin through a domain other than the licensed one makes it fail.
We are happy to help with provider choice, prompt instructions or licensing. Get in touch via the contact page or write to info@stoneandwater.online.
Every setting of the plugin with its default value and explanation, in the same order as in the administration. The texts come straight from the plugin (version 1.1.1).
AI provider used to generate the questions.
Your API key for the selected provider. Stored in the Shopware configuration.
Model of the selected provider.
How many questions (incl. answer) are generated per product. Allowed range: 1 to 12.
How many products are generated simultaneously. Higher = faster, but more load on the AI provider (mind rate limits). Recommended: 5. Allowed range: 1 to 10.
Language in which questions and answers are generated.
Additional instruction for the AI, e.g. tonality or content focus.
Existing product reviews are given to the AI, e.g. for a “What do customers say?” question.
Cross-selling or similar products (same manufacturer) are given as context to enable comparison questions.
Products of these manufacturers are skipped in bulk runs (all products, category, period, stream). An explicit selection - specific products or a deliberately chosen manufacturer - still takes precedence.
Also creates a short pros/cons list and shows it as a block on the product page
Headline above the question chat on the product page.
Where the question chat is inserted on the product page.
Custom logo/icon for the chat header. Empty = Stone & Water logo.
Colour for buttons, chips and accents in the chat widget.
Maximum number of pre-built questions shown in the shop. Allowed range: 1 to 12.
The selection applies to all questions: the chosen product details are used as key facts during AI generation and woven directly into the texts of the questions, answers and pros/cons. No selection means no requirement. Takes effect from the next generation.
Lets customers type their own questions - the answer is generated live by AI.
When enabled, the question chat is completely hidden on product pages without generated questions - even if free-text input is enabled.
Shows the generated questions as an expandable FAQ list (click a question → answer) and hides the input field. Free-text input is ignored in this mode.
HTML tag for the block heading and (in FAQ mode) the questions, e.g. h2-h6 for a clean HTML structure. The look stays the same.
Small note below the chat (e.g. AI disclaimer).
Shows or hides the pros/cons block on the product page (regardless of whether it was generated).
Hides the cons and shows only the pros in the block. Cons are still generated but not displayed.
Appends the product name to the heading of the block, e.g. „Highlights: Product name“. Applies to the detail page and the FAQ subpage.
Custom heading for the pros/cons block. Empty = default text. Handy when only pros are shown (e.g. What we like).
Outputs questions/answers and pros/cons directly as HTML in the source (as an FAQ) instead of loading them via JavaScript - better for search engines. If free-text input is enabled, the input field also appears below the FAQ.
Every question gets its own directly shareable URL and opens automatically when visited. Applies to the FAQ display and the SEO mode.
Shows a small chain icon next to every question that opens and shares the question's own URL directly.
Additionally outputs the questions and answers as FAQPage markup (structured data). Helps search engines understand the page and is a prerequisite for rich results. Only takes effect in SEO mode (server render).
Phrases generated questions like typical Google search queries and targets 40-60 words per answer (featured snippet format). Takes effect from the next generation.
Automatically links products mentioned in the answers to their product pages (links per answer configurable). Only takes effect in SEO mode (server render).
Automatic: cross-selling products, supplemented by same-manufacturer products when fewer than 3. Cross-selling only: links and mentions during generation exclusively the cross-selling products maintained on the product - if no cross-selling is maintained, no links are set.
How many product links are set in a single answer at most (1-10, default 3).
Comma-separated terms. Products whose name contains one of these terms are never linked and no longer provided to the AI as similar products. Linking takes effect immediately after cache clear; existing mentions in the text only disappear after a regeneration.
Provides all generated questions of a product as an own indexable page at /produkt-fragen/... (incl. FAQ schema) and links it below the widget. Enables standalone rankings for long-tail search queries.
Defines what the URL is built from. Already shared links keep working after a change because all variants are resolved.
Generates only for active (visible) products, so the whole catalogue including disabled products is not filled with texts.
What happens to products that already have questions/texts.
Two routes, the same feature set - pick whichever suits your setup.
All prices net, plus VAT.
You buy the licence directly from us and get an invoice from us. Install by upload in the admin without SSH, updates through our licensing platform, support straight from the developer.
Listed in the official Shopware Community Store. Installation and updates run through the plugin manager, billing through your Shopware account. Store conditions differ from ours.
All features on the product page. We are happy to help with setup and integration.