Product FAQ AI — Complete installation & configuration guide
Install and configure Product FAQ AI 4.0.0: connecting Claude or OpenAI, writing settings, reviewing drafts, bulk generation with a budget, accessible display and structured data.
Product FAQ AI uses AI (Anthropic Claude or OpenAI) to write the FAQ for your PrestaShop product pages using only the data in your catalogue. The answers are saved as drafts: you read them, correct them if needed, then validate them before they appear on the product page, in an accordion that is keyboard-accessible and works without JavaScript.
Version 4.0.0. Rewritten module: PrestaShop 1.7.8, 8 and 9, PHP 7.1 to 8.5, back office in 8 languages, drafts with review, bulk generation with cost estimate and budget, usage log, 5 accordion styles. Upgrading from 3.x keeps your FAQs, their order, their visibility and your settings.
Requirements
- PrestaShop 1.7.8, 8.x or 9.x, with the Classic or Hummingbird theme, or a theme that calls one of the product page hooks offered.
- PHP 7.1 to 8.5 with the cURL and OpenSSL extensions (API key encryption).
- An Anthropic (console.anthropic.com) or OpenAI (platform.openai.com) account with an API key. Calls are billed by the provider to your own account.
- The server must be able to make outgoing HTTPS requests to
api.anthropic.comorapi.openai.com.
Installation and upgrade
- In Modules > Module Manager, click Upload a module and choose
productfaqai-4.0.0.zip. - The module creates the
ps_product_faqtable and three technical tables (ps_productfaqai_job,ps_productfaqai_job_item,ps_productfaqai_usage), and registers its hooks and a hidden admin controller. - Click Configure: the AI connection tab guides you through setting up the API key.
Upgrading from 3.x. Upload the 4.0.0 zip over the old version. Nothing is deleted: an FAQ that was active in 3.x becomes “published”, and an inactive FAQ becomes a draft. An API key stored in plain text is encrypted, model identifiers are replaced with current models (for example claude-haiku-4-5-20251001 with claude-haiku-5-5, gpt-4o-mini with gpt-6-luna), and the displayProductFooter position, which neither Classic nor Hummingbird calls, switches to displayFooterProduct if your theme does not use it.
Dashboard
The dashboard shows the share of active products that have a published FAQ, the number of product pages with drafts to review, the month’s spending against your cap, and the indicative cost of 1,000 FAQs with the selected model. Coverage is broken down by language: published FAQ, drafts only, or no FAQ.

Connecting to the AI provider
The wizard takes three steps: choose the provider, paste the key, choose the model. The key is first tested with a free call to the provider’s Models API; it is saved only if the test succeeds.

The key is stored encrypted (AES-256-GCM) in the database. It is never displayed again (only the last 4 characters are shown), it is never sent to the browser and it is never written to the logs. To replace it, paste a new key.
Choosing the model
The list shows the models available on your account, read from the Models API during the test. Recommended models display their indicative list price and the estimated cost for 1,000 FAQs with your settings. Without a test, a fallback list is offered.
- Anthropic:
claude-opus-5-5(default, best quality),claude-sonnet-5-5(balanced),claude-haiku-5-5(fast and economical). - OpenAI:
gpt-6.1-sol(default),gpt-6-astra,gpt-6.1-terra,gpt-6-luna,gpt-5.4-mini,gpt-5.4-nano.

Reasoning effort is set to “low” by default: writing an FAQ from a product page is a simple task, and a higher effort increases the time taken and the number of tokens.
Writing settings
Choose the number of questions per product (1 to 15), the length of the answers (short, medium, detailed) and the tone (professional, friendly, conversational, technical, enthusiastic or reassuring). You can specify a target audience and style instructions, for example “address the customer formally”.

Style instructions cannot override the module’s rules: use only the facts on the product page, never invent a price, delivery time, warranty, certification or health claim, and write in plain text. For commercial information (warranty, returns), write it in the product description or add the question by hand.
Data sent: the name, the short and long descriptions, the dimensions and the weight are always sent; the main category, brand, features and combinations are sent depending on your settings. No customer, order or employee data is ever sent.
New products: the “Put new products in the FAQ queue” option does not call the AI while the product is being saved; the dashboard shows the pending products and you start the queue whenever you want. Publishing without review is a separate option, off by default.
Products and FAQ editor
The Products & FAQ tab lists the shop’s products with, for each language, the number of published or draft questions and any warnings. Filter by products without FAQ, with drafts or with a published FAQ.

The editor opens a product’s FAQ in one language:
- Generate the FAQ: replaces the drafts for that language; questions already published are kept unless you ask to replace them.
- Regenerate a single question: the new version, different from the other questions, is saved as a draft.
- Reorder by drag and drop or with the arrows, edit, delete, or add a question by hand.
- Validate a question, or Validate and publish all. Only published entries appear on the product page.

Verification warnings
Each answer shows the facts from the product page that the AI relied on. The module displays a warning when a figure does not appear in the product data, when the answer mentions warranty, delivery, returns, certification or health while the product page says nothing about it, or when no fact is cited. An entry with a warning is never published automatically.

Data sent to the AI
The Data sent to the AI button shows exactly what the module sends for this product: the instructions (identical for every product, which allows them to be cached) and the product message, with an estimate of the tokens and cost.

Bulk generation
Choose the languages, the products (only those without FAQ, or all active products), optionally a list of IDs, and what happens after generation (drafts to review, recommended, or publication of entries without warnings). Click Estimate the cost: before anything is sent, the module displays the number of product pages × languages, the estimated tokens and a cost range, with and without caching of the instructions.

- The queue progresses in short steps as long as the module page is open: no never-ending request, no cron job.
- A lock prevents two tabs or two employees from processing the same queue; only one queue is active per shop.
- The queue stops at the run’s budget and at the monthly cap. You can pause it, resume it, cancel it, or retry the errors (3 attempts per product page).
- If the provider refuses every request (key, credit, model), the queue pauses with a message telling you what to fix.
Economy mode: requests are sent to Anthropic’s Message Batches API or OpenAI’s Batch API, at about half price. Results usually arrive within a few hours, at most within 24 hours, and are collected when the module page is open.

Costs, budget and log
Set the default budget for a bulk run, the monthly cap (all generations combined) and the maximum number of tokens per answer. The log lists every call: operation, product, language, model, input, output and cache-read tokens as returned by the provider, indicative cost and any error.

The prices used are indicative list prices recorded on 11 November 2026. The estimate counts about 3.5 characters per token: it is an order of magnitude. Your provider’s invoice is the reference.
Display on the product page
In the Display tab, choose the style, the accent colour, whether the first question is open, the position on the page and the block title in each language. A position that your theme does not call is flagged.

Five styles are available: minimal, cards, framed, plus buttons and all open. The block uses the native HTML elements <details> and <summary>: it works with the keyboard (Tab, Enter, Space) and with screen readers, without any JavaScript file. The HTML is generated by the server, so nothing shifts when the page loads, and the stylesheet is loaded only on product pages that display an FAQ.








FAQPage structured data
The module can add schema.org FAQPage markup (JSON-LD) to product pages that display an FAQ. The option is off by default.
Google restricted FAQ rich results to authoritative government and health websites in August 2023, then stopped showing them for all sites on 7 May 2026. The markup is still valid schema.org and does no harm, but it produces no rich results in Google: do not expect any SEO gain from it.
If you enable the option and the DataFirefly All in One SEO module already outputs an FAQPage block on the page, Product FAQ AI does not add a second one; the other module’s settings are never changed. The JSON-LD is encoded so that it stays valid and can never close the script tag.
Security and privacy
- The product text is placed in a delimited data block and treated as untrusted content: a description containing instructions (“ignore the instructions…”) remains plain data.
- The AI’s answer is constrained by a JSON schema, then cleaned: plain text only, with no HTML, links, email addresses or Markdown.
- Back-office actions go through a token-protected controller; write operations require a POST request from the module and the “edit” permission on the Product FAQ AI tab. An employee who only has the “view” permission can consult but cannot generate or edit.
- Multistore: settings, key, FAQs, queue and log are specific to each shop; select a single shop to configure the module.
Troubleshooting
- “The provider refused the key (401)”: check the key in the provider’s console or create a new one, then paste it again.
- “Unknown or unavailable model (404)”: choose another model; the Test the saved key button reloads the list for your account.
- No credit left or billing limit reached: add credit with the provider, then resume the queue.
- Too many requests (429), service overloaded: the module has already retried, respecting the delay indicated by the provider; wait, or use economy mode.
- Answer cut off: increase the maximum number of tokens per answer, or ask for fewer questions or shorter answers.
- Refusal by the provider: review the product text or write the questions by hand.
- The FAQ is not displayed: check that the questions are published in the language being viewed and that the chosen position is called by your theme (warning in the Display tab).
Uninstallation
Uninstalling deletes all the FAQs, the generation queue, the usage log, the settings and the admin controller. You are asked for confirmation.