Sales rules and guardrails
Commerce Pro gives you two kinds of control over what the assistant says. Sales rules are short notes you write, such as “Sizes run small”, that shoppers can open on a product. AI guardrails shape the answers the AI writes: their tone, their language, and which claims to avoid.
Write sales rules
Section titled “Write sales rules”- Open Geeky Bot → Commerce Pro.
- Select the Sales assistance tab.
- Turn on Sales rules.
- In Rules, write one rule per line as
target | note. - Select Save settings.
The target says which products the note applies to:
| Target | Applies to | Example |
|---|---|---|
global |
Every product | global | Free shipping over $50 |
product:ID |
One product, by its WooCommerce ID | product:123 | Engraving takes two extra days |
category:slug |
Products in a category, by its slug | category:shoes | Mention that sizes run small |
tag:slug |
Products with a tag, by its slug | tag:organic | Certified organic cotton |
keyword:word |
Any product, when the shopper’s latest message in the chat contains the word or phrase. Letters, numbers, spaces and hyphens, 2 to 60 characters. Since Commerce Pro 2.1.3 | keyword:gift | Free gift wrapping on every order |
Lines that don’t follow the format, or use an unknown target, are dropped when you save. You can keep up to 50 rules, and each note is cut at 240 characters.
How shoppers see sales rules
Section titled “How shoppers see sales rules”Each product card in the chat gets a Sales note button in its More actions menu (•••). When a shopper selects it, the chat shows “Additional store information for this product:” with up to three notes that match the product, in the order you wrote them. If none match, the chat says “No additional sales information is available for this product.”
Sales rules are shown as you wrote them. They aren’t sent to an AI provider and don’t change the AI’s answers. To give the AI facts it may use, use Approved guidance below.
Set AI guardrails
Section titled “Set AI guardrails”The guardrails only affect answers written by an AI provider, so they need OpenAI or Zywrap as your answer mode. In local mode Geeky Bot never writes free text, and the screen marks Multilingual replies and Extra guardrails as On, inactive. See Answer modes.
- Open Geeky Bot → Commerce Pro.
- Select the AI guardrails tab.
- Change the settings below.
- Select Save settings.
| Setting | What it does |
|---|---|
| Multilingual replies | Tells the AI which language to reply in. Product names, SKUs and coupon codes stay unchanged. |
| Language | Match the shopper (the default), Use the site language, or Force one language. Applies only while Multilingual replies is on. |
| Forced language | The language to use with Force one language, for example English, Arabic or Spanish. Only a language name is accepted, up to 40 characters. |
| Tone | Helpful (the default), Friendly, Premium retail or Concise. Changes the style of AI answers, never the product facts. |
| Extra guardrails | Adds stricter rules against unsupported claims, invented offers and guarantees. When the shopper asks for something the store hasn’t provided, the AI says so and suggests the product page or contacting the store. |
| Topics to avoid claims on | One topic per line. The AI is told to be extra careful with these. Up to 10 lines of 80 characters. Applies only while Extra guardrails is on. |
| Approved guidance | Optional facts the AI may use, such as size advice. Up to 8 lines of 140 characters. Applies only while Extra guardrails is on. |
Topics to avoid claims on starts with medical, legal, financial, safety and guaranteed delivery. Add topics that matter for your products, such as allergens or warranties.
Geeky Bot’s own rules already tell the AI to use only your catalog and Store Knowledge pages, and never to invent prices, stock, coupons, delivery promises or refund rules. The guardrails add to these rules and can’t remove them.
How the settings reach the AI
Section titled “How the settings reach the AI”When the assistant asks your AI provider to write an answer, Commerce Pro adds your language, tone and guardrail settings to the instructions sent with that request. With OpenAI they go to OpenAI, and with Zywrap to your Zywrap endpoint. Nothing is sent in local mode.
Before they are sent, your topics, approved guidance and forced language are filtered so that text that looks like an instruction to the AI is removed. Write plain facts and topics, not commands.