AI at PubLocal
PubLocal uses artificial intelligence (AI) for two things only: helping you write a listing and protecting you from scams in the messages. Here is exactly what it does, and what it doesn’t.
Filling in a listing with AI
- On the Post page, the “Fill in with AI” button suggests a title, a description, a category and a price range based on your photos.
- Nothing is sent until you click. Only small versions of your photos (3 at most) are sent, without your name or your email.
- The AI posts nothing: you read it over, correct it and post it yourself, and the listing goes to review as usual. The posted text remains yours: check that it is accurate. The suggested price is a general estimate, not an appraisal.
- Suggestions per month: 1 with the Free plan, 5 with Pro, 10 with Seller and 20 with Business. An attempt that gives nothing is not counted.
The anti-scam shield
- Some messages you receive are read automatically by the AI to spot known scams: a request for a code received by text, payment in advance, a fake payment link, a fake courier, etc.
- What is analyzed: the first messages of each conversation, messages from accounts less than 30 days old and those that mention payment, links or contact details. The AI receives the text of the message and of the few messages before it, the listing’s title and price and the age of the sender’s account, without any name or email.
- If a message looks like a scam, a warning appears under it, only for the person who receives it. When the risk is high, our team is also notified and can read the conversation.
- The AI never blocks, deletes or changes a message, and makes no decision about an account: only a person on our team can suspend an account, after reviewing the conversation.
- It is a help, not a guarantee: the AI does not catch every scam, and the absence of a warning does not mean a message is risk-free. Keep the good habits.
The shield on listings
- Every new listing, and every change to its text, price or photos, is read by the AI to spot known scams and prohibited items: a deposit required before a visit, a fake job, an unrealistic price, counterfeits, weapons, etc.
- The AI receives the listing’s title, description, price, category, town, website and first photo, along with the plan and the age of the seller’s account, without any name or email.
- Its opinion is read only by our team: it is never shown to buyers. A listing that is already waiting for the team’s review stays pending as usual.
- A listing already online that the AI judges very risky is taken off the site and sent back to review while a person on our team examines it. It comes back online if the team approves it. The AI never rejects or deletes a listing: only the team can.
Who runs the AI
- The AI is provided by Anthropic (Claude model), a company based in the United States, where the data is processed.
- Anthropic does not use the photos or messages we send it to train its models.
- We use AI for nothing else: no profiling, no targeted advertising, no final automated decision about your account or your listings.
AI can make mistakes
A strange suggestion, an unjustified warning, a shocking text produced by the AI? Let us know: we fix the AI’s instructions when it gets things wrong.