AI will suggest. Your team will decide.
The AI package arrives with v2. It will flag suspicious reviews, draft seller replies, work out what reviews talk about, and translate and summarise them.
It will flag suspicious reviews, never reject them.
It will spot reviews that look fake, incentivised, off-topic or abusive and say why they look suspicious.
- The review will go to the approval queue and never be rejected automatically
- The reason it suggests will sit next to the review
- Someone on your team will make the decision
It will write the first version of your seller reply.
It will prepare a draft in the tone you pick in the reply form, written straight into the reply field. You will still edit it and send it.
- The tone will be picked for each draft: formal or warm
- It will never ask the shopper to change or withdraw their review, and never offer a discount
- The reply will still be approved before it goes live
What do reviews talk about most?
It will work out which topics each review mentions and whether it speaks about them positively or negatively. The results will appear by topic on the Statistics screen.
- Ready-made topics per category, such as fit, quality, delivery, price and packaging
- Anything that does not fit will go under “other”
- You will be able to compare one period with another
Shoppers will read reviews in their own language.
When a review goes live, it will be translated into the languages of your open channels, so shoppers in another country can read it in their own language.
- Translated reviews will carry a “Translated” label
- “Show original” will keep the original one click away
- The original text and the rating will never change
A product's reviews in a few sentences.
The product page will show a summary of the reviews. It will say it was generated with AI and how many reviews it is based on.
- Built only from published reviews, with negatives in proportion
- It will appear on products with at least 10 reviews
- You will be able to turn it off, but not edit it
- Every version will be recorded, so you can prove in an audit what was shown on a given day
People still make the decisions.
AI will never reject a review or send a reply on its own. It will suggest; your team will decide.
A flag is not a rejection.
A review flagged by the review checks will go to the approval queue. Someone on your team will approve or reject it.
Personal data will be masked.
Before any text reaches the model, personal data such as email addresses and phone numbers will be masked.
The shopper's words stay theirs.
Shoppers write their own reviews; there will be no AI in the review form and no ready-made sentences. Translations and summaries will never touch the original text or the rating.
Summaries will be labelled and recorded.
The summary will say it was generated with AI. Every version will be recorded, so you can prove in an audit what was shown on a given day.
- Regulation on Commercial Advertising and Unfair Commercial Practices, Art. 28/B (Official Gazette 33297)
- Advertising Board, Guideline on Consumer Reviews (12 September 2023)
This page explains how Sayproof applies these rules and is not legal advice. Consult your legal team about your own situation.
Let's see it on your store.
Tell us about your platform and the markets you sell in, and we will show you how Sayproof would work on your site with your own data.
Read the documentation first(opens in a new tab)We have your request.
We will get back to you at the email address you gave.