Reviews AI
Choose what AI does with your reviews: which functions run, which only make suggestions, and the instructions they follow.
Reviews AI reads every review that comes in. It detects the review's sentiment and how credible it looks, finds profanity and personal data, suggests replies, and writes the summaries shoppers see. You choose which functions run and which only make suggestions for your team.
Navigate to Settings → AI & Agents → Reviews to manage Reviews AI.
How Reviews AI works
Reviews AI analyzes each review when it comes in, and again after the reviewer edits it. It only runs for workspaces whose plan includes Reviews.
- The master switch: Reviews AI turns the functions you can switch on or off. The always-on functions keep running when it's off.
- Modes: Functions that only add information to a review are Off or On. Functions that propose something for your team to do are Off or Suggest. The rest, such as Content safety, are Always on.
- People decide: AI never publishes, rejects, or hides a review on its own. Your hold rules and your team decide what's published: see Moderation.
Cevoid turns on Reviews AI the first time your plan includes Reviews, with every function at its default. If your workspace already had AI settings saved by then, Reviews AI keeps the state it had.
Cevoid also detects the language of each review and translates reviews, whatever these settings say. See Translations.
Functions and defaults
The page groups the functions by stage: Understanding, Moderating, and Distributing.
| Function | Stage | What it does | Modes | Default |
|---|---|---|---|---|
| Sentiment analysis | Understanding | Detects the review's tone and emotions, and flags when the text disagrees with the star rating | Off, On | On |
| Credibility scoring | Understanding | Scores how trustworthy a review looks, with a reason | Off, On | On |
| Content safety | Understanding | Finds profanity, personal data, and harmful content | Always on | Always on |
| Signal extraction | Understanding | Picks out pros, cons, issues, and questions | Always on | Always on |
| Replies | Moderating | Flags reviews that need a reply, and drafts one | Off, Suggest | Suggest |
| Misfiled reviews | Moderating | Spots reviews filed as the wrong type | Off, Suggest | Suggest |
| Media safety | Moderating | Flags unsafe or sensitive content in photos and videos | Always on | Always on |
| Review summary | Distributing | Writes the summary shoppers see, once there are 30 written reviews | Always on | Always on |
| Media accessibility | Distributing | Writes alt text, video descriptions, and captions | Always on | Always on |
Where you see what Reviews AI does
Each function shows its results in a different place, in the review lists, on the review itself, or on your store.
| Function | Where you see it |
|---|---|
| Sentiment analysis | A sentiment icon on each review in the lists, the Sentiment and Emotions filters, and the note The text doesn't match the star rating |
| Credibility scoring | A credibility meter on the review: Credible, Uncertain, Low, or Spam. Hover over it for the score, the signals, and the reason |
| Content safety | The Profanity, Personal info, and Harmful content badges on the review, and masked text on your store |
| Signal extraction | The Highlights, Cons, and Question badges on the review. Review summaries and suggested replies build on it |
| Replies | The Reply suggested badge in the review lists, and a drafted reply in the reply composer |
| Misfiled reviews | A suggestion on the review to Convert to company review or Convert to product review |
| Media safety | Media flagged on the review, and the Has media findings filter |
| Review summary | AI summary widgets and review list summaries on your store, and What people say in the dashboard |
| Media accessibility | Alt text, video descriptions, and captions in review widgets, translated into each market's language |
Content safety and Credibility scoring also feed your hold rules. A review with profanity, personal data, or harmful content can be held for your team, and so can a review whose credibility is Low, through the Low-credibility / likely-fake rule. With Credibility scoring off, that rule can't hold anything. See Moderation.
Turn Reviews AI on or off
Reviews AI is the master switch for the functions you can switch. With it off, Sentiment analysis, Credibility scoring, Replies, and Misfiled reviews stop, and the always-on functions keep running.
- Navigate to Settings → AI & Agents → Reviews
- Enable or disable Reviews AI
- Click Save changes
Change a function's mode
Each function you can switch has its own mode, under the Reviews AI master switch.
- Navigate to Settings → AI & Agents → Reviews
- Select a mode on the function's row: Off or On, or Off or Suggest
- Click Save changes
The mode controls are inactive while Reviews AI is off.
Add instructions to a function
Instructions steer what a function does, in your own words, up to 2,000 characters per field. Sentiment analysis, Credibility scoring, Replies, and Misfiled reviews take instructions. Replies has two fields: When to reply decides which reviews get a suggestion and how, and How to write the reply shapes the draft. For example: "Reply privately to anything about a damaged or wrong item."
- Navigate to Settings → AI & Agents → Reviews
- Click the function's row to open it
- Enter your instructions in Instructions, or for Replies, in When to reply and How to write the reply
- Click Save changes
A row opens only while Reviews AI is on. Always-on functions take no instructions here.
Set your brand voice
Your brand voice describes how your brand writes. Reviews AI uses it when it drafts replies, on top of the instructions for Replies.
- Navigate to Settings → AI & Agents → Brand voice & tone
- Enter a sentence or two in Describe your voice. The field's example is "Warm, concise, a little playful. Never corporate.".
- Click Save changes
Suggested replies
Replies decides which reviews need a reply, and whether the reply should be public or private, from the whole review: its text, rating, and question answers. It then drafts a reply in the reviewer's language and your brand voice. A person always sends the reply. With Replies on Off, there are no suggestions. To answer reviews, see Replies.
Misfiled reviews
Misfiled reviews spots reviews filed as the wrong type, such as a product review that's about delivery, and suggests converting it. Your team decides: Cevoid never converts a review on its own. To convert a review, see Review details.
Review summaries
Review summary writes a short summary of what reviewers say: one for each product, one for your company reviews, and one for each widget that shows a selection of reviews. It's always on: whether shoppers see a summary depends on the widgets you add. See Review widgets.
- When it appears: Once there are 30 published reviews with text. Star-only ratings don't count.
- How it reads: It says how common a point is in words, such as "most reviewers" or "a few reviewers", and never quotes numbers
- Languages: It's written in your workspace's primary language and translated into every market language
- Updates: It updates when enough new written reviews arrive, or when the average rating changes by 0.5 or more, at most once every 7 days