In a multi-store retail chain, not every review carries the same weight. Some are insults or threats aimed at staff; others describe an experience that never happened, whether it is an ad, a bot or a competitor. Hate speech and fake review detection is the first step in separating both from the genuine voice of the customer. This post explains what BranchVoice's free Moderation and trust tools do, how to read their results and how the same analysis runs automatically on your reviews inside the platform.
Why moderation is an operations job
For a regional manager, the problem is not just bad reviews. A threat naming a cashier is something HR and security need to know about. A cluster of similar negative reviews posted on the same day with no concrete detail can drag a store's rating down unfairly. The reverse hurts too: "great product, recommend to everyone" posts with a discount code inflate the average and crowd out real satisfaction in your reports.
Both tools share one goal: keep harsh but legitimate criticism, and filter out content that is harmful or not genuine. "The product didn't meet my expectations, honestly disappointing" is a complaint that deserves attention. Moderation is not about silencing it; it is about finding what should be set apart from it.
What the Hate Speech Detector returns
The Hate Speech Detector first shows an overall likelihood of harmful content for the text you paste. It then scores each of six labels separately and lists the clear ones under "Detected":
- Hate speech: attacks on ethnicity, religion, gender, sexual orientation, disability or nationality.
- Insult: insulting or belittling a person or group.
- Profanity: swearing and vulgar language.
- Threat: threats of violence or harm.
- Harassment: bullying aimed at a specific person.
- Discrimination: calls to exclude a group or treat it unequally.
"Do that again and I'll find you" stands out as a threat, while "The manager of this store is useless" falls on the insult side. The distinction matters: an insult calls for a reply and, if needed, a report to the platform; a threat starts your employee safety process.
How to read the Fake Review Detector
The Fake Review Detector shows how likely a review is not to be a genuine customer experience, plus the warning signals behind that:
| Label | What it means |
|---|---|
| Promotional | Promo codes, links or self-promotion |
| Off-topic | Content unrelated to a product, service or experience |
| Bot-like | Template phrases and repetition |
| Competitor attack | Exaggerated attack without detail that pushes another brand |
| Incentivised | Written for payment, a gift or a discount |
| Overly generic | So generic it could describe any business |
The tool tries not to flag short but sincere reviews. A concrete detail such as "the trousers were a bit long but I had them hemmed" is a strong sign of a real experience. "Brand X is much better, don't buy from this store" carries a competitor attack signal.
The result is a probability, not a verdict. Before disputing a review with Google or a marketplace, you can use the flagged signals as your reasoning, but the decision stays with your team.
How to use the free tools
Both tools sit on the AI tools page under Moderation and trust. There is no sign-up or login: paste the text and press "Analyse". A text can be up to 1,500 characters, and the tools understand multilingual text, Turkish and English first among them.
A few practical notes:
- Your own labels: the "Customise labels" panel lets you rename, rewrite, delete or add labels, up to 12 in total. The AI reads only the description, so state in one clear sentence what you mean.
- Privacy: your text is not stored; the result for the same text is cached only briefly.
- Usage limits: the tools are free, but heavy use triggers a short wait.
How the same analysis runs inside BranchVoice
The free tools work one text at a time. BranchVoice analyses every review it collects from Google Maps, Şikayetvar (Turkey's largest complaint site), Ekşi Sözlük (a popular Turkish forum), X, social media, app stores and marketplaces. Two of the core questions asked during AI review classification are about moderation: whether a review looks like spam, bot output or part of an orchestrated copy-paste campaign, and whether it contains profanity, insults or inappropriate language aimed at staff or the brand.
These probabilities are saved on the review and shown on review screens. More importantly, the platform watches behaviour in aggregate: when 3 suspicious reviews arrive within a week, it opens a suspicious review burst anomaly and sends a warning notification, as described in review anomaly detection. A competitor attack or bot campaign is noticed without anyone reading reviews one by one; the regional manager opens the store's reviews and, if needed, starts a dispute with the platform.
Next steps
Start by pasting a few of your real reviews into the free tools and see which signals matter to you. Then request a demo to see the same analysis running automatically every day across all your stores.