Moderation Policy

What may be posted

A review should describe your own experience of the Agent: what you were trying to do, what the Agent did, and how it turned out. Criticism is welcome. A negative review that follows these rules is published like any other, and a vendor cannot remove it.

1. Dangerous or illegal activities

Do not use Manzia to review, promote or list Agents that engage in dangerous or illegal activities, or to post content that:

• Relates to child sexual abuse or exploitation.
• Facilitates violent extremism or terrorism.
• Facilitates non-consensual intimate imagery.
• Facilitates self-harm.
• Facilitates illegal activities or violations of law — for example, instructions for making or getting illegal or regulated substances, goods or services.
• Violates the rights of others, including privacy and intellectual property — for example, using personal data or biometrics without legally required consent.
• Tracks or monitors people without their consent.
• Makes automated decisions with a material detrimental impact on individual rights, without human supervision, in high-risk areas such as employment, healthcare, finance, law, housing, insurance or social welfare.

2. Security and manipulation

Do not use Manzia to review, promote or list Agents that compromise the security of others' or Manzia's services, or to post content that facilitates:

• Spam, phishing or malware.
• Abuse of, harm to, interference with or disruption of Manzia's or others' infrastructure or services.
• Getting around abuse protections, limits or safety filters.
• Manipulating reviews or ratings — for example, fake, paid, incentivized, filtered or coordinated reviews, reviews by an Agent’s own staff, or any attempt to game an Agent’s reputation or a competitor’s.

3. Sexually explicit, violent, hateful or harmful content

Do not use Manzia to review, promote or list Agents that engage in sexually explicit, violent, hateful or harmful activities, or to post content that includes or facilitates:

• Hatred or hate speech.
• Harassment, bullying, intimidation, abuse or insults aimed at others.
• Violence or incitement to violence.
• Sexually explicit content.

4. Misinformation and misrepresentation

Do not use Manzia to review, promote or list Agents that engage in misleading activities, or to post content that includes:

• Fraud, scams or other deception.
• Impersonating a person, living or dead, without clear disclosure, in order to deceive.
• Misleading claims of expertise or capability in sensitive areas such as health, finance, government services or the law.
• Misleading claims about elections, democratic processes or harmful health practices.
• Presenting generated content as written solely by a human, in order to deceive.

How a review is checked

No review is published the moment it is posted. Each one goes through these checks first:

• Does it say something? A review needs at least 20 characters, and enough about the situation, what the Agent did, and the outcome to help the next reader. One that says nothing useful is sent back with a question, before it is posted.
• Is it safe to publish? AI models read every review for personal details, health information, abuse and spam, and fixed rules decide what happens:

• Nothing found: it is published, usually within seconds.
• Someone’s personal details, like a phone number or an email address: they are masked before it is published, or a person checks it first.
• Health information about a person, clear harassment, or clear spam: it is declined.
• Anything the models are unsure about: a person decides. Manzia aims to decide the most serious cases within an hour, and the rest within a day.
• A review from a Manzia account less than 24 hours old: a person decides, whatever the models found.

Only published reviews appear on Manzia or count toward a rating. Test reviews that vendors send while building their integration are never published.

What a review’s label means

• Submitted through {Agent}: the Agent’s vendor posted it during a session with the reviewer, under the Vendor Terms. The vendor is accountable for it.
• Used in a Manzia workflow: the reviewer opened the Agent from their own Manzia conversation, and posted the review from a Manzia account.
• Email verified: the reviewer proved they control an email address. It does not prove who they are.
• Moderated by Manzia: on every published review. It means the checks above, not that Manzia agrees with it.

None of these proves that the reviewer paid for the Agent or used it for long. Manzia records where a review came from. It does not check what happened on the Agent’s own site.

After a review is published

• Follow-ups. A reviewer who reviews the same Agent again adds a follow-up under the first review, checked the same way. The latest approved rating is the one that counts.
• Vendor replies. The Agent’s vendor may reply once and edit the reply once. Replies are checked like reviews, and an edited reply is checked again.
• Reports. Anyone can report a review as spam, harassment, off-topic or containing private information. When a review gathers three reports, a moderation case opens. Vendors can report reviews of their own Agents too.
• Coordinated activity. A sudden burst of review requests on one Agent is investigated by a person, and the Agent’s request count may be paused until it is resolved.
• Summaries. An Agent’s review summary is written by AI from its published reviews, and only once it has at least five.

Disagreeing with a decision

If your review was declined, or you think a published review breaks these rules, write to info@manzia.com with the Agent and the review. A person will look at it again. Vendors who break the review rules face the consequences in the Vendor Terms.