> ## Documentation Index
> Fetch the complete documentation index at: https://docs.bodyguard.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Text classifications
> A complete reference to Bodyguard's text taxonomy: the signals we use to describe content and the classifications we detect.
Bodyguard reads every comment, message, or post and describes it with a set of labels. Those labels tell you **what kind of content it is**, **how serious it is**, and **who it is aimed at**. Together, they form the text taxonomy.
The taxonomy has two layers:
* **Moderation signals** — the *type*, *severity*, and *target* of the content. These qualify the content and help decide what to do with it.
* **Classifications** — the detailed labels that describe precisely what the content contains (an insult, a scam, an advertisement, and so on).
Both layers appear in every analysis result, whether you access Bodyguard through the dashboard or through the API.
***
## Moderation signals
Three signals sit alongside the classifications to refine every moderation decision. The same classification can require a very different action depending on these signals — an insult marked `LOW` and aimed at no specific person is not handled like an insult marked `CRITICAL` and aimed at a user.
### Type
The type is the overall nature of the content. It groups classifications into families so you can filter and prioritise at a glance.
| Name | Technical value | Definition |
| ----------- | --------------- | ----------------------------------------------------------------------------------------------------------------------------------------- |
| Hate speech | `HATE_SPEECH` | Content that is explicitly violent or discriminatory — racism, LGBTQIA+ phobia, misogyny, ableism, moral or sexual harassment, and so on. |
| Hateful | `HATEFUL` | Content with malicious intent such as insults, denigration, aggression, condescension, or body-shaming. |
| Undesirable | `UNDESIRABLE` | Content that disturbs the experience of the community or the normal flow of conversation (spam, floods, ads). |
| Criticism | `CRITICISM` | Content with a negative tone that is not hateful — typically complaints or unfavourable opinions. |
| Neutral | `NEUTRAL` | Content that displays neither supportive nor problematic signals. |
| Positive | `POSITIVE` | Supportive, encouraging, fair-play, or body-positive content. |
| None | `NONE` | Content that doesn't belong to any of the categories above. |
### Severity
Severity tells you how intense the content is. It's the main lever to adapt moderation: the same type of content can be merely inappropriate, genuinely harmful, or an immediate safety issue.
| Name | Technical value | Definition |
| -------- | --------------- | ------------------------------------------------------------------------------------------------------------------------------------- |
| None | `NONE` | A message not considered harmful or offensive. |
| Low | `LOW` | A message that may be offensive or inappropriate. |
| Medium | `MEDIUM` | A message with the potential to cause harm or distress, but not urgent or severe. |
| High | `HIGH` | A message likely to cause significant harm or damage, such as hate speech or discrimination against protected groups. |
| Critical | `CRITICAL` | A message that poses an immediate threat to the safety or well-being of individuals or groups, such as content that incites violence. |
### Directed at
The target tells you who the content is aimed at. The same insult directed at a public figure, a family member, or an entire community does not carry the same weight, and moderation policies often differ accordingly.
| Name | Technical value | Definition |
| --------------------- | ------------------- | ----------------------------------------------------------------------------------------------------------------------------- |
| User | `USER` | The comment responds to the author of the original content or directly to another user. |
| User's family | `USER_FAMILY` | The comment targets the family of the author of the original content or of another commenter. |
| Single person | `SINGLE_PERSON` | The comment refers to a specific third person (typically named or referred to with he/she). |
| Group | `GROUP` | The comment targets a group of individuals (a community, a demographic, a team). |
| Everyone | `EVERYONE` | A general statement with no specific target or aimed at a wide range of people. |
| Author of the comment | `AUTHOR_OF_COMMENT` | The commenter is talking about themselves, or including themselves in the sentiment. Important for signals such as self-harm. |
| Haters | `HATERS` | The comment addresses critics or detractors of the author or of the brand. |
| Entity | `ENTITY` | The comment refers to a company, brand, or other non-human entity. |
| None | `NONE` | No target identified. |
Moderation signals are returned on every analysis. Classifications — described below — are the actionable labels applied to each piece of content.
***
## Classifications
Classifications are the detailed labels Bodyguard assigns to content. A single piece of content can carry **several classifications at once** — an insulting comment containing a phone number will be labelled both as an insult and as personal information.
Classifications come in three tiers:
* **General** — available to everyone. Used in moderation decisions and always returned when detected.
* **Additional** — monitoring-only labels available on the Advanced plan and negotiated per contract. They enrich reporting and community insights without driving automated moderation.
* **Custom** — classifications built on request for a specific customer. See [Custom classifications](#custom-classifications).
Each card below shows the human-friendly name, the technical value used by the API, a short description, and a real example.
### Harmful content
`INSULT` · Hateful
Disrespectful or abusive language targeting an individual.
*"You all are cocksuckers!"*
`THREAT` · Hate speech
Comments designed to intimidate or scare another person by promising to endanger them mentally or physically.
*"I'm going to find you, and then I'm going to kill you."*
`RACISM` · Hate speech
Discrimination or prejudice on the basis of race or ethnicity, typically towards minorities or marginalised groups.
*"Go back to your country."*
`HATRED` · Hateful
Comments expressing fundamental hostility, aimed at insulting or injuring a person, group, or entity.
*"I hate you, you deserve nothing."*
`MISOGYNY` · Hate speech
Discrimination or prejudice against women, including comments reflecting hateful patriarchal or transphobic attitudes.
*"Go make me a sandwich, woman."*
`ABLEISM` · Hate speech
Discrimination or prejudice against individuals living with disabilities — mental, psychological, or physical.
*"The paralympics are shit."*
`LGBTQIAPLUS_PHOBIA` · Hate speech
Discrimination or prejudice against people who identify as members of the LGBTQIA+ community.
*"Trans people are sick in the head. They need therapy."*
`BODY_SHAMING` · Hateful
Discrimination or prejudice targeting someone's physical appearance or body modifications.
*"You look like a fat, disgusting cow."*
`SEXUAL_HARASSMENT` · Hate speech
Inappropriate sexual comments or unwelcome advances, from flirtatious remarks to violent sexual fantasies.
*"Wanna see you naked."*
`MORAL_HARASSMENT` · Hate speech
Comments and abusive behaviour designed to undermine and humiliate a person.
*"You really are useless aren't you."*
`TROLLING` · Hateful
Confusing or manipulative messages posted to provoke a reaction.
*"Just like Belgium at the World Cup."*
`VULGARITY` · Neutral
Socially unpleasant, offensive, or obscene language without necessarily targeting anyone.
*"Oh fuck, I can't believe this shit!"*
`DOXXING` · Hateful
Publishing, or threatening to publish, someone's personal information or intimate content without consent.
*"Here's her address: 1234 S Canal St, Chicago, IL 60607."*
`SELF_HARM` · Hate speech
Content expressing distress, intent to harm oneself, or suicidal ideation. Flagged so teams can offer support — never removed automatically.
*"I can't do this anymore. This life's not worth living."*
`REPUTATION_HARM` · Hateful
Comments intended to damage the reputation of an individual or an entity.
*"\[Brand] is in cahoots with pedophiles."*
`GEOPOLITICAL` · Neutral
Comments about politics, especially international relations influenced by geographical factors.
*"#FreePalestine"*
### Explicit content
`SEXUALLY_EXPLICIT` · Neutral
Comments that talk about or detail sexual acts or genitalia.
*"I'll let you bend me over."*
`PEDOPHILIA` · Hate speech
Comments exhibiting or promoting sexual attraction towards minors.
*"kali ur 13 but respectfully u are hot"*
`PEDOPHILIA_REFERENCE` · Neutral
Reference to child sexual abuse or illegal sexual activity with minors, excluding first-person statements of intent.
*"Nude photos with under 12 year olds are available on the dark web."*
`DRUG_EXPLICIT` · Neutral
Comments that talk about or encourage the use of drugs.
*"I'm so looking forward to dropping acid at music festivals this summer."*
`WEAPON_EXPLICIT` · Neutral
Comments that talk about or encourage the use of arms and weapons. Adapted to context — gaming weapons are treated differently from real-world promotion.
*"I love my AK47 with its high capacity drum mag."*
`TERRORISM` · Hate speech
Content supporting intimidation or coercion through the threat or perpetration of violence.
*"Totally support his action, should have killed more of them."*
`TERRORISM_REFERENCE` · Neutral
Explicit mentions or references to acts of terrorism, terrorist organisations, or associated individuals.
*"This is just like what ISIS did in 2015."*
`UNDERAGE_USER` · Neutral
Content where the user discloses or is understood to be underage. Age thresholds (under 13, 16, or 18) are configurable.
*"Hi, I'm John and I'm 9 years old."*
### Spam & low quality
`SPAM` · Undesirable
Undesirable comments sent repeatedly to a large number of people. Detection relies on time and repetition patterns, not just content.
*Behaviour-based detection — no single example applies.*
`FLOOD` · Undesirable
Comments meant to disturb a platform's normal operation, such as the mass publishing of senseless or repetitive text.
*"ooooooooooooooooooooooooooooooooo"*
`FLOOD_ASCII_ART` · Undesirable
Flood content composed of ASCII art characters.
*"(╯°□°）╯︵ ┻━┻ ┻━┻ ︵ ╯(°□° ╯)"*
`FLOOD_EMOJI_REPETITION` · Undesirable
Flood content consisting of repeated emojis.
*"😂😂😂😂😂😂😂😂😂😂😂😂😂😂😂"*
`FLOOD_STYLISED_TEXT` · Undesirable
Flood content using stylised, decorated, or Unicode-manipulated characters.
*"ｈｅｌｌｏ ｅｖｅｒｙｏｎｅ ｈｅｌｌｏ ｅｖｅｒｙｏｎｅ"*
`FLOOD_UPPERCASE_TEXT` · Undesirable
Flood content written entirely or predominantly in uppercase.
*"WHY IS NOBODY ANSWERING ME WHY IS NOBODY ANSWERING ME"*
`FLOOD_WORD_REPETITION` · Undesirable
Flood content consisting of the same word or phrase repeated.
*"come on come on come on come on come on come on"*
`FLOOD_UNKNOWN_TEXT` · Undesirable
Flood content that does not fit a more specific flood subtype.
*"asdfjkl; asdfjkl; asdfjkl; asdfjkl;"*
`USELESS` · Undesirable
Comments without meaningful content that do not add to or enrich the conversation.
*"1st person in the comment section"*
`SCAM` · Undesirable
Deceptive messages designed to trick people out of money, personal information, or account access.
*"Click my link "Millionaire 111" so that you too can make money."*
`LINK` · Neutral
Comments containing a link to another page or site.
*"[https://www.bodyguard.ai](https://www.bodyguard.ai)"*
`FORBIDDEN_LINK` · Undesirable
Comments containing a link blacklisted for your channel.
*Triggered when a link matches your configured blacklist.*
`FORBIDDEN_LANGUAGE` · Undesirable
Comments written in a language blacklisted for your channel.
*Triggered when content is outside your configured language allowlist.*
`PLATFORM_BYPASS` · Undesirable
Comments inviting users to leave the current platform and continue the conversation elsewhere.
*"Wanna talk on snap?"*
`DATING` · Neutral
Comments indicating intent or interest in pursuing romantic or sexual relationships.
*"Are you single? I'd love to get to know you better 😉"*
### Personal information (PII)
PII sub-labels are returned alongside the parent `PII` label. For example, a comment containing an email address is labelled with both `PII` and `PII_EMAIL`.
`PII` · Neutral
The act of publishing, willingly or not, personally identifiable information online.
*"Actually my number is 9958592028, please call me."*
`PII_EMAIL` · Neutral
A comment containing a personal email address.
*"Reach me at [john.doe@gmail.com](mailto:john.doe@gmail.com) anytime."*
`PII_PHYSICAL_ADDRESS` · Neutral
A comment containing a home or physical address.
*"I'm at 12 Oak Street, Springfield — come find me."*
`PII_PHONE_NUMBER` · Neutral
A comment containing a personal phone number.
*"Call me on +1 555 234 5678."*
`PII_BANK_INFORMATION` · Neutral
A comment containing banking details such as account or card numbers.
*"My IBAN is FR76 3000 6000 0112 3456 7890 189."*
`PII_NATIONAL_ID` · Neutral
A comment containing a national identification number or document reference.
*"My SSN is 078-05-1120."*
`PII_REGISTRATION_PLATE` · Neutral
A comment containing a vehicle registration plate.
*"That's the guy — his plate is AB-123-CD."*
`PII_SOCIAL_HANDLE` · Neutral
A comment containing a social media handle pointing to a personal account.
*"Find me on Instagram @john.doe.real"*
`PII_REAL_NAMES` · Neutral
A comment containing a real first or last name.
*"My name is Jean Dupont."*
### Advertising
ADS sub-labels are returned alongside the parent `ADS` label when the type of advertisement is identified.
`ADS` · Undesirable
Promotional messages advertising products, services, events, or personal brands.
*"Hello, please visit my page, I need a follower 😉"*
`ADS_COLLAB_REQUEST` · Undesirable
Offers to collaborate with other users on social media.
*"Hey, I'd love to collab with you — DM me!"*
`ADS_DIRECT_SALE` · Undesirable
Explicit attempts to sell products or services directly to users.
*"Selling brand new AirPods Pro, €150. DM to buy."*
`ADS_PROFILE` · Undesirable
Attempts to drive other users to the commenter's social media profile.
*"Follow my page for daily tips 👇 @mypage"*
`ADS_PROMO_CODE` · Undesirable
Offers of discounts or promo codes for products and services.
*"Use code SAVE20 for 20% off your first order!"*
`ADS_SELF_PROMOTION` · Undesirable
Attempts to get users to engage with the commenter's events or projects.
*"I just dropped my new track — link in bio 🔥"*
`ADS_SEXUAL_CONTENT` · Undesirable
Explicit attempts to sell sexual material or services.
*"Check my OnlyFans for exclusive content 🔞 link in bio."*
`ADS_TICKET_SALE` · Undesirable
Offers to sell tickets to events.
*"Selling 2 floor tickets for the concert Saturday, €80 each."*
`BETTING` · Neutral
Comments including links to betting websites.
*"Best tips for all today's games — join the link below."*
`STREAMING_LINK` · Neutral
Comments with links to illegal live event broadcasts.
*"Watch the match live for free here → bit.ly/stream99"*
`PLAGIARISM` · Neutral
Comments raising concerns or accusing a brand of stealing concepts or ideas from other brands or creators.
*"Copying other brands won't make you successful. Come up with your own ideas."*
### Positive signals
`SUPPORTIVE` · Positive
Comments showing appreciation for a user or their content.
*"This product is amazing."*
`ENCOURAGEMENT` · Positive
Comments giving an individual or entity your support.
*"Keep it up guys, you can do it!!!"*
`FAIRPLAY` · Positive
Comments expressing fairness and honour, following the spirit of the rules.
*"GG guys!"*
`NEUTRAL` · Neutral
Comments that do not display any supportive or problematic content.
*"It's gonna rain tomorrow."*
### Brand & community
`NEGATIVE_CRITICISM` · Criticism
Critiques of a user's content or actions without the intent of harming the user.
*"I don't like this bag, I think it's not fashionable at all."*
`BOYCOTT` · Criticism
Comments urging users to stop buying, using, or supporting a company's products or services.
*"No one should buy from this brand ever again."*
`DEFAMATORY` · Hateful
False statements of fact presented as true, with intent or effect of damaging reputation.
*"I heard they tested their products on children without consent."*
`CUSTOMER_COMPLAINTS` · Neutral
Comments where a customer criticises the brand's quality, representatives, or service.
*"No call back; can't ever speak with management!!! Poor customer service!"*
`CUSTOMER_REQUESTS` · Neutral
Customer enquiries and requests for assistance about a brand's service or product.
*"Can you please help me identify the name of this particular bag?!"*
`CUSTOMER_SATISFACTION` · Neutral
Comments where a client expresses appreciation for the brand's service, representatives, or product.
*"I met this amazing sales person, made shopping there very easy."*
`HEALTH_CONCERNS` · Neutral
Reports of negative physical reactions, health concerns, or safety risks related to a product.
*"Your product gave me an allergic reaction. Thanks for nothing!"*
### Context & topics
All classifications below are **Additional** — monitoring-only labels that provide community insights.
`POLITICS` · Neutral
Comments related to or mentioning governments, political parties, and political figures.
*"Imagine people using social media to discourage artists after speaking out against Trump."*
`FINANCE` · Neutral
Comments related to monetary and economic issues — management, impact, and ethics of financial practices.
*"Their prices keep rising while quality drops — classic corporate greed."*
`ENVIRONMENT` · Neutral
Concerns or campaigns regarding ecological issues tied to a brand's production processes.
*"This brand's packaging is completely non-recyclable — shameful."*
`ANIMAL_WELL_BEING` · Neutral
Concerns or campaigns regarding a brand's use of animal products.
*"I won't buy from them until they stop using fur."*
`CYBERSECURITY` · Neutral
Comments related to digital security topics.
*"@brand you had a data breach — someone tried to order \$166 worth of product and my bank caught it!"*
`PRICING` · Neutral
Comments criticising high prices or price increases of a brand's products.
*"€400 for a basic t-shirt? Completely out of touch."*
`SEXTING` · Neutral
Sexual comments or invitations that are playful or respectful, within mature-user norms. Differs from sexual harassment in that it does not include violent or degrading content.
*"You looked incredible in that photo last night 🔥"*
`CHEAT_REFERENCE` · Neutral
Mentions or references to cheating websites or practices, primarily in gaming contexts.
*"Report mojo for cheats — pretty sure he is using one."*
***
## Custom classifications
Beyond the catalog above, Bodyguard can build **custom classifications** tailored to your business. Custom classifications are created on request by the Bodyguard NLP team, trained on your industry and languages, and delivered typically within two weeks.
They are the right choice when you need to detect something specific that isn't covered by the general or additional classifications — a brand-specific terminology, an industry-specific risk, or a niche community behaviour.
If you have a need the standard taxonomy doesn't cover, [reach out to our team](https://www.bodyguard.ai/contact-us).