The Words TikTok Parent ByteDance May Be Watching You Say (2023)

Forbes obtained a trove of internal documents showing how ByteDance tracks “sensitive words” mentioned on its social media apps. Hundreds of vocabulary lists housed in the company’s “detection tool” illustrate the range of political, social and cultural topics that the Chinese giant is monitoring or suppressing.

_By Alexandra S. Levine_{, Forbes Staff}

Most major social media companies have rules and tools that control what people on their apps can and cannot see. Generally, they’re in place to track different types of content and keep dangerous or illegal material off their platforms—like posts dealing with terrorism, abuse and suicide.

TikTok parent ByteDance is no different. But the Chinese company’s content moderation and monitoring appears to go far beyond what’s standard among American peers like Alphabet-owned YouTube and Meta-owned Instagram.

A Forbes investigation into TikTok and ByteDance revealed that a ByteDance tool, run by staff in China, is tracking mentions of what it considers “sensitive words” across the company’s products. In some cases, where words are marked “must kill,” “forbidden” or “prohibited,” ByteDance may be blocking related posts altogether. The tool is also aiming to track every time one of these words comes up—recording who said it and where they’re located, including for people in the United States.

The Chinese government has targeted people in the U.S. who’ve spoken out against it online, but experts have warned that the average American can be naive about how far their words can travel on the internet, who may be watching and the potential consequences.

ByteDance’s library of “sensitive words,” which are organized into thousands of vocabulary lists, “is proof-positive that there are specific things that they are concerned about and they want to monitor who was saying them, when and how often,” William Evanina, the former head of counterintelligence for the U.S. government, told Forbes.

"They're not just collecting it for collection's sake,” he added.

Forbes obtained internal documents showing hundreds of the word lists housed in this “detection tool” and is publishing them in full below. While this collection of word lists is not exhaustive and there are undoubtedly more in ByteDance’s system, they illustrate the range of political, social and cultural topics that ByteDance is keeping an eye on or suppressing.

ByteDance declined to comment on what these hundreds of lists mean, how or where they are applied, and who created them. Spokesperson Jennifer Banks said only that “there are separate and distinct keyword tools used for different products, each with strict permissions and access controls that enable a select few people within each product or platform to manage or add to them.”

TikTok spokesperson Jamie Favazza said that “we believe many of these list titles have translation errors and are not relevant to TikTok.” (More than 50 that she confirmed are used on the platform, for safety purposes, are noted below.) “Regardless of wordlist names, TikTok's keyword platform operates separately from Douyin's and other China market products, with separate code, separate databases, and is maintained by separate personnel.”

"As a responsible platform, TikTok uses wordlists to help protect our community from hate speech, misinformation, and other harmful content,” Favazza added after publication. “The majority of the list names Forbes provided to us are not used on TikTok, and anyone can easily see content about those topics is available on TikTok through a simple search on the app."

[Editorial note: A handful of list names were partially cut off in documentation reviewed by Forbes. In those cases, numbers were omitted or hyphens have been added.]

^{CHINA & POLITICS}

Lists about Chinese power or culture

At TikTok CEO Shou Zi Chew’s first-ever testimony before Congress, he told lawmakers under oath that “we do not promote or remove content at the request of the Chinese government.” Word lists in ByteDance’s “sensitive words” tool appear to deal with content that Beijing would likely disapprove of—including language critical of China’s government, military, leaders and history. TikTok denied any of the below have ever been used on TikTok.

2346-Tiktok IM Party and Government Negative Word-

Party, government and military negative core words

504 - negative core words of the party, government

505-Name of Government Agency

777-Party and Government Negative Words List

978-Special prohibited words for Xi and Peng

773-Xi Peng variant vocabulary

774-Mao Deng Jianghu and its variants

975-Leader Preference Targeted Prohibited Words

981 - Leader Relatives Orientation Prohibited Words

2391-FS Project-Xi Peng-Neutral & Negative Inaccurate

2392-FS Project - Party, Government and Military Relat-

1673-FS Project - Universal (Core Leader/Kind Variant)

1255-FS Project - Fraud User Glossary

2389-FS Project- June 4th Related-Negative Inaccurate

956-June 4 must-kill word list

1312-Vertical June 4th Special Review Speech

1482-Falun Gong

2535-Thematic Strategy for China’s Strategic Policy

Lists that mention “Taiwan” or “Hong Kong”

Taiwan and Hong Kong have long represented two of the largest threats to the Chinese Communist Party and authoritarian rule because they enjoy considerable autonomy from mainland China and are democratic in their ideals. Under Xi Jinping, Hong Kong has seen a fierce crackdown on pro-democracy activists and other challengers of his regime, prompting anti-government protests in recent years. Taiwan’s own push for independence has repeatedly prompted retaliation from the Chinese military (most recently, in April, after Taiwanese and American leaders met on U.S. soil). The island has long been at the center of a geopolitical tug-o-war between the U.S. and China, and some fear it could be at the heart of a war between the superpowers in the future. TikTok denied any of the below have ever been used on TikTok.

3034-Local Life-Taiwan Independence & Hong Kong Independence & Tibet Independence & Xinjiang Independence

1140-Hong Kong-related special vocabulary for live bro-

1861-Live China and Taiwan Title Activity Vocabulary

2284-High-risk phrases in Taiwan emergency queue

2180-Prohibited artist words in music in Taiwan

2120-Taiwan’s emergency queue in the live broadcast

2220-Taiwan Blind Box Kaiping Comment Blocking Voc-

2642-High-risk Adjuvant Vocabulary for Taiwan Emerg-

2753-Taiwan-related theming (title)

2758-Taiwan-related theming (comment)

2758-Taiwan-related theming (review)

2776-im Taiwan sensitive word test

Lists about geopolitics

As tensions intensify between the U.S. and China and other authoritarian adversaries, some lists in the tool could shape global discourse around American politics, U.S.-China relations and war in Ukraine and Russia or other parts of the world. TikTok denied any of the below have ever been used on TikTok.

977-Trump Directed Prohibited Words

982-Sino-US trade directional prohibited words

508-North Korea-related core words

976-Putin Directed Prohibited Words

2350-Interactive Russian-Ukrainian Tag Temporary Vo-

623 - Current Leader Words

1632-G Leaders’ Special Retracement Speech

2007-G-Competitor block word list

503-separatist forces core words

2749-LS Theming - Coup, War (title)

2754-LS Theming - Coup, War (Review)

3253-im Private Chat Politics-related Active Control

^{MARGINALIZED GROUPS}

Lists that mention “Tibet”

Tibet, governed as an autonomous region of China, is similarly viewed as a threat by the Chinese government. Beijing’s persecution of Tibetans—and efforts to quash their political, religious and cultural freedoms—is well documented by the State Department and human rights groups. That has included the jailing of supposed dissidents (everyone from teachers to musicians to religious leaders) and restrictions on freedom of expression in the media and online. TikTok denied any of the below have ever been used on TikTok.

2553-Tiktok audio sensitive words in Tibet region

2553-Audio Sensitive Words in Douyin Tibet Region (P-

2471-Tibetan Blocked Words for Douyin Comments

1486-Toutiao Tibetan Poetry Risk Vocabulary

3034-Local Life-Taiwan Independence & Hong Kong Independence & Tibet Independence & Xinjiang Independence

Lists that mention “Uyghur” (also spelled “Uighur”) or “Xinjiang”

Uyghurs, an ethnic minority living primarily in China’s Xinjiang region, have been the victim of Chinese genocide. China has in recent years built a sprawling operation of internment camps and fortified detention centers across Xinjiang, where Uyghurs and other Muslim minorities have been subject to torture and a range of human rights abuses. The U.S. government has labeled the yearslong escalation a “genocide,” while the U.N. Human Rights Office has in the last year described violations as potential “crimes against humanity.” Reports have also shown a rise in forced marriages between Uyghur women and Han men (China’s ethnic majority), calling these partnerships “forms of gender-based crimes that violate international human rights standards and further the ongoing genocide.”

TikTok said none of the below word lists referencing 'Uyghur' are used on TikTok. “TikTok's policies prohibit claims that Uyghur camps in China don't exist or are fake,” Favazza said in a statement. “However, content that is educational or raises awareness about Uyghur camps is allowed on TikTok. One of the ways we enforce this policy is through keywords.”

2470-Tiktok comment Uighur blocked word

2528 - Theming Strategies of Uyghur-Han Couples (Tit-

2540-Uyghur and Han Couples Theming Strategies

2798-IM Uyghur private letter review word list

3245-Uighur Audio Review Vocabulary-Uyghur

2103-Special vocabulary related to Xinjiang

3244-Xinjiang Region Audio Review Vocabulary

718-Sensitive words in Douyin videos in Xinjiang (prelim-

3034-Local Life-Taiwan Independence & Hong Kong Independence & Tibet Independence & Xinjiang Independence

^{SCIENCE & CULTURE}

Lists about science and medicine

Some lists in the tool seem to monitor conversations around China and the Covid-19 pandemic. They appear to reference one epicenter of the outbreak—the east China city of Putian—as well as a “leaked experiment” and pangolins, a species of mammal that early on was rumored to be responsible for spreading the coronavirus from animals to humans. TikTok denied any of the below have ever been used on TikTok.

2894-2196ab Leaked Experiment Vocabulary

2895-1880ab Vocabulary for Missing Experiments

2896-2561ab Omission Experiment Vocabulary

2897-2560ab Leaked Experiment Vocabulary

2898-1168ab Leaked Experiment Vocabulary

2715-Pangolin Title Sensitive Vocabulary

1532-Putian Hospital Vocabulary

2654-Medical Special Review Vocabulary

3031-Medical Related

3035-ugc Medical Audit Highlights

3041-Medical

3283-Medical ASR black word

3285-Medical Title Black Thesaurus

Lists about global culture

Some lists in the tool appear to be concerned with free expression in autonomous parts of China and beyond—across the West. They deal with everything from music, poetry and books to sports leagues and pro-athletes (the NBA, World Cup and soccer player Mesut Özil). They also touch on the stock market, real estate and foreign languages and cultures. TikTok denied any of the below have ever been used on TikTok.

1539-Words banned from the media (other)

2180-Prohibited artist words in music in Taiwan

1486-Toutiao Tibetan Poetry Risk Vocabulary

997-Spirit Dog Added Poetry Vocabulary

1983-FS Project-Special “Night Watchman” Vocabulary

2288-Fizzo Erotica List

1808-NBA Copyright Glossary

3481-Qatar World Cup Official Cooperation Exemption

3487-Qatar World Cup Characteristic Words

2778 - Ozil Theming Strategy (Title)

2779-Ozil Theming Strategies (Review)

1582-hebrew_sensitive_text

2538-Islamic Theming Vocabulary (Review)

2526-Backpack-Islamic Theming Glossary (Title)

517-regional group black

2765 - Sicilian Business Vocabulary

983-Stock Market Orientation Prohibited Words

1802-Specially sensitive words in real estate content in-

1803-Specially sensitive words in real estate content in-

^COMPANIES

Lists about ByteDance rivals

Several lists viewed by Forbes concern TikTok’s biggest competitor in the U.S., YouTube, as well as products from ByteDance’s fiercest challengers in China—including Alibaba’s cloud software Aliyun and Tencent’s messaging app WeChat. TikTok denied any of the below have ever been used on TikTok.

1618-YouTube Domestic Surveillance

1710-TikTok-Twitter public opinion keywords

476-aliyun_sensitive_test

Wechat business control words

1058-Fly chat group search block

1060-Fly chat user search blocking

1061-Search blocking for users in Feichat group

1123-Flychat file name forbidden words

Lists that mention “Douyin” and other ByteDance products or companies

While some of the ByteDance word lists directly cite TikTok, others explicitly mention the Chinese version of the app, Douyin, which is heavily censored by the Chinese government. The lists also reference ByteDance’s news service Toutiao, music-streaming platform Resso and workplace software Lark (known as Feishu in China), among other products past and present. TikTok denied any of the below have ever been used on TikTok.

718-Sensitive words in Douyin videos in Xinjiang (prelim-

2471-Tibetan Blocked Words for Douyin Comments

2553-Audio Sensitive Words in Douyin Tibet Region (P-

1486-Toutiao Tibetan Poetry Risk Vocabulary

2817-Toutiao’s private letter abuse risk warning words

1593-Qingbei Online School User Feedback Keywords

1599-Dali Desk Lamp User Feedback Keywords

1830-Xingfuli Agent Questions and Answers Sensitive

1592-Guagualong Enlightenment User Feedback Keywords

1928-Resso-GP Feedback Matching Vocabulary

Testing Lark

858-Company Product Negative Sensitive Vocabulary

Lists that mention “TikTok” (also “TT”) or “U.S.”

Forbes found almost 100 lists in the tool with “TikTok” or “U.S.” in their name—some concerned with language critical of the Chinese government or blocked speech about persecuted Uyghurs. TikTok denied that about half of them had ever been applied to its platform, suggesting many could be the result of translation errors from Chinese to English. (See below for those the company confirmed.) Favazza, the TikTok spokesperson, said access to TikTok’s word lists is controlled by TikTok’s trust and safety team and that any changes to those lists goes through a U.S. team member. Internal materials show that employees in China are also among those managing some TikTok lists.

LISTS THAT TIKTOK SAID WERE NOT USED ON TIKTOK:

2346-Tiktok IM Party and Government Negative Word-

2553-Tiktok audio sensitive words in Tibet region

2470-Tiktok comment Uighur blocked word

TikTok Japanese Comments Suppress Words

1710-TikTok-Twitter public opinion keywords

1565-Tiktok Government Affairs Media Topic Vocabulary

1688-Tik Tok Burmese Vocabulary

1283-Tiktok pedophile specific words

1420-Tiktok group chat pornographic blocking vocabulary

1426-Tik Tok Porn Drainage Recognition Vocabulary

2186-Tiktok Single Male & Celebrity Themed Vocabulary

2271-Tiktok Sensitive Person Theme Vocabulary (Title)

2272-Tiktok Sensitive Person Themed Vocabulary (Co-

2780-Tiktok push-sensitive people/prohibited audio

1507-Tiktok pretending to be a celebrity name list

2231-Tik Tok-Data impersonation-Special List Policy P-

2232-Tik Tok-Data impersonation-Political Media Account

2234-Tiktok-Data impersonation-Strategic protection

120-TikTok Private Message Sensitive Words

2746-Tiktok Xiaoan Private Message Sensitive Word Fil-

838-TikTok special activity—Comment review first (p-

1585-TikTok Special Events-Comment Full Review Voc-

2109-Tiktok comment bottom word list

2548-TikTok Historical Nothingness Video Recall Vocab-

839-TIkTok Special Event—Video First Review (Prelim-

1576-TikTok Video Title No. 1 Itinerary Special Vocabulary

2196-TikTok video title & nip review + word list release

2416-Tiktok video title & nip release + must-kill word list

2561- TikTok long video title & nip review + word list re-

2597-TikTok long video title & nip release + kill word list

1807-TikTok Audio Review Keyword Prompt

679-Tik Tok Hot Search Filter Words

84-Prohibited words for TikTok user nicknames

457-Tiktok corporate private letter sensitive words

635-User Feedback Filter Did/Uid-Tik Tok

946-Tiktok user nickname is not recommended word list

1075-User Feedback Filter Words-Tik Tok Red Packet

1159-Tiktok user feedback keywords

1403-TikTok POI Blocking Vocabulary (Activity)

1455-Tiktok high-risk prohibited word list (queue high-)

1490-Tiktok report user feedback keywords

1542-TikTok-Tns user feedback keywords

1653-Tiktok Topic Graded Vocabulary (Personal Review)

1798-TT Feedback Algorithm Matching Keywords

1820-Tiktok poi No. 1 itinerary special vocabulary

1843-Tiktok Live No. 1 itinerary special vocabulary

1981-Tik Tok Collection Title Prohibited Words List (Co-

2139-Tiktok commercialization snapshot review queue

2201-TikTok Special Time Node Themed Vocabulary (T-

2202-TikTok Special Time Node Themed Vocabulary-

2725-Vocabulary for TikTok special area titles to be rev-

2874-Tik Tok Emergency Response-Topic Automatic P-

(US 1707)DM - Grayscale test wordlist

LISTS THAT TIKTOK CONFIRMED ARE USED ON TIKTOK FOR SAFETY REASONS:

1327-(US 570) MT_User Profile Tier1 Wordlist

577-MT_Search filter Words (User)

1333-(US 577)MT_Search Tier1 Words (User)

579-MT_Search ban Words (Hashtag)

1335-(US 579) MT_Search ban Words (Hashtag)

908-Hatespeeech PSA Ban Search (All)

1344-(US 908) Hate. PSA Ban Search (All)

909-Hatespeech PSA Ban Search (Music

1345-(US 909) Hates PSA Ban Search (Music)

932 - tiktok-m personalized push sensitive vocabulary

990-General PSA Ban Search List (All)

1348-(US 990) General PSA Ban Search (ALL)

991-General PSA Ban Search (Music)

1349-(US 991) General PSA Ban Search (Music)

1041-Ban Search Suicide & Self Harm

1352-(US 1041) Ban Search Suicide Wordlist

1080-TT-Search Sug Auto-Remove Words

1353-(US 1080) Search Sug Sensitive Wordlist

1150-User Feedback-Overseas Bulk Reply

1243-MT-DM Ban Words List

1275-MT-Search Sug Whitelist

(US 1293) NLP Urgent Support Words List

1322-(US) musical.ly_emergency words

1417-ED Ban Search List

1481-(US 1417) ED Ban Search List

1548-(US)DM Adult wordlist-T1

1554-TT Anti-semitic Search Ban wordlist

2010-(US 1554) TT Anti-semitic Search Ban word list

1700-TT Search Suicide T2 wordlist

1734-TT Distressing Search Reminder Wordlist

1737-(US 1734) TT Distressing Search Reminder Wordlist

1735-TT Inappropriate Search Reminder Wordlist

1738-(US 1735)TT Inappropriate Search Reminder Wordlist

1736-TT Anti wildlife trafficking Search wordlist

1739-(US 1736)TT Antiwildlife trafficking wordlist

1791-TT Project Survivor SearchBan wordlist

1792-(US 1791)TT Project Survivor Search Ban wordlist

1793-TT Sug Variants moderation word

1886-TT Search Aid-Dangerous challenges wordlist

1889-(US1886)-TT Search Aid-Dangerous challenges wordlist

1887-TikTok Search Aid-SSH Hoaxes wordlist

1888-(US1887)TikTok Search Aid-SSH Hoaxes wordlist

1975-TT Search Aid-CSAM wordlist

1976-(US1975)TT Search Aid-CSAM wordlist

2111-TT-Search Sug-Variant Seed Wordlist

(US 2193) TikTok suggested search inter-

2297-(US 2193)TikTok suggested search intervention wordlist

2291-SEA Personalized Candidate Words Black Words List

2292-TikTok User Profile Enqueue Wordlist

TikTok Search Result Layered Intervention

TikTok Q&A question automatic takedown

MT_Comment Auto-remove words

Trust+Safety lists common to major social media companies

To protect social media users, most major social media platforms—including TikTok—have policies and tools aimed at filtering out content and language that is violent, vulgar or illegal.

908-Hatespeeech PSA Ban Search (All)

1041-Ban Search Suicide & Self Harm

1432-Bullying/harassment

1433-Misinformation/Media manipulation

1434-Terrorists & criminal organizations

1439-Political

1623 - Children’s Soft Porn OCR Vocabulary

child porn_one_level

child porn_two_level

2282-tf-idft word v2 9014 harassment

2284-tf-idf word v2 0914 misinformation

2286-tf-idft word v2 0914 suicide

2289-tf idf word v2 09114 misinformation new method

771 - Full text review of false content

772 - False content title for review

Miscellaneous lists in the tool

The meaning of many lists in the tool is unclear, and some appear to be more technical or related to the basic functioning of ByteDance apps. Still, many of these lists contain words that are “forbidden,” “banned,” “high-risk” or “sensitive.”

1173-Intelligent Monitoring Alarm Sensitive Vocabulary

1318-Forbidden word list

1523-Forbidden Word List-BR

2589-Private letter abduction vocabulary list

2284-High-risk phrases for emergency queues in live b-

high risky words

1838-xxxx sensitive words

1147-SMS activates sensitive words

1260-G-Global Distribution Nickname Sensitive

849-Search Operation Push Word Vocabulary

Feedback Form - Access Party Diversion

816-User Feedback Filter Region-X Project

812-X Project User ID Banned Words

815-X Project Vocabulary (Row-inducing words)

1179-DMT Virtual Vocabulary

1512-mt_web

1513-mt_transitive

658-M_ban

M-Review

3032 - Inferior Marketing

3033 - Marketing Sense

3036 - test grayscale

3037-dry

2298-Live comment AB test

3276 - Test openAPI

peadar test 15test test test test test test

2285-tf-idf word v2 0914 ansa

This story has been updated to provide additional comment from TikTok.

Emily Baker-White contributed reporting.

MORE FROM FORBES

MORE FROM FORBESTikTok Parent ByteDance's 'Sensitive Words' Tool Monitors Discussion Of China, Trump, UyghursBy Alexandra S. LevineMORE FROM FORBESIndia Banned TikTok In 2020. TikTok Still Has Access To Years Of Indians' DataBy Alexandra S. LevineMORE FROM FORBESTikTok's China ProblemBy Emily Baker-WhiteMORE FROM FORBESHow A TikTok Ban Would Deal A Blow To Creators, Businesses And The American EconomyBy Alexandra S. LevineMORE FROM FORBESSecurity Failures At TikTok's Virginia Data Centers: Unescorted Visitors, Mystery Flash Drives And Illicit Crypto MiningBy Emily Baker-White