X Bot Detector

X Bot Detector analyzes tweets and accounts on X (formerly Twitter) to identify bot activity and AI-generated content using 22 heuristic signals, 9
swarm detection checks, and 15 AI text detection signals. Everything runs locally in your browser — no data leaves your device, no API keys needed.

HOW IT WORKS

The extension intercepts X's internal GraphQL API responses to extract rich user and tweet metadata — the same data X's own frontend uses. This gives
it access to account creation dates, follower counts, posting sources, listed memberships, and more, enabling far more accurate detection than DOM-only
analysis.

WHAT IT DETECTS

Account signals (11): account age, reputation score (followers/following ratio), posting rate, list memberships, favourites ratio, default profile
detection, username patterns, display name analysis, bio spam patterns, media ratio, and follow-ceiling behavior.

Content signals: hashtag density, spam keywords (crypto scams, engagement bait), emoji abuse, mention bombing, suspicious URLs, text entropy, and
template/low-variance text detection.

Engagement signals: engagement ratio analysis for bought likes/views/retweets, generic reply farming, and vocabulary richness.

Source analysis: identifies bot frameworks, automation tools, and suspicious posting clients from the tweet source app field.

Temporal analysis: detects unnaturally regular posting patterns and rapid-fire tweeting.

REPLY SWARM DETECTION (9 checks)

Groups replies by conversation and checks for text similarity clustering, bot username patterns, generic reply floods, new account prevalence, default
avatars, low reputation repliers, automation tool usage, synchronized account creation dates, and shared suspicious URLs — catching coordinated
inauthentic behavior that single-tweet analysis would miss.

AI TEXT DETECTION (15 signals)

Independent from bot scoring, this module flags tweets likely written by large language models (ChatGPT, Claude, Gemini, etc.) using two complementary
approaches:

Linguistic patterns: signature LLM phrases ("it's important to note", "delve into", "in the realm of"), corporate AI buzzwords, bulleted/numbered list
structure, avoidance of contractions, hedging language density, and a casualness discount for clearly human-sounding text.

Statistical fingerprint: em dash density, semicolon usage, comma density, sentence burstiness (variation in sentence length), word complexity,
sentence-end punctuation distribution, clause complexity, sentence uniformity, and passive voice frequency.

Flagged tweets get a distinct violet "AI Text" badge alongside any bot badges, so you can tell apart bot accounts, human accounts posting AI-written
content, and legitimate organic posts.

SCORING

Each tweet gets a weighted score across all signals:
- 70%+ = Likely Bot (red badge)
- 40-69% = Suspicious (amber badge)
- 20-39% = Mild Signals (blue badge)
- Below 20% = Clean (no badge)

Reply swarms get a pulsing purple badge. AI-generated text gets a violet badge. Badges show "API" (green) when scored with full intercepted data, or
"DOM" (grey) for fallback DOM-only analysis.

WHAT'S NEW IN v3.2.0

- AI text detection with 15 pattern + statistical signals, independent toggle
- Reply swarm detection expanded to 9 checks (added creation-date clustering and shared-URL detection)
- 3 new account-level signals (favourites ratio, media ratio, follow ceiling)
- Redesigned popup with live status indicator and per-category accent colors

100% private. Zero data collection. No external API calls. Everything stays in your browser.