Detect the ads
X doesn’t label.

After you explicitly agree in the extension and sign in, posts near your X feed viewport, author information, quoted posts, and existing ad labels are sent to PromoSift for processing by Jev. Keyword topics are also sent for semantic matching; literal keyword matching and feed controls run locally. Automatic scanning excludes direct-message conversations and unpublished drafts, but a manual selected-text check sends the text you select, which may be private. See our Privacy Policy for providers, retention, and data requests.

No. It detects signals of commercial intent; it cannot prove a private financial relationship. That is why the product uses labels such as “likely promotion” rather than making definitive accusations.

Any classifier can be wrong. An uncertain post can stay visible with a label. If you fold a post, you can expand it again.

PromoSift separates likely lead capture, subtle promotions, direct sales pitches, and ordinary posts. It shows the probability and signals behind a flag.

PromoSift is designed for both English and Chinese posts, including common promotion patterns, discount language, affiliate calls to action, and self-promotion.

Each successful post check uses one credit, including posts already labeled as ads by X and requests served from the result cache. New accounts receive 500 free credits once and can manually claim another 50 once per Beijing day (UTC+8). Signing in does not claim them, missed days cannot be claimed later, and credits never expire. Paid credits are one-time USDC top-ups to your personal deposit address: 1,000 checks per 9.9 USDC, or 5,000 per 29.9 USDC for a single transfer of 29.9 USDC or more. Failed checks are not charged.

Continue with Google or enter your email to receive a six-digit verification code. The first successful verification creates your account automatically—there is no password and no waitlist.

Jev is TypeSafe AI’s System One decision model. Instead of generating free-form text, it returns typed choices, probabilities, and confidence values. PromoSift uses that structure to classify likely promotions and map the result to a reversible feed action.

Jev returns typed probabilities instead of generating a paragraph word by word. PromoSift combines promotion, topic, and optional AI questions into one Jev request per post. The extension can analyze several visible posts concurrently, and a matching completed result can be reused across accounts for 24 hours to avoid another model call. Each successful server request still consumes one credit, including a cache hit. This reuse window is not a deletion deadline. Actual response time varies with the network and whether a result is cached.