What counts as a hidden promotion?
A hidden promotion is content that behaves like marketing even though the platform does not label it as an advertisement. It may include an affiliate link, a creator discount code, a product-focused call to action, repeated selling points, or a funnel to an external signup page.
That does not automatically mean the author was secretly paid. PromoSift therefore distinguishes likely commercial intent from evidence of a financial relationship. The product should flag what can be observed in the post without inventing facts about private arrangements.
Words in context
Interpret marketing language as part of the full post.
Link signals
Recognize product destinations and affiliate-style URLs.
Promotion patterns
Combine discount, urgency, feature, and CTA signals.
Clear uncertainty
Show confidence and reasons instead of pretending certainty.
Traditional blockers match labels and selectors. Jev handles the bounded semantic decision: what kind of promotion is this, how strong is the commercial intent, and how confident is the result?
Signals a hidden-ad detector can consider
No single word proves that a post is promotional. Stronger classification comes from combining multiple signals: product links, discount codes, pricing, urgency, purchase language, signup prompts, repeated brand framing, affiliate URL patterns, and the relationship between the surrounding text and the destination.
Context also prevents obvious mistakes. A journalist reporting on a company, a user criticizing a product, and a creator offering a discount may mention the same brand for very different reasons.
I showed her how I make over $400K a month from digital products. Here are the seven concepts.
Explain the flag instead of issuing a verdict
A useful detector should show reasons such as “discount code,” “purchase call to action,” or “external product link.” It should use language such as “likely promotion” or “commercial intent detected,” not “this creator was definitely paid.”
That keeps the interface accurate and gives the user enough information to make the final decision.
Analyze only what the feature needs
Keyword matching and display controls run in the browser. Visible post text and existing X ad labels are sent to Jev for classification. Each successful classification uses one credit, including a cached result. Direct messages and unpublished drafts are outside scope.
Frequently asked questions
Can an AI know whether a post is legally an advertisement?
Not from public text alone in every case. PromoSift is designed to detect observable promotional signals, not make a legal determination.
What happens when the model is uncertain?
The safest default is to keep the post visible and add a low-key label rather than hiding it.
Does the detector analyze images or video?
PromoSift currently focuses on text, links, labels, and surrounding post context. Image and video analysis can be added as a separate detection layer later.
What is Jev in PromoSift?
Jev is TypeSafe AI’s System One decision model. PromoSift uses it to classify visible post context into typed promotion categories and confidence values that can be mapped to label, dim, fold, or hide actions.
Detect what basic blockers miss.
Enter your email, verify the six-digit code, and receive a one-time grant of 500 free Jev decisions that never expire. Click daily check-in to claim 50 more each Beijing day (UTC+8); missed days cannot be claimed later.
Log in to PromoSift →