Transparency Center

A social platform makes thousands of judgement calls a day: which post to remove, which account to restrict, which report to escalate, which government demand to refuse. Most of those decisions are invisible to the people affected by them. That invisibility is what erodes trust, not the decisions themselves. We publish this centre so that three questions always have a public answer: what rule applied, who applied it, and how you challenge it.

We commit to four things. We will state our rules before we enforce them, not after. We will tell you which rule your content broke rather than sending a generic notice. We will keep an appeal route open for every enforcement action that affects reach, visibility or account access. And we will report on our own behaviour on a fixed schedule, including the numbers that do not flatter us.

Policy at iTweet starts with a written problem statement: a pattern of harm we have observed, evidence from user reports, and the categories of speech or behaviour that would be caught by any new rule. Every draft is reviewed for over-breadth — the test is whether an ordinary user acting in good faith could be caught by it. Rules that fail that test are narrowed or dropped.

Substantive changes to the Community Guidelines, Terms of Use or Privacy Policy are versioned and dated. We publish material changes before they take effect and, where a change reduces your rights or expands our permissions, we give notice in the app rather than relying on you re-reading a page. Minor clarifications — a rewritten sentence, a fixed link — are made inline and noted in the change log. We do not silently broaden a policy and then enforce it retroactively.

Enforcement combines automated detection with human review. Automated systems handle scale: hash matching for known illegal imagery, spam and bot-network signals, coordinated-behaviour detection, and classifiers for high-confidence categories. Human reviewers handle context: satire, reclaimed slurs, counter-speech, newsworthiness, and any decision that would remove an account rather than a post.

We use graduated consequences. Depending on severity and history, an action may be a label, reduced distribution, removal of the single item, a feature restriction (for example losing the ability to comment or go live), a temporary lock, or permanent termination. Severity — not virality — decides the tier. A first-time borderline post is not treated like a repeat pattern of targeted harassment, and content in the zero-tolerance categories described in our Community Guidelines is removed on first instance regardless of account standing.

Reviewers are measured on accuracy, never on volume of removals. Sampling audits re-review a random slice of decisions each cycle, and categories with elevated error rates are pulled back into human-only review until quality recovers.

Every enforcement notice tells you what was actioned, which policy applied, and how to appeal. Appeals are routed to a reviewer who did not make the original decision. If an appeal succeeds we restore the content, remove the strike from your record, and reverse any downstream effect the strike had on distribution or monetisation — a reversal that leaves the penalty in place is not a reversal.

We also correct at the category level. When an appeal reveals that a rule or classifier is misfiring for a whole class of content, we fix the rule and re-run the affected decisions rather than fixing only the case in front of us. To start or check an appeal, use iTweet AI — it verifies your identity, collects the case context and routes the appeal to the specialist queue.

We publish a transparency report twice a year covering: content actioned by policy category; the share detected proactively before any user report; appeals received, granted and denied; accounts actioned for coordinated inauthentic behaviour; legal removal requests by country with a compliance rate; government and law-enforcement requests for user information with the share where we produced data; and preservation requests received.

Numbers are reported as ranges when a raw count could deanonymise an individual, and we say so instead of rounding silently. Where a metric changes definition between reports, we restate the prior period so the trend is honest. Where we cannot publish something — because a statute forbids it — we state that a gap exists rather than presenting an incomplete figure as complete.

We require valid legal process. Requests for user information must come through the correct instrument for the requesting jurisdiction, be specific about the account and the data sought, and be proportionate to the investigation. We reject requests that are overbroad, that lack legal basis, or that ask for content data where the process supports only basic subscriber information. Where we are permitted to do so, we notify the affected user before disclosing anything, and we delay notice only where a statute or a court order requires it.

Emergency disclosures are limited to a good-faith belief of an imminent risk of death or serious physical harm, are logged, and are reviewed after the fact. Government demands to remove lawful content are treated as removal requests, assessed against local law and our own policies, and counted separately in our reporting. Full detail on formats, retention windows and international process is in our Information for law enforcement article.

Ranking on iTweet is a prediction, not an editorial verdict. Feed, Reels and Explore score candidate posts on signals such as who you follow and interact with, recency, dwell and completion, topic affinity, and negative signals like "not interested", mutes and reports. Negative signals are weighted heavily — telling us you do not want something is more informative than a passive scroll.

We apply demotions rather than removals to borderline content: clickbait, engagement bait, unoriginal re-uploads and repeatedly reported accounts get less distribution while remaining available to people who follow them. Sensitive categories are excluded from recommendation surfaces entirely. You can see and reset the interests we infer for you in Personalisation settings, and switching to a following-only feed disables recommendation ranking altogether.

Ads are labelled as ads, and every ad can be inspected: who paid for it, why you are seeing it, and which broad interest or context triggered it. We do not build advertising audiences from sensitive categories, we do not sell your personal information, and advertisers receive aggregated performance reporting rather than lists of individuals. Ads that make prohibited claims, imitate platform UI, or target minors with age-restricted products are rejected at review and removed if they slip through.

Transparency includes our own weak points. We publish the security posture we hold ourselves to: transport encryption everywhere, row-level access control on user data, device attestation on the Android app to block tampered clients, end-to-end encryption for one-to-one messages, and least-privilege access for internal tooling with logged administrative actions.

We welcome external scrutiny. Researchers and users who find a vulnerability or a systematic policy failure can report it through iTweet AI, which routes disclosures to the security team; we do not pursue good-faith researchers who follow responsible disclosure. If a breach affects your personal data we notify affected users and the relevant supervisory authority within the timelines the applicable law requires, including the Digital Personal Data Protection Act, 2023 framework in India, and we publish a post-incident summary once the immediate risk has passed.