Reddit has published its transparency report for January to June 2026, and the intellectual-property sections are unusually blunt about how much platform enforcement fails. On the copyright side the platform received 66,221 takedown notices, 4% fewer than in the preceding six months, but those notices identified 465,803 individual pieces of content, up 9%. Reddit removed 258,855 of them — a 55% content actionability rate — and actioned 46,470 notices in whole or in part, a 70% notice actionability rate. The other side of that ledger is the part worth reading twice: Reddit rejected 20,143 copyright notices outright as invalid, 30% of everything it received, and declined to touch 109,706 reported items. Among the grounds it lists for invalidity are missing required elements, notices that are really privacy or harassment complaints in copyright clothing, and content that uses the work in a way copyright law permits. TorrentFreak, which reported the figures on 15 November 2026, noted the parallel collapse in escalation: 1,266 accounts banned for repeat copyright violations, down 21%, and just 118 subreddits banned, down 79% from 563 in the second half of 2025.
The trademark numbers are starker still. Reddit received 1,084 trademark takedown notices in the half, up 24%, covering 4,026 pieces of content. It actioned 110 of those notices — a 10% actionability rate — and removed 346 items, 9% of what was reported. It rejected 978 notices, 90% of the total, and declined to remove 3,616 items. Twenty accounts and five subreddits were banned for repeat trademark violations in six months. Reddit’s own stated reasons for rejection read like a checklist of drafting failures: a notice that does not identify a valid, registered trademark; a notice aimed at content making fair use of the mark; a notice that confuses trademark with copyright, impersonation, privacy or harassment; and duplicate URLs already removed under an earlier notice. Reddit attributes the rising volume to better reporting from legitimate rights holders and less low-quality bulk reporting — which means 90% is the improved figure.
A registered mark, named in the notice, is the threshold
The first rejection ground is the one most within a brand owner’s control. A platform trademark notice that cannot point to a valid registration is not a weak notice; at Reddit it is an invalid one, and the content stays up. That makes two portfolio questions operational rather than theoretical. Is the mark actually registered in a jurisdiction the platform will recognise, and is it registered in the classes that cover the goods being counterfeited or the services being impersonated? Enforcement teams routinely discover the gap at the worst possible moment — a counterfeit storefront is live, the notice goes out on a common-law or pending-application basis, and it bounces. If an upcoming enforcement push depends on marks that are not yet registered, the filing is the enforcement step, not a parallel administrative chore: see our trademark filing service for what that timeline actually looks like.
Communities that merely name your brand are not a takedown problem
Reddit highlighted a notice from a brand in Cyprus targeting two subreddits whose titles contained the brand’s name. Both were unaffiliated communities sharing reviews and warning users about possible scams, and Reddit declined to remove either, stating that using a brand name to identify the subject of user discussion is permitted under its trademark policy. Compare the outcome in the case Reddit also flagged, where a fashion subreddit was posting links to buy counterfeit clothing: the content came down, the moderators were warned, the links went back up, and the subreddit was banned. The distinction is not the volume of complaint but the nature of the use. Notices aimed at commercial counterfeit links work; notices aimed at criticism, reviews or discussion fail and consume the credibility of every later notice from the same sender. The practical response is to triage findings by use type before anything is sent, which is also the argument for structured trademark search and monitoring rather than keyword alerts that cannot tell a counterfeit listing from a complaint thread.
The repeat-infringer lever is weakening, and Cox is why
Strategies built on getting accounts and communities terminated are running into a changed legal background. In Cox Communications, Inc. v. Sony Music Entertainment, No. 24-171, decided on 25 March 2026 and reported at 607 U.S. 583 with the opinion written by Justice Thomas, the Supreme Court reversed the Fourth Circuit. As TorrentFreak reported at the time, the 7-2 decision held that contributory liability requires proof the provider intended its service to be used for infringement, and that merely providing a service to the public while knowing some users will infringe is not enough. The pressure that once pushed intermediaries toward aggressive termination of repeat infringers is therefore lighter, and Reddit’s 79% drop in subreddit bans sits in that context whatever its internal cause. The first transparency report covering a full post-Cox period is still to come. For now, treat termination as an outcome you may occasionally earn on a documented pattern, not as a remedy you can plan a programme around.
Counter-notices now put a litigation clock on your desk
Reddit received 688 counter-notices in the half, 60% more than in the previous period, of which 221 were valid and restored 599 pieces of content. A further 40 withdrawal requests restored 764 items — rights holders themselves pulling back. Under 17 U.S.C. §512(g)(2)(C), a service provider that receives a valid counter-notice must restore the material not less than 10 and not more than 14 business days later, unless the party who sent the original notice first tells the provider it has filed an action seeking a court order against the subscriber. That is the part teams forget when they send notices in bulk. A counter-notice converts a cheap administrative action into a two-week decision about whether to sue, and the default answer — do nothing — puts the content back with a record that your claim went untested.
What to change
Four adjustments follow directly from the data. Confirm registration and class coverage before a trademark notice is drafted, because the platform will not infer either. Screen candidate targets by use type and send only against commercial counterfeiting and source confusion, not commentary. De-duplicate URLs against earlier notices, since Reddit names duplicates as an independent rejection ground. And size the programme to the roughly 45% of reported copyright content that survives the process, rather than to the volume you report, so that the gap is a planning assumption instead of a quarterly surprise.
