TikTok’s Feed Filters Add Control—but Still No AI-Free Switch
TikTok now offers topic preferences and AI-assisted keyword filtering as established ways to influence recommendations. The central limitation remains unchanged: these controls can alter the mix of videos in a feed, but they do not give users a reliable switch for excluding every unwanted subject or all AI-generated content.
The picture has become clearer since Smart Keyword Filters and Manage Topics arrived in June 2025. Both features remain relevant, while a later control aimed specifically at AI-generated videos was still described as being tested in TikTok’s July 2026 update rather than as a universally available opt-out.
Two established controls solve different problems
Manage Topics changes the frequency of broad categories in the For You feed. A user can request more or less material about subjects such as sports, food and drinks, or lifestyle. The preference applies to recommendations from creators outside the accounts that person follows; it does not change the Following feed, profiles or inbox.
TikTok’s current recommendation documentation lists Manage Topics alongside signals such as “Not interested” feedback and also describes keyword filters for removing content containing selected words, variations and hashtags from named feeds. Its AI option can extend a filter to similar language, including synonyms.
The distinction matters because a topic preference is not a blocklist. Reducing travel or cooking changes the weight given to a broad category while leaving the system free to recommend some videos from it. A keyword filter is better suited to a specific name, phrase, product, event or recurring trend that a user wants to avoid.
Neither setting makes the feed deterministic. Recommendation systems continue to rank eligible videos using other signals, including viewing behavior and interactions. The controls add explicit preferences to that process, but TikTok still decides which individual videos appear and in what order.
The original rollout mixed released features with future plans
TikTok’s June 3, 2025 product announcement introduced Manage Topics for more than 10 categories and began rolling out Smart Keyword Filters, while noting that existing filters had been used to restrict content containing more than 200 million words globally. The same announcement described the ability to select individual AI-suggested terms, bulk entry and an increase from 100 to 200 keywords as changes planned for the following months.
That chronology creates an important status distinction. AI matching for related words formed part of the rollout, but several interface and capacity changes were presented in the future tense. The launch post alone therefore does not establish that every planned option subsequently reached every account.
The “smart” label also referred to how a keyword filter expands a user’s instruction. It did not mean TikTok had created a general filter for all content produced with artificial intelligence. A user who blocked a term such as “renovation” could also limit semantically related wording, but that mechanism would not identify every synthetic video unrelated to the selected term.
The AI-generated-content slider is still a test
A separate control for synthetic media appeared later inside Manage Topics. TikTok’s July 10, 2026 AI-content update continued to describe Manage Topics as a feature being tested for choosing how much AI-generated content users see, rather than presenting it as a fully released exclusion tool.
This is not merely a new name for Smart Keyword Filters. Keyword matching uses AI to find language related to a term entered by the user. The newer setting targets the volume of videos identified as AI-generated or substantially AI-edited, regardless of the words they contain.
The wording “how much” also defines the control’s practical boundary. It is a recommendation preference for increasing or decreasing a class of content, not a promise to remove that class entirely. Users who do not see the setting should not assume that the ordinary topic or keyword controls provide the same function.
What the controls change for creators and businesses
These settings make recommendation exposure somewhat more dependent on explicit audience choices. A viewer can reduce an entire subject area, filter a campaign phrase or brand name, or potentially request fewer AI-generated videos where the experimental control is available. None of those actions means a video violated TikTok’s rules; personalization and content moderation are separate systems.
The difference between the For You and Following feeds is particularly relevant to distribution. Manage Topics primarily affects discovery recommendations from outside the accounts a user follows, while keyword filtering can operate across multiple feeds. A creator’s existing follower and a new viewer may therefore encounter the same material through different ranking paths.
Businesses should not treat topic preferences as audience-exclusion controls comparable to advertising settings. They are user-side signals, not targeting parameters visible to a brand. A campaign may receive less exposure in one person’s recommendations without providing the advertiser or creator with a direct explanation of which preference produced that result.
The emerging AI-generated-content control adds another uncertainty. Brands using synthetic media cannot infer that every viewer can reduce it, while brands avoiding such media cannot assume their posts will benefit when someone lowers the slider. Classification, labeling and recommendation ranking remain platform processes rather than guarantees available to publishers.
TikTok’s feed controls are therefore best understood as adjustable inputs. They give users more agency over recurring topics and identifiable language, but the platform still assembles the final feed—and its dedicated control for AI-generated videos has not yet become a confirmed universal off switch.
Also read:
Subscribe to our newsletter
Get the latest Web3, AI, and crypto news delivered straight to your inbox.