Brand suitability has been a publishing concern for decades. The formats changed from print to banner to programmatic, but the core problem stayed constant: advertisers do not want their brand sitting next to content that contradicts their values or unsettles their customers. In AI chat, that problem sharpens considerably.
A static article page has a fixed context. The brand safety check happens once, at crawl time or in the SSP's classification layer, and the result is cached. A conversation is different. Each turn generates new context. The assistant's response to "what are the best budget mattresses" is a perfectly comfortable environment for a home goods advertiser. The next turn in that same session, if the user pivots to asking about opioid addiction resources, is not.
This is why Velocity treats brand suitability as a per-turn decision rather than a per-session label. The scoring pipeline that determines what context a turn carries is covered in the contextual relevance technical overview.
The Two Layers of Safety Enforcement
Velocity's content safety system works in two distinct layers that operate at different points in the pipeline.
The first layer is platform-level exclusions. These are categories that we suppress globally regardless of publisher configuration. The list covers content involving minors in unsafe contexts, content that facilitates self-harm, and a handful of other categories defined in our publisher terms. No ad is inserted into turns where the assistant's response falls into these categories. This layer is not configurable by publishers, advertisers, or any other party.
The second layer is publisher-configured category exclusions. Each publisher can specify up to 20 category exclusion rules in their Velocity dashboard under Ad Preferences. These are applied using the same taxonomy structure as our relevance scoring, so if you exclude the Gambling and Firearms categories at the top level, all subcategories within those are also excluded without needing to list them individually.
How Publisher Exclusions Work in Practice
When you configure exclusion categories, you are doing two things simultaneously. You are telling Velocity not to serve ads from advertisers in those categories. You are also telling the safety layer to suppress any ad insertion into turns where the detected context topic falls within those categories, regardless of what the advertiser is.
The second part is the one publishers sometimes do not anticipate. Consider a publisher running a health and wellness AI assistant who excludes Alcohol from their ad categories. If a user asks about the caloric content of beer, that turn will be flagged under the Alcohol topic, and no ad will be inserted, even if the matched advertiser would have been a fitness brand. The exclusion applies to the conversation context, not just the advertiser category.
This is intentional. A fitness brand appearing immediately after a detailed discussion of alcohol consumption is a brand suitability problem for the advertiser regardless of whether the advertiser's own category is clean. Protecting the context protects both the publisher's reader relationship and the advertiser's brand placement quality.
Sensitivity Categories and When to Use Them
Beyond outright exclusions, the Velocity publisher dashboard has a second tier of configuration called Sensitivity Flags. These do not exclude categories entirely but reduce the relevance score threshold required for a match, making it harder for an ad to pass the scoring gate when the conversation involves those topics.
Sensitivity Flags are currently available for: Mental Health, Personal Finance, Medical Conditions, Legal Advice, and Relationship Topics. If you enable the Mental Health sensitivity flag, the relevance score threshold for a match in a turn where that topic appears rises from the default 0.50 to 0.75. In practice, this means only high-confidence, strongly on-topic matches survive in sensitive conversation contexts.
A publisher running a general-purpose AI assistant, where conversations genuinely span everything from recipe suggestions to grief support, will typically benefit from enabling several sensitivity flags even if they are not excluding those categories outright. The goal is not to zero out revenue in those turns; it is to ensure that when an ad does appear in a sensitive context, it fits well enough that no reader would find the placement jarring.
What Brand Suitability Does Not Cover
It is worth being direct about the limits of what this system does. Brand suitability enforcement in Velocity is about the semantic context of the conversation turn. It is not a real-time content moderation system for user-generated content. If a user sends abusive or harmful messages to your assistant, Velocity's safety layer will typically suppress ads in those turns based on context classification, but that is a side effect of safety logic, not the primary purpose.
Publishers are still responsible for their own content moderation policies for the assistant product itself. Velocity does not and should not be treated as a substitute for your own trust and safety infrastructure around what users can say and what the assistant will respond to.
We are also not saying that our category taxonomy is comprehensive. There are edge cases where a conversation touches a sensitive area that does not yet map cleanly to an exclusion category in the Velocity taxonomy. We update the taxonomy quarterly, but gaps exist. If you have publisher use cases where certain topic areas matter and you cannot find a clean match in the current exclusion category list, contact us. We have handled custom exclusion logic for publishers whose assistant products operate in specialized domains like legal or clinical contexts.
Advertiser Suitability From the Other Side
Advertisers using Velocity also have their own brand suitability controls. They can specify topic adjacency exclusions on the buy side, which means an advertiser can decline to serve their ads in sessions where the conversation has touched certain topics, even if the specific turn being scored is clean.
This is a relatively uncommon configuration, used by advertisers in categories like financial services or healthcare where brand positioning around certain content is regulated or carefully managed at the company level. When an advertiser applies session-level exclusions, it reduces their fill footprint, which publishers may notice as slightly lower fill rates for sessions involving those topic areas.
Configuring Your Safety Setup
The practical starting point for any publisher joining Velocity is to spend time in the Ad Preferences section of the dashboard before going live. The default state, with no exclusions and no sensitivity flags configured, will work technically. But it reflects a generic configuration, not one tuned to your specific assistant product and your readers' expectations.
Think about the subject matter your assistant handles most commonly and where your readers would be most sensitive to an ad appearing. Think about whether there are entire advertiser categories that would feel misaligned with your product's identity. Those judgments belong to you as the publisher, and the configuration tools exist to express them precisely.
The brand suitability layer is there to enforce what you specify. The specification itself is where the meaningful work happens.
