Brand Safety on X: Adjacency Controls and Reply Moderation

Brand safety on X is the question most likely to come up when a marketing leader proposes spending on the platform, and it deserves a precise answer. Advertisers pulled back from X in 2023 and 2024 over concerns about the content their ads appeared beside. Since then X has rebuilt its control set, extended third-party measurement and, in March 2026, published a Brand Suitability Playbook that positions its Grok AI model as part of the classification system. X’s global head of advertising said at CES in January 2026 that 97 of the top 100 advertisers had returned, according to trade press reports. This guide explains what the controls do, where their limits are, and how to run reply moderation and crisis response as an operating process. It is part of the Kurums Social Media Marketing Hub.

What does brand safety on X mean for advertisers and organic accounts?

Brand safety on X covers two distinct exposures: the content that appears next to your paid ads, and the replies and quotes that appear around your own posts. Different tools address each, and most incidents come from treating them as one problem.

The industry usually separates brand safety from brand suitability. Safety is the floor: content almost no advertiser wants to be near, such as illegal material, graphic violence or hate speech. Suitability is the judgment above that floor, which varies by brand. A news publisher may accept adjacency to war coverage; a toy company will not. X’s controls map to both ideas. Its rules remove content that violates policy for everyone, and its advertiser settings let each brand draw its own line above that.

Organic exposure works differently. When the brand account posts, anyone can reply by default, and those replies sit directly under your message. Spam, abuse, competitor mentions and off-topic arguments are all possible. No ad setting touches them. They are managed through conversation controls and by the people who watch the account.

A complete policy therefore assigns two owners. The media buyer or agency owns ad adjacency settings and verification reports. The social or community manager owns reply moderation. Both report into one escalation path when something goes wrong.

How do adjacency controls work?

Adjacency controls let an advertiser name keywords and accounts whose posts should never appear directly above or below its ads. X allows up to 4,000 excluded keywords and up to 2,000 excluded accounts per advertiser.

X defines ad adjacency narrowly: one organic post above and one organic post below the ad placement. Keyword exclusions are case-insensitive, cover hashtags and plurals, and work in all languages, according to X’s brand safety documentation. X states that the system achieves over 99% efficacy. The controls apply across the Home timeline, in both the For You and Following tabs, and to profiles, search results, post replies and the media viewer.

Source: X’s brand safety controls page, checked October 2026.

One distinction trips up many teams. Campaign-level keyword targeting exclusions are a different feature from adjacency controls. X’s keyword targeting documentation says explicitly that excluding a keyword in targeting prevents ads from showing in search results for that term, and does not prevent ads from appearing near or adjacent to that word in the timeline, on profiles or in post details. If the goal is to stay away from a topic, the term belongs in the adjacency list. Putting it only in campaign targeting leaves the exposure open.

Adjacency is also a definition with edges. It covers the posts immediately around the ad. It does not describe the account whose profile the ad appears on in every case, the wider conversation thread, or what a user saw three posts earlier. That is why the other layers exist.

What do sensitivity settings change?

The sensitivity setting is an account-wide dial with two positions. Standard, the default, shows ads beside a reduced range of content. Limited applies stricter filtering for brands with low tolerance, and X warns it may lower reach and raise costs.

Both settings exclude the same categories, according to X: illegal or regulated products, explicit adult content, graphic violence, sensitive topics such as terrorism, hate and political content, and excessive profanity. The difference is in degree. X describes Limited as also filtering out educational or newsworthy discussion of those topics, which Standard may allow. Content that breaks X’s rules is excluded under either setting.

Choosing between them is a business decision with a price. Limited reduces the inventory your ads can win, which tends to raise the cost of each impression and can slow delivery for narrow B2B audiences. A regulated financial firm or a healthcare brand will usually accept that cost. A developer tools company advertising to engineers may find Standard with a well-built exclusion list gives adequate protection at lower cost. Record the choice and the reasoning in the media plan so it survives a change of agency or staff.

X’s March 2026 playbook, as reported by trade publication PPC Land, lists seven advertiser tools in total: placement filters, adjacency controls, sensitivity settings, vertical video pre-bid, Trend Genius, Amplify video and always-on suitability protections. It describes Grok as providing contextual understanding of posts and text recognition across formats for classification. That is X’s own description of its system. Independent measurement, covered below, is how an advertiser checks it.

How should you build and maintain blocklists?

Build the keyword list from four sources: universal unsafe terms, industry-specific risks, your own brand’s sensitive history and current events. Build the author list from accounts that repeatedly produce content you would not want beside your logo. Review both monthly.

A list of 4,000 keywords sounds large and fills quickly. Structure helps.

• Core list. Violence, slurs, adult content, tragedy and disaster vocabulary. Many agencies and verification vendors maintain starting lists.
• Category list. Terms tied to risks in your sector: data breach and fraud vocabulary for a bank, recall and injury terms for a manufacturer, layoffs for an HR software vendor.
• Brand list. Names of past controversies, litigated matters, former executives or nicknames critics use for the company.
• Event list. Temporary additions during elections, conflicts, disasters or a breaking story in your industry. Date each entry so it can be removed.

Over-blocking has a real cost. Excluding a common word such as “crash” keeps a carmaker away from accident reports and also away from discussion of software crashes, market commentary and sports. Each broad term shrinks available inventory. Prefer specific phrases, and test additions by searching the term on X to see what kind of posts it actually matches.

For the author list, start with accounts flagged in your verification reports and in your own monitoring, then add accounts known for content outside your suitability line. Keep the list in a shared document with the date and reason for each entry. An unexplained blocklist becomes impossible to maintain within a year.

What does third-party verification cover on X?

Integral Ad Science and DoubleVerify independently measure whether ads on X appeared beside safe and suitable content. Post-bid measurement is offered in six markets and two languages, and IAS additionally offers pre-bid filtering for vertical video.

X’s documentation lists the supported markets as the United States, United Kingdom, Australia, Canada, New Zealand and Japan, covering English and Japanese content. IAS offers both pre-bid and post-bid products, and DoubleVerify provides adjacency measurement. Trade press reports that IAS extended its measurement from the timeline to profile placements in January 2026, and that its pre-bid product, which screens inventory before an ad is served, applies to vertical video in the US only.

Read those headline scores for what they are: platform-wide averages published by X, drawn from partner data. They are encouraging and they do not describe your campaign. Ask your verification vendor for a campaign-level report with the incidents listed, and compare it against your own suitability line.

Note the gaps as well. A campaign running in German, Spanish or Turkish falls outside the listed language coverage, so independent measurement may be unavailable and you will rely on X’s own controls. Post-bid measurement reports what happened after the impression was served and paid for; it informs blocklist updates and make-good conversations and cannot prevent the placement. Verification is a paid service, usually priced per thousand impressions measured, so include it in the media budget. The Kurums X Ads guide for business covers how the rest of that budget is structured.

How do you moderate replies on your own posts?

Use reply settings to decide who may respond, hide replies that are abusive or off-topic, and mute or block persistent offenders. All of these are available to the author of a post at no cost and can be applied after the post is live.

X offers four reply settings for each post.

X’s help center confirms that the setting can be changed after publishing through the post menu, and that people will see that replies have been limited. That visibility matters. Restricting replies on a routine product post looks defensive, and critics will simply quote the post instead, where you have no control at all. Reserve restrictions for posts where an open thread would cause harm.

Hiding a reply moves it behind a separate view; it is not deleted and people can still choose to see it. Hide spam, scams, slurs and personal attacks. Leave legitimate criticism visible and answer it. A thread where a company responds calmly to a hard question does more for trust than a thread scrubbed clean, and hidden criticism tends to resurface as a screenshot.

For accounts with high reply volume, automation is now documented. X’s developer changelog for October 1, 2026 describes an updated guide for real-time reply moderation: the Activity API delivers each new reply as an event, the xAI API classifies it using structured outputs, and replies that breach the brand’s rules are hidden through the Hide Replies endpoint. Only the author’s authenticated account can hide replies in its own conversations. API usage is billed per event and per read, as explained in the Kurums guide to X API pricing, and any automated classifier should be reviewed by a person for false positives before it runs unattended.

What does a reply moderation workflow look like?

A workable workflow classifies every reply into a small number of categories, assigns a fixed action and response time to each, and names who decides the unclear cases. It should fit on one page.

The matrix below is a template to adapt to your own policies and legal advice.

Three practices make the matrix work. Keep a log of hidden replies and blocked accounts with a short reason, so decisions are consistent across team members and defensible if challenged. Write response templates for the common cases and require them to be adapted, since identical replies read as automated. And define coverage hours openly: if nobody monitors at weekends, say so in the bio and schedule sensitive posts for weekday mornings.

Impersonation deserves a line of its own. Fake support accounts that reply to your customers with phishing links are a recurring problem on X. Report them, warn customers in a pinned post during active waves, and consider whether the gold checkmark and impersonation monitoring included in X’s business tier are justified. The Kurums guide to X Premium for business sets out what those tiers cost.

What should a crisis protocol on X include?

A crisis protocol names the triggers, the decision-maker, the first three actions and the criteria for returning to normal. On X the first actions are usually to pause scheduled posts and ads, assess the facts and publish a holding statement.

Crises on X fall into three types, and each needs a different first move.

1. Adjacency incident. A screenshot circulates of your ad beside objectionable content. Pause the affected campaigns, capture evidence, add the terms and authors to exclusion lists, request a report from your verification vendor and contact your X representative. Decide whether to comment publicly based on how widely the image has spread.
2. Brand-originated incident. Something the company did or posted has drawn criticism. Pause all scheduled content immediately, since a cheerful product post during a backlash compounds the damage. Publish a short factual statement, set reply controls only if abuse is severe and move detailed responses to an owned page.
3. External event. A tragedy, disaster or major news story makes normal marketing inappropriate. Pause scheduled posts and reach campaigns, add event terms to the blocklist and resume when the news cycle has moved on.

Write the protocol before you need it. It should list who can pause ads and scheduled posts, with at least two people holding the necessary logins; who approves public statements; the internal channel where decisions are recorded; and contact details for the agency, the verification vendor and X. Run a 30-minute tabletop exercise twice a year using a realistic scenario.

Set return criteria in advance too. A common rule is to resume normal posting when mention volume has fallen back toward baseline for 48 hours and no new facts are expected. Afterward, hold a short review: what was seen first and by whom, how long each decision took, and which control would have prevented or limited the incident. Feed the answers back into the blocklists and the workflow.

How do you report on brand safety to leadership?

Report four numbers each month: the share of measured impressions classified as safe and suitable, the number of adjacency incidents, the volume of replies hidden or reported by category, and median response time to customer issues. Add a short note on changes made.

Leadership does not need the blocklist. It needs evidence that the exposure is understood and managed, and a clear statement of residual risk. A one-page monthly summary with the four measures, a trend line and a list of actions taken is sufficient. Where a campaign ran outside third-party coverage, say so plainly.

Connect the report to the wider measurement framework so that safety is weighed against performance. Stricter settings cost reach, and the trade-off should be visible. The Kurums guide to social media analytics metrics explains how to present channel results, and the X marketing strategy guide for B2B covers the organic plan that the moderation workflow protects.

Finally, revisit the decision to be on the platform at least once a year, using your own data. If verified suitability is high, incidents are rare and the channel produces pipeline, the case is made. If the team spends more hours on moderation than the channel returns, that is a finding as well. A broader framework for weighing channels is in the Kurums guide on building a social media marketing strategy.

Frequently Asked Questions