AI brand safety in B2B paid media: the guardrail policy your team can copy
Most AI brand safety controls in B2B paid media are now settings and permissions, several default to on, and this guide maps them and gives a one-page policy a team can copy.

AI brand safety in B2B paid media means controlling three things once AI writes, places and changes your ads: the words the platforms generate in your name, the places and audiences they expand into, and the customer data your team feeds to models. As of November 2026, most of those controls are settings and permissions, and several default to on.
You'll know the moment. Someone in sales forwards a screenshot of your Google search ad with a headline nobody on the team wrote, promising something your product doesn't quite do. Legal asks who approved it. The honest answer is that a setting did, on a campaign built last spring, and nobody remembers switching it on because nobody did.
This guide is for the marketing leader who has to sign off on how far AI goes in the ad accounts, especially in regulated or enterprise markets where one bad claim costs more than a quarter's savings. It covers where each platform's AI writes and expands, the claim and disclosure rules in force, a data rule for CRM exports, and a one-page policy you can paste into your team's wiki. The four decisions a person must keep signing are in our guide to where humans stay in the loop on ad spend.
AI brand safety covers the words, the places and the data
AI brand safety used to mean one thing in paid media: keeping your ad away from content you'd rather not sit beside. The platforms now also write copy, assemble creative and widen audiences on your behalf, so the risk has moved inside the ad itself.
We'd split the work into three surfaces, because each one has a different owner and a different control:
- Words. Headlines, descriptions and images the platform generates or alters from your inputs. The control is a campaign or ad setting, and the owner is whoever owns the brand.
- Places. The sites, apps, videos and audiences the platform expands into beyond what you chose. The control is mostly at the account level, and the owner is the head of paid media.
- Data. The CRM exports, call notes and customer lists your team gives to AI tools and ad platforms. The control is a written rule plus the right software plan, and the owner is whoever owns privacy.
Claims cut across all three. A claim your team wrote and a claim the platform generated carry the same legal weight, and the regulator won't ask which one came from a model.

Guardrails settled early let a team hand AI more of the week's work without a nervous review of every ad, and Scalewell supports paid and demand gen teams through that shift across strategy, planning, implementation and operations.
The platforms switch some AI writing on by default
Google Ads turns on AI-generated text for new Search campaigns unless you opt out, and Meta and LinkedIn offer generation at the ad level with a review step.
Google's AI Max help page says that "New Search campaigns have the AI Max feature turned on by default," and that inside AI Max, "The 'Text customization' setting will be enabled by default" and "The 'Final URL expansion' setting will be enabled by default." Text customization is the feature formerly known as automatically created assets: Google writes extra headlines and descriptions from your landing page and existing ads. Final URL expansion lets Google send the click to a different page on your site from the one you picked. Both are switched in the campaign's AI Max section under Asset optimization, and turning off text customization switches off final URL expansion with it. Our guide to AI Max for Google Ads covers whether to keep it on for performance reasons; this section is only about control.
If you keep text customization on, use Google's text guidelines. The help page describes two controls, term exclusions and messaging restrictions, with "up to 25 term exclusions per campaign" and "up to 40 restrictions per campaign," and warns that existing assets that break new guidelines "will stop serving and be marked as removed." Google's February 26, 2026 announcement gives the kind of plain-language rule it accepts, such as "don't imply our products are cheap." The feature is still labeled an "experimental beta," and it governs text only, so it says nothing about generated images.
Meta handles generation per ad. Its Marketing API documentation for Advantage+ creative lists each enhancement with an enroll status of OPT_IN or OPT_OUT, including AI-generated ones such as image background generation (labeled Generate backgrounds in Ads Manager) and image uncrop (Expand image), plus text optimizations that rework the text you supplied, and marks one, adapt to placement, as "Default is opt-in." The API sets these on each ad creative, so check them on every new ad rather than assuming the choices you made on the last one carried over.
LinkedIn's generation sits inside the ad builder. Its help page for single image ads describes a Draft with AI button and AI ad variants that use your intro text and headline to create three variations in the same ad set, and says "Once generated, you can review and edit before saving." That review step is the guardrail, and it works only if someone reads the variants as drafts.
| Platform | AI feature that writes or alters ads | Default as of November 2026 | Where to control it |
|---|---|---|---|
| Google Ads | AI Max text customization and final URL expansion | On for new Search campaigns | Campaign settings, AI Max, Asset optimization |
| Google Ads | Text guidelines (term exclusions, messaging restrictions) | Off until you add them | AI Max Asset optimization panel; Performance Max brand guidelines |
| Meta | Advantage+ creative enhancements, including generated backgrounds and reworked text | Set per ad creative; adapt to placement defaults to opt-in | On each ad, or creative_features_spec in the API |
| Draft with AI, AI ad variants, Brand Kit | Used only when someone clicks them | Create ad, inside the ad set in Campaign Manager |
Placement and expansion controls belong at the account level
Placement and audience controls protect a B2B brand best when they're set once at the account level, because campaign-level settings get forgotten every time someone builds a new campaign.
In Google Ads, the content suitability center sits under Content suitability in the Tools menu. Google's help page says you can exclude "up to 1,000 content keywords at the account level" and up to 65,000 placements across YouTube, the Display Network, Google Video Partners and the Search Partner Network. Brand exclusions are a separate, campaign-level list that, per Google's brand exclusions page, applies in Performance Max "to only Search, Shopping, and YouTube search inventory."
Meta's inventory filter offers Expanded, Moderate and Limited levels, and its placement targeting reference says "When a filter is applied at the ad account level, only more restrictive options will be available at the campaign level." Publisher block lists sit alongside it for Audience Network.
LinkedIn has two settings worth checking on every ad set. Its audience expansion page says "Audience expansion is automatically enabled on ad sets where it's available," with the checkbox in the Audience section of the Set up ad set page. The LinkedIn Audience Network checkbox sits under Placements, and LinkedIn's help page says it "is automatically enabled for new single image, carousel, document, and video ad sets." Its Audience Network FAQ says you can use "category-level blocking or a block list of specific websites and apps."
Expansion matters more in B2B than in consumer advertising, because your audience is a list of accounts and buying committees rather than a crowd. An ABM program built on, say, 2,000 named accounts loses its point when the platform decides people who look a bit like them will do. We'd clear audience expansion on every account-list campaign, and the reasoning for that sits in our guide to AI for account-based marketing.
AI marketing compliance rests on claims you can prove
AI marketing compliance in the US comes down mostly to an old rule: an advertising claim needs evidence behind it when it runs, whoever or whatever wrote it. The Federal Trade Commission has applied that rule to AI companies and to AI-generated content, and B2B software firms are squarely in range.
The clearest recent case is Workado. In its August 28, 2025 final order, the FTC barred the company from advertising the accuracy of its AI content detector "unless it has competent and reliable evidence showing those products are as accurate as claimed," and required it to keep that evidence. If your ads say, to take a hypothetical, that your platform cuts close time by 40% or catches 98% of fraud, that's the standard your claim will be held to, and a generated headline that rounds a case-study figure up is your claim too.
Fake social proof is the second trap. The FTC's August 14, 2024 rule on fake reviews covers "reviews and testimonials that misrepresent that they are by someone who does not exist, such as AI-generated fake reviews," and lets the agency seek civil penalties against knowing violators. A generated "VP of Finance" quoting your ROI in a LinkedIn ad fits that description uncomfortably well.
The tribunal that ruled against Air Canada on February 14, 2024 put the principle plainly. The airline had argued, in effect, that its website chatbot was a separate legal entity responsible for its own actions, and the tribunal held that Air Canada is responsible for all the information on its website, chatbot included, as Cassels' note on Moffatt v Air Canada summarizes. It's a Canadian small-claims decision that binds nobody in the US, but expect the same argument about your ads.
The practical tool is a claims register: a short, shared list of every number, comparison and outcome your ads are allowed to say, each with the source document and the date it was last checked. The platforms' text guidelines can then exclude anything not on the list. AI marketing ethics adds one question the law doesn't always ask: would your buyer feel misled if they knew how this ad was made?
Disclosure rules in force as of November 2026
Disclosure of AI in ads is now required in some markets for some content, mostly synthetic people and realistic generated imagery, while AI-written text largely sits outside the rules. Here's where it stands, kept short because our piece on AI-generated ads for B2B covers what each platform labels and what the research says about labels and performance.
| Rule | What it covers | In force |
|---|---|---|
| New York General Business Law ยง 396-b | Ads containing a synthetic performer the advertiser knows about must disclose it; $1,000 for a first violation and $5,000 after, per the bill text | June 9, 2026 |
| EU AI Act transparency rules | Generated content must be identifiable and deep fakes clearly labeled, per the European Commission | August 2026 |
| Google Ads AI label setting | Google says rules in the EU, India and New York require labels on certain AI-generated or edited assets, and that its label "doesn't guarantee compliance," per its policy update; its Ads Editor help page now also lists California, South Korea and Vietnam for on-ad labels | July 2026 |
| IAB AI Transparency and Disclosure Framework, version 2 | Industry self-regulation; disclose when AI "materially affects authenticity, identity or representation," and text or copy should not automatically require labels, per the IAB | August 18, 2026 |
The Commission also notes that the AI Omnibus entered into force on July 27, 2026 and pushed some high-risk rules to December 2027, which is why you'll read conflicting dates. The transparency rules weren't among the delayed ones on the Commission's page. For a B2B team the working default is simple: real people with signed releases for any face that stands for a customer or employee, labels on realistic generated imagery in markets that require them, and counsel's sign-off before you run synthetic people anywhere.
CRM data needs a rule before anyone pastes an export into a chatbot
The fastest brand safety failure in a B2B team has nothing to do with ads: someone pastes a CRM export of named contacts and deal notes into a personal AI account to "find patterns." Whether that data can be used to train a model depends on the plan, and personal plans and company plans now differ.
Anthropic's consumer terms update applies to "Claude Free, Pro, and Max plans," lets those users choose whether their chats train models, and extends retention "to five years, if you allow us to use your data for model training." The same update excludes commercial use, and Anthropic's privacy center says that "By default, we will not use your inputs or outputs from our commercial products" for training. OpenAI's API data documentation says "data sent to the OpenAI API is not used to train or improve OpenAI models," while abuse monitoring logs are "retained for up to 30 days" by default.
Ad platforms set their own terms on the other side. Google's Customer Match policy says "you may only upload customer information that you collected in the first-party context," requires your privacy policy to disclose sharing with third parties, and requires consent where law or Google policy demands it.
So the data rule we'd write is short. CRM data goes only into company-approved AI tools on commercial terms, with the fewest fields the job needs. Customer lists go to ad platforms only through their official upload routes. And nobody, including the CMO, uses a personal AI account for anything with a customer's name in it.
Copy this one-page AI guardrail policy into your wiki
The policy below is meant to be edited: replace the bracketed roles with names.
AI GUARDRAIL POLICY FOR PAID MEDIA
Owner: [VP Marketing] Reviewed: [quarterly, first week] Version: [1.0]
1. SCOPE
Covers every AI feature that writes, alters, places or targets our ads,
and every AI tool that touches customer data. Applies to staff, agencies
and contractors.
2. WORDS (owner: [brand owner])
2.1 Google Ads: switch AI Max text customization and final URL
expansion OFF on every new Search campaign. Turning either on needs
[brand owner] approval and text guidelines set before launch.
2.2 Meta: Advantage+ creative enhancements that generate images,
backgrounds or text are OPT_OUT unless approved per campaign.
Check each new ad before it launches.
2.3 LinkedIn: Draft with AI and AI ad variants are drafts. Nobody saves
a generated variant without reading every field.
2.4 Every new headline, description and image is reviewed by
[named reviewer] before launch. Generated assets are logged.
3. CLAIMS (owner: [compliance owner])
3.1 Ads may only state numbers, comparisons and outcomes listed in the
claims register, with source and date checked in the last 90 days.
3.2 No testimonials, quotes or reviews from people who do not exist.
3.3 Regulated categories (financial, health, employment, housing,
credit): [compliance owner] signs every ad before it runs.
4. PLACES (owner: [head of paid media])
4.1 Account-level content suitability and inventory filters are set
and documented. Campaign-level settings may only be stricter.
4.2 Placement block lists and brand exclusions are reviewed monthly
from the placement reports.
4.3 LinkedIn audience expansion is OFF on all account-list (ABM)
campaigns. LinkedIn Audience Network needs a block list before use.
5. DISCLOSURE (owner: [compliance owner])
5.1 Real people with signed releases for any face representing a
customer, employee or expert.
5.2 Realistic generated imagery or synthetic people carry the label
required in each target market. Counsel signs off before any
synthetic person runs.
6. DATA (owner: [privacy owner])
6.1 Customer data goes only into approved AI tools on commercial terms:
[list approved tools and plans].
6.2 No customer names, emails or deal notes in personal AI accounts.
6.3 Customer lists reach ad platforms only through official upload
routes, from first-party data our privacy policy covers.
7. AGENTS AND ACCESS (owner: [head of paid media])
7.1 AI agents get read-only access by default. Write access is granted
per tool, in writing, with the decisions in section 8 kept human.
7.2 Change history is exported and kept beyond the platform's window.
8. HUMAN SIGN-OFF
Budget ceilings, launches, placement lists and regulated claims are
signed by a named person. See: humans-in-the-loop rule.
9. INCIDENTS
Anyone who sees an ad that breaks this policy pauses it and tells
[head of paid media] the same day. We log what happened and which
rule or setting would have stopped it.
The settings in sections 2 and 4 are a checklist you can work through in Google Ads, Meta and LinkedIn. The claims register in section 3 is the slow part, because someone has to find the source behind every number your ads have used for the last year, and some of them won't have one.
Guardrails fail when nobody owns the settings
AI guardrails rarely fail because the policy was wrong. They fail because the settings drift, the policy grows until it's ignored, or the controls block so much that the account stops working.
The settings drift. Google's AI Max default applies to every new Search campaign, and Meta's enhancements are set on each ad creative, so a policy set once goes out of date as soon as someone builds a new campaign without checking. A quarterly settings audit across all three platforms is the minimum; a monthly one is better if agencies build campaigns for you.
The policy grows. Every incident adds a rule, and the document grows until nobody reads it before launching. Keep the one page, and move detail into the claims register and the settings checklist where it belongs.
The controls over-block. B2B audiences are small already, and a block list copied from a consumer brand can cut you off from trade publications and industry forums your buyers actually read. Check the placement report for the good sites your exclusions removed as well as the bad ones they blocked.
Nobody owns it. If the policy has no named owner and no review date, it becomes a document people cite after an incident rather than one that prevents it. Before you write it, an AI readiness assessment will show which of these roles already exist in your team and which you'll need to create.
Where to start
This week, open your three ad platforms and write down where AI Max, Advantage+ creative enhancements and LinkedIn audience expansion are switched on, then paste the policy above into your wiki with real names against each section. Putting AI to work in your accounts with rules like these in place from day one is what Scalewell does with B2B paid media teams, so the time it saves isn't spent cleaning up after it.
Frequently asked questions
Is AI brand safety different from traditional brand safety?
Traditional brand safety is about where an ad appears, such as next to harmful content. AI brand safety adds what the ad says and shows when a platform generates or alters it, how far AI widens your audience, and what data you give AI tools. The first problem is solved with block lists, the second with settings, review and a claims register.
Do I need to disclose AI-written ad copy?
Usually not, as of November 2026. The IAB's August 2026 framework says text or copy should not automatically require labels, and the main disclosure laws target synthetic people and realistic generated imagery. The claims in that copy still have to be true and supported, so AI-written text needs the same review as any other copy.
Can I turn off AI Max in Google Ads completely?
Yes. Google's help page says you can open the campaign's settings, go to the AI Max section and toggle it off, which opts the campaign out of all AI Max settings. Existing URL inclusions and exclusions are then ignored. New Search campaigns start with AI Max on, so each new campaign needs the same check.
Who should own AI guardrails in a B2B marketing team?
Give the policy one owner, usually the VP Marketing or head of demand generation, and split the sections. The brand owner covers generated words and images, the head of paid media covers placements and agent access, a compliance owner covers claims and disclosure, and whoever owns privacy covers customer data. Agencies follow the same policy in writing.
How often should we review AI ad settings?
Review the platform settings at least quarterly, and monthly if agencies or several people build campaigns, because defaults apply to every new campaign. Review the claims register whenever a case study, benchmark or product capability changes. Put a date on the policy so a reader can tell whether it is current.