Mental health advertising: the approval you apply for, the audience tools that are switched off, and the order that followed a questionnaire

Mental health advertising is treated as its own case by both the platforms and the regulators, and the three rules that matter most are not about what your copy says. One decides whether you may run at all and in which countries. One removes most of the audience tools a media planner would reach for by default. One is an enforcement order about a web form, not an advertisement. Settle those three before anybody writes a headline, because each of them can invalidate a plan that is already built.

This page covers which services sit inside Google's restricted addiction policy and which sit outside it, what "sensitive interest category" removes from your account, what the FTC said BetterHelp did and what its order requires, the measurement setup that keeps you out of that shape of trouble, and the copy that survives review. The general version of this subject, covering claims, platform certification and patient data across all of healthcare, is a separate page linked near the end.

What is restricted, and what is not

Google's healthcare and medicines policy on addiction services is specific about its own scope, and the scope is narrower than most people assume. The policy states that "Google restricts the promotion of recovery-oriented drug and alcohol addiction services", and gives clinical treatment providers, sober living environments, referral agencies and crisis hotlines for drug and alcohol addiction as examples of what falls inside it.

Two details decide whether you can run a campaign at all.

• Six countries, and an application in each. The policy permits advertising in Australia, Canada, France, Ireland, New Zealand and the United States, and only for advertisers who have applied and been approved through Google Ads. Each location requires a separate application, so a provider running in two of those countries goes through the process twice.
• The approval is a timeline item, not a form. Build it into the plan ahead of creative and media, because a campaign that is ready to launch and not yet approved is a budget sitting still.

The policy is equally clear about what it does not cover. In its own words, "Promotion of services unrelated to drug and alcohol addiction, like services for impulse control disorders, behavioral addiction, or nicotine addiction, are out of scope of this policy." That cuts both ways in practice. A therapy practice, a psychiatry group or a counselling service that does not treat drug and alcohol addiction is not inside this particular restriction, which people often assume it is. A gambling or gaming addiction programme is also outside it, which people often assume it is not. Read your own service against those words rather than against the category you file yourself under.

On enforcement, the same page states that "A warning will be issued at least 7 days prior to any suspension of your account." That is useful to know and a poor thing to rely on. Treat the warning as a deadline you hope never arrives, and keep the person who can answer a policy query named in your own documentation.

The audience tools are switched off, and that changes the media plan

Google's personalized advertising policy keeps a list of sensitive interest categories. Three entries on that list sit directly over this field: Health, Abuse and trauma and Relationship hardships. There are 22 categories in all, and they include political affiliation, sexual orientation, religious beliefs, negative financial status and clinical trial recruitment, which tells you what kind of list it is.

The restriction is stated once for the whole group: advertisers promoting within these categories "are unable to use advertiser-curated audiences". In the account, that shows up as four things marked not supported.

Tool	Status for a sensitive interest category
Customer Match	Not supported
Your data segments	Not supported
Audience expansion	Not supported
Lookalike segments	Not supported
Predefined Google audiences	Available

So the whole family of plans that starts with "upload the client list and build a lookalike" is unavailable here, and a media proposal built on it is wrong rather than ambitious. What remains is strong enough to work with, and it is the honest version of this category anyway.

• Query intent. Somebody typing the name of a modality, a condition, a city plus "therapist", or "accepts" plus an insurer has told you what they want without you profiling anybody.
• Geography bounded by licensure. A clinician may treat people located in the states or jurisdictions they are licensed in, so the service area is a legal fact rather than a radius setting. That is the single most common source of wasted spend in this field.
• Availability as the offer. When the thing in shortest supply is an appointment this month, saying so is better targeting than any segment.

What the BetterHelp order actually says

The case worth knowing in full is the Federal Trade Commission's action against BetterHelp, announced on 2 March 2023. It is the clearest published statement of where the line sits, and the conduct it describes is a tracking and data sharing problem rather than an advertising copy problem.

According to the FTC, BetterHelp promised consumers it would not use or disclose their personal health data except for limited purposes such as providing counselling, and then disclosed email addresses, IP addresses and health questionnaire information to Facebook, Snapchat, Criteo and Pinterest. The complaint describes using consumers' email addresses together with the fact that they had previously been in therapy so that Facebook could find similar consumers for advertising. The FTC also alleged the company obtained no affirmative express consent before disclosing health data, placed no limits on what third parties could do with it, and in 2020 publicly denied news reports that it shared consumer information.

The order requires BetterHelp to pay 7.8 million dollars to fund partial refunds for consumers who paid for its services between 1 August 2017 and 31 December 2020. It also bans disclosing health information for advertising and re-targeting using certain consumer data, prohibits misrepresenting its sharing practices, requires affirmative express consent before disclosing personal information to certain third parties, requires a comprehensive privacy programme with data safeguards, requires the company to direct third parties to delete the data it had revealed to them, and limits how long personal and health information is retained.

Read the list of remedies rather than the number. "Direct third parties to delete the consumer data it revealed to them" is the part that is genuinely hard to do, and it is the reason to get this right at setup: once a signal has left your site and entered somebody else's advertising system, retrieving it is a legal exercise rather than a technical one.

The setup that keeps a campaign out of that shape

Every item here is about what you do not collect and do not send. Go through it before launch, with whoever actually controls the tag manager in the room.

• No condition in any string a tag can read. URLs, page titles, form field names, button labels, conversion event names and campaign names all travel. "anxiety-intake-complete" as an event name is a health disclosure with a technical costume on.
• Intake questionnaires live on pages with no advertising pixels. That is the specific pattern the FTC complaint describes. If the questionnaire has to be on the site, the pixel does not have to be on the questionnaire.
• Never upload a client or enquiry list to an advertising platform. In this category the tool is switched off anyway, and the attempt is the risk.
• Count appointments, not conditions. One conversion called "appointment booked", with nothing attached about who or why, is enough to run a campaign on.
• Configure call tracking not to retain content, and check what your chat tool stores and who it is stored with. Both are vendors processing health information on your behalf.
• Write down what you refuse to collect, and give the document a date. It is the thing that keeps a well meaning growth suggestion from quietly reversing all of this in six months.

Copy that survives review, and the line the landing page has to carry

The strongest advertisement in this field is a specific one, because the reader is trying to work out whether you can actually help them this month.

• Name the modality and who delivers it. The approach, the licence held, and the state or jurisdiction it is held in.
• Name the practical facts. Telehealth or in person, which insurers are accepted or that the practice is private pay, the fee band, and whether there is a waitlist.
• Do not state an outcome you cannot evidence, and do not imply one you would not be allowed to state directly. No recovery rates without a source, no promises of a result, no before and after framing.
• Do not write copy that asserts something about the reader. Addressing the person as though you already know their diagnosis is both a platform problem and the wrong tone for the one moment they are deciding whether to contact anybody at all.
• Carry the crisis route on the landing page. A campaign aimed at people looking for help will reach some people in crisis. In the United States that route is the 988 Suicide and Crisis Lifeline, reachable by call or text on 988, which its own site describes as available 24/7/365, free and confidential, with text and chat services in English and Spanish. Put it where it is visible without scrolling, not in the footer.

For the general frame across the whole sector, the claims standard, the platform certifications and the patient data constraint, see what healthcare advertising allows you to say and how to brief an agency. For the unpaid side of the same buyer, where the directory usually owns the search and the licence draws the map, see digital marketing for a therapy practice.

Questions about mental health advertising

Can you advertise mental health services on Google?

Yes, and the answer depends on which service. Recovery-oriented drug and alcohol addiction services are restricted: they may run in Australia, Canada, France, Ireland, New Zealand and the United States, and only for advertisers approved through an application made for each location. Services unrelated to drug and alcohol addiction, which Google's policy illustrates with impulse control disorders, behavioural addiction and nicotine addiction, are outside that particular restriction.

Why can I not build a lookalike audience for a therapy practice?

Because Health, Abuse and trauma and Relationship hardships are sensitive interest categories in Google's personalized advertising policy, and advertisers in those categories cannot use advertiser-curated audiences. Customer Match, your data segments, audience expansion and lookalike segments are marked not supported. Predefined Google audiences remain available.

Can we retarget people who read a page about a condition?

Treat that as off limits and build the campaign so that it never becomes possible. Retargeting assembled from health interest sits against platform policy and against data protection rules at the same time, and the signal that makes it possible is the same signal that creates the exposure. Retarget from a neutral page such as a general services page if you retarget at all.

Is HIPAA the only rule that applies?

No, and the BetterHelp case is the reason to know that. The FTC's action rested on the company's own promises and on consent, which applies to businesses whether or not they are HIPAA covered entities. Platform policy is a third separate rulebook, and in the United Kingdom and the European Union health data is a special category under data protection law with its own consent requirements. Check all of them for every market you run in.

What should the ad land on?

A page that answers availability, licensure, fee and insurance, and modality, with one clear action and the crisis route visible. Not the homepage, and not a page carrying an intake questionnaire with advertising pixels on it.

How do we measure without collecting health information?

Count appointments booked, calls answered and enquiries received, with no condition attached to any record, and take a baseline before the campaign starts so the comparison is yours rather than an industry figure. Report by service line in aggregate rather than by person.

Where people describe the problem before they search for a provider

Before anybody types "therapist near me", a lot of people now describe how they are feeling to ChatGPT, Gemini, Perplexity, Claude or Google's AI Overviews, and then ask what kind of help that is and where to find it. Those answers name a small number of providers and directories and cite a small number of sources, and the five do not agree with each other, so being named by one says nothing about the others. For this subject they lean on recognised clinical publishers, professional directories and provider pages with clear clinical authorship, which means an accurate, properly authored services page is doing work in a place you have probably never looked at. Ask the questions a prospective client would ask in two assistants, keep your own name out of the prompt so the answer comes back clean, and write down who is named and which sources are cited. AI Knows Us runs those questions across the main assistants and records what the answers were built from.