An AI content system for a psychologist's private practice, with professional and legal limits
A private psychology practice's Facebook posts used to be long, hand-written educational pieces. Now there is a 13-week content plan, two text variants and an illustration for every post, an approval interface, and a 19-point rule list so the copy never promises a cure.
- Client
- Private psychology practice with group sessions; internal pilot at a partner business
- Status
- Being rolled out, internal pilot, November 2026
The starting point
The practice is a sole-practitioner psychology consultancy with group sessions: a women’s self-awareness circle and a meditation group. It has a Facebook page and a blog, and the earlier posts were long educational pieces that took a lot of time to write. In a helping profession content is not advertising but trust-building, so it cannot be dashed off, yet alongside consultations there is no room for it week after week.
The constraints are strict, too. Copy published in a psychologist’s name cannot promise results, cannot claim a healing effect, and cannot contain client testimonials. The psychologists’ code of ethics and the rules on health claims are not the small print here; they are the heart of the task.
The solution
A system that produces the raw material for the posts but publishes nothing on its own.
- A 13-week content plan built around the practice’s own events and topics: group launches, meditation sessions, educational themes, spread out at an even rhythm.
- Two text variants and one illustration for every post, matched to the website’s visual identity. The practitioner picks one, or rejects both.
- A web approval interface: approve, edit by hand, send back with a comment, request a new image with specific guidance, or upload her own picture instead of the generated one.
- Scheduled publishing to Facebook, only after approval, at the time fixed in the plan.
- Email notifications and a results dashboard on the same interface: the practitioner gets an email when a new post is waiting, and sees in one place what went out and how it performed.
The hard part: writing in a health professional’s name
Technically, writing the copy is the easy part. The hard part is getting the machine to produce text that a psychologist can stand behind with her name, and that does not break any rule on Facebook either.
The answer is a 19-point professional and legal rule list built into the writing instructions. Its sources: the psychologists’ code of ethics, the EU rules on health claims, Meta’s content policies and the transparency requirements of the EU AI Act. It is not an after-the-fact check but a condition for the text to be produced at all.
The trial post was about a herbal topic. Without any effect claims, approached from tradition and symbolism, it passed the filter and still read well. Along the way a risk surfaced in the old content too: earlier posts contained healing claims, while the practice also sells a product of its own. Introducing the system brought this to light, and reviewing the old content went on the task list.
Deliberate limits. Nothing is published without approval. No promise of a cure, no diagnosis, no client testimonials, no case descriptions, no dosage. On crisis topics (self-harm, suicide) the system does not write at all. Prices and available places go into a post only if the plan says so.
What we measured, and what we did not
The numbers that can be read from the system: 26 posts planned for 13 weeks, one post with two text variants and an image produced in 35-60 seconds based on three measured runs, and two posts have gone through the full process so far: one sent back for revision, the other with a new image requested.
Business results, meaning reach, reactions and time saved, do not exist yet: they are measured after live use. Machine cost is only known as an estimate, so it is not shown as a headline number. There is no precise data on the starting workload either, only that a post took a lot of time; so this study makes no “saves this many hours” claim.
Status and next steps
The first instance of the system runs on Matrix AI’s own Facebook page, with the first post scheduled for 9 November 2026. The practice’s instance is waiting for Facebook access, after which the internal pilot starts. After the first published posts and the weekly measurement we will update the study with real numbers.
The data in the screenshots is fictional sample data. The practice’s name is withheld at the practitioner’s request.