AI in the psychology practice stopped being a debate about the future somewhere in the last two years. Today there are therapists transcribing sessions with an automated tool, drafting reports with a generative assistant, or seeing clients who have spent months talking about their problems with a chatbot before booking an appointment.
The American Psychological Association has put a name to that reality.
Trend 1: AI as clinical decision support
The first trend the APA points to is the use of personal data and AI models to sharpen which treatment works best for each person.
Historically, diagnosis and treatment planning have leaned on what the client reports and on their history. The APA describes how that is beginning to be complemented by other sources: phone and wearable data, clinical records, sleep and activity patterns. Tools able to analyse large volumes of information and hand the professional something actionable: patterns that would otherwise go unnoticed, signals that it is worth intervening sooner, criteria for adjusting the plan without resorting to trial and error.
The word the APA repeats here is permission. This data is shared only with the client's consent. And the association's conclusion is just as explicit: with the right regulation and research, this line can widen access to better treatments, and psychology professionals should take a leadership role both in developing that technology and in educating their clients about it.
In other words: not "wait and see", but "be in the conversation".
Trend 2: digital companions and the bond that is not one
The second trend runs in the opposite direction, and the APA treats it with considerably more caution.
Applications designed specifically to simulate human company — Replika, Character.AI and dozens of imitators — have become mainstream.
Character.AI has around 20 million monthly users and more than half are under 24. Several studies place seeking mental health support and companionship among the two main reasons people use generative models.
APA, Monitor on Psychology (2026)The problem the association identifies is not that these tools exist, but how they are designed. Many chatbots are optimised to maximise engagement, and that objective is not the same as keeping someone safe or improving their mental health. Research is starting to show that heavy use can isolate a person further, worsen loneliness and erode social skills.
Hence the APA's insistence on a very concrete role for professionals: to remind people, with evidence in hand, why human connection is irreplaceable.
The distinction that orders the whole debate
Set side by side, the two trends draw a fairly sharp line:
| AI that supports the clinician | AI that replaces the clinician | |
|---|---|---|
| What it does | Organises information, reduces admin, surfaces patterns | Talks to the client, offers containment, simulates a bond |
| Who decides | The therapist | The system |
| What the design optimises for | The professional working better | The user carrying on interacting |
| The APA's position | An opportunity, with professional leadership | Cause for concern and vigilance |
The useful question for judging any tool is not "does it use AI?". It is "which side of that table is it on?". A system that hands you your own clinical information better organised does not compete with the therapeutic bond: it frees up time and attention for you to hold it. A system that puts itself in the therapist's place does.
What the rules require in Spain (and what has just changed)
Here a therapist practising in Spain has an extra layer that the APA, for obvious reasons, does not cover. And there has been recent movement.
Data protection
Health data is a special category under the GDPR (Art. 9). Any tool that processes session content implies a processing arrangement with a signed agreement (Art. 28 GDPR), clear information to the client and a risk analysis. If the provider trains models on that data or transfers it outside the EEA without safeguards, the problem is yours as much as theirs: the controller is the professional.
The European AI Regulation
The calendar moved this summer. Regulation (EU) 2026/1744, known as the digital omnibus on AI, entered into force on 27 July 2026 and postponed the obligations for Annex III high-risk systems from 2 August 2026 to 2 December 2027 (and those in Annex I to August 2028).
That postponement is worth reading carefully, because headlines told it as if the AI Act had been stopped. It has not:
- The prohibited practices (Art. 5) have applied since 2 February 2025. The omnibus does not soften them: it adds two more, non-consensual intimate imagery and child sexual abuse material generated or manipulated with AI.
- The AI literacy duty in Art. 4 does not disappear, but it is softened. Where providers and deployers previously had to ensure a sufficient level of literacy, the new wording has them "adopt measures to support the promotion of AI literacy among their staff". The duty to know what you use is still there; what drops is the bar you can be held to.
- The transparency obligations in Art. 50 have applied since 2 August 2026 — a chatbot has to identify itself as one, and AI-generated content has to be marked — with a four-month transitional period for systems already on the market before that date.
Translated to the practice of a consulting room: what was postponed mostly affects manufacturers of systems classified as high risk. What was not postponed is exactly what touches your day to day — knowing what you use, on what data, and telling your clients about it.
Professional code of ethics
The code of ethics of the Spanish General Council of Psychology does not mention AI, but its principles apply without effort: professional secrecy, informed consent, competence (do not use what you do not understand) and responsibility for clinical judgement, which cannot be delegated to a system. If you need the base document, we have a free informed consent generator.
Seven questions before putting an AI tool in your practice
A practical filter, in the order worth applying it:
- Where is the data hosted? Servers in the EU is not a technical detail: it is the difference between a simple processing arrangement and an international transfer file.
- Are models trained on the content of my sessions? The acceptable answer is "no", and it has to be in the contract, not on a marketing page.
- Is there a signed and accessible data processing agreement? If you cannot download it before signing up, that is a signal.
- Who makes the clinical decision? If the tool issues diagnoses or treatment recommendations instead of organising information for you to decide, your liability profile changes.
- What do I tell my client, and when? Informed consent about the use of technology has to be specific, prior and revocable. And the conversation usually goes better than you fear.
- Can I export and delete? Portability and erasure are not optional, and they are also your insurance against the provider disappearing.
- Do I understand enough about how it works to explain it to a client in two sentences? If not, Art. 4 of the AI Act has something to say to you, and so does your code of ethics.
Where Alchely stands
We built Alchely squarely on the left-hand side of that table, and deliberately so.
Alchely does not talk to the client. It does not replace the therapist's judgement or simulate a bond. What it does is take on the part that consumes time and erodes attention: longitudinal clinical memory, structured notes, continuity between sessions. So that when someone comes back after five weeks, you do not depend on your memory or on three loose sheets of paper to pick up exactly where you left off.
As for the framework above: servers in the European Union, no model training on client data and a formal data processing agreement. Not as a sales argument, but because any other configuration passes the professional a risk that is not theirs to carry. We set it out on the privacy page.
The APA asks psychology professionals to lead how this technology is developed. That seems to us the right ask, and we would rather build with clinicians in front of us than for them.
Frequently asked questions
Can I use AI to transcribe or summarise my sessions?
Yes, provided there is specific informed consent from the client, a data processing agreement with the provider and guarantees about where the data lives and that it is not reused for training. A session recording is health data: the standard of care is high.
Does the European Regulation ban AI in psychology?
No. The AI Act does not ban the use of AI in mental health. It bans specific practices and sets obligations graded by risk level. The digital omnibus of July 2026 postponed the high-risk obligations to December 2027, but kept the prohibitions — adding two more — and transparency in force, and kept AI literacy in a softer wording.
What does the APA think of therapy chatbots?
The APA warns that many are optimised to maximise engagement, an objective distinct from keeping the person safe or improving their mental health, and points to evidence that heavy use of digital companions can increase isolation. Its position is that professionals should take part in developing and regulating these products.
Do I have to tell my client I use an AI tool?
Yes. Because of informed consent and because of the duty of transparency. On top of that, if the tool interacts directly with the client, Art. 50 of the AI Act requires it to identify itself as an AI system, in force since August 2026.
Can AI replace the therapist?
The APA's position points clearly the other way: professionals have a central role in reminding people why human connection is irreplaceable. AI can widen access and reduce administrative load; the therapeutic bond is not a process that can be outsourced.
Sources
- Straight, S. (2026). What's ahead for psychology? 9 trends to watch in 2026. APA Monitor on Psychology, 57(1), p. 54.
- Infocop / Consejo General de la Psicología de España (15 January 2026). Principales tendencias que la Psicología debe tener en cuenta en 2026, según la APA.
- Regulation (EU) 2024/1689 (European Artificial Intelligence Regulation).
- Regulation (EU) 2026/1744 of the European Parliament and of the Council of 8 July 2026 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 as regards the simplification of the application of harmonised rules on artificial intelligence (digital omnibus on AI). OJEU of 24 July 2026; in force since 27 July 2026.
- Regulation (EU) 2016/679 (GDPR), Arts. 9 and 28; Spanish Organic Law 3/2018 (LOPDGDD).