Recovery & Wellbeing

Would you tell an AI something you haven’t told anyone else?

By Craig Bilton, Founder & Clinical Director · 8 September 2026 · 8 min read

It can feel easier to type a difficult thought into a chatbot than say it aloud to another person. Digital tools may offer a place to organise thoughts or prepare for a conversation, but a reassuring response is not evidence of clinical understanding, confidentiality or safe advice. AI should not be relied on to assess withdrawal, manage a crisis or replace addiction treatment.

Imagine opening a chat window late at night and typing a sentence you have been avoiding: “I think my drinking is getting out of control.” There is no face watching you, no appointment to arrange and no immediate need to explain everything. That apparent distance can make the first words feel more manageable.

The important question is what happens next. Does the conversation help you move towards appropriate support, or does it become somewhere to keep the problem contained and hidden?

Why talking to a machine might feel easier

Shame, uncertainty and fear of judgement can make an ordinary conversation feel consequential. You may worry that a professional will insist on a label, that a partner will be hurt or that admitting the problem makes it impossible to take back.

A chatbot can appear to offer control over the pace. You can pause, rewrite a sentence or close the window. That may explain some of its appeal, but it does not mean the system understands you in the way a person does.

The potential opening is worth taking seriously. So is the possibility that an apparently helpful conversation never reaches someone able to assess the situation and act.

What the research actually shows

A study published in JMIR Formative Research on 20 May 2026 described the development of a chatbot called Suzy for people receiving substance use disorder care. Patient usability testing involved eight participants using a rule-based version. A later large-language-model version underwent expert co-design and safety review.[1]

The study reported encouraging feasibility and acceptability findings. It did not demonstrate that the chatbot reduced substance use, prevented relapse or could replace a recovery professional. The authors identified evaluation of effectiveness in larger, real-world studies as a next step.[1]

That distinction matters. A tool can be easy to use and well liked without having established clinical benefit. Results from a specialist research prototype also cannot establish that a general-purpose chatbot, or a different commercial recovery app, is effective or safe.

When you read a headline about AI treatment, ask what was actually measured: satisfaction, the quality of a conversation, symptom change, substance use or long-term outcomes. Those are different questions.

A useful role may be preparing for human support

The following are examples of limited, practical tasks to consider. They are not benefits established by the Suzy study, and they do not require disclosing identifying information.

• Turn a rough list of concerns into questions for a GP or drug and alcohol service.
• Practise a simple opening sentence for a conversation with someone you trust.
• Organise your own notes about situations you want to discuss at an appointment.
• Ask for a plain-language explanation of a term, then check it against a reliable source.

For example, you could ask for help wording a request for an appointment without entering your full substance-use history. The useful output might be a short message you can review and send yourself.

Keep factual details in your own words and check that the tool has not added symptoms, assumptions or promises. If an output does not sound like your experience, you do not have to adopt it.

You can also take these steps without AI. A notebook, a helpline or a conversation with a trusted person may meet the same need more comfortably.

Warmth and confidence are not a safety check

AI-generated language can sound certain even when the answer is incomplete or wrong. The World Health Organization's guidance on large multimodal models in health identifies risks including inaccurate outputs, bias, privacy concerns and people placing too much trust in automated answers.[2]

In recovery, that matters when a response appears to approve a home detox, dismiss a physical symptom, make a diagnosis or reassure someone that a risky pattern is under control. You should not treat conversational confidence as an assessment.

Be cautious if the answer gives precise medical instructions without a qualified professional assessing you, invents a source, or frames disagreement from people around you as proof that only the chatbot understands. An answer that feels supportive still needs to be checked against reality.

A tool should leave room for uncertainty and appropriate professional involvement. It should not make you feel obliged to trust it, remain in the conversation or withdraw from other support.

A private-feeling conversation may still create a data record

Before sharing sensitive information, check who operates the service and read its privacy information. Look for clear explanations of whether conversations are stored, who can access them, how long they are retained, whether they may be used to improve systems and how deletion works. If those answers are unclear, avoid adding details you would not be comfortable disclosing.

Do not paste another person's messages, clinical notes or identifying history into a chatbot without appropriate permission. You can ask about a general situation without including names, addresses, employer details or dates that identify someone.

Different tools and account settings may have different terms. A familiar name, a padlock in the browser or a friendly tone does not by itself establish clinical confidentiality. WHO guidance treats privacy and data governance as core requirements for AI in health.[2]

Notice whether digital support is connecting or isolating you

One useful personal check is what happens after the conversation. Have you contacted a service, prepared an honest discussion or followed an agreed recovery plan? Or have you spent another evening seeking reassurance while avoiding the action you already know needs attention?

This is not a diagnostic test for dependence on technology. It is a way to consider whether a tool is serving the purpose you chose it for.

If you find yourself repeatedly asking the same question until you receive the answer you want, bring the question to a trusted person or qualified professional. If you feel increasingly reluctant to discuss the chatbot's advice with anyone else, that is also worth pausing over.

Recovery support should widen the people and options available to you. Our guide to structured recovery support looks at the role of a continuing plan beyond moments of motivation.

Situations that need a real person now

Do not use a chatbot to decide whether you can safely stop alcohol or benzodiazepines, to design a taper, or to manage suspected overdose. If alcohol dependence may be present, get medical advice before stopping or sharply reducing your intake.[3]

Call 999 for an overdose, seizure, collapse, severe breathing difficulty, an inability to wake someone or immediate danger to life. If you are in a mental-health crisis, use an appropriate NHS or local crisis service. In England, urgent mental-health help is available through NHS 111; immediate danger requires 999 or A&E.[4]

An AI conversation should not delay that contact. Nor should you assume a chatbot is monitored by a person or able to summon help unless the service explicitly establishes those capabilities.

Let the first words become a next step

If writing to a chatbot has helped you admit that something is wrong, the next useful step may be sharing one sentence with a person: “I have been trying to work this out privately, and I need some help.” You do not have to arrive with a diagnosis or a polished explanation.

The NHS guide to getting help for drug problems explains routes into support. IRN can also offer a non-urgent conversation about its services and treatment-navigation options. It is not an emergency service and does not prescribe or provide medical detox.

This article provides general education for adults. It does not endorse a particular chatbot or claim that IRN's digital products have the outcomes of another system's research. The images are illustrative and do not depict real clients or a functioning clinical chatbot.