BACP AI Guidance: A Checklist for UK Counsellors Before Using AI

Before using AI, a UK counsellor needs to meet the AI clause in BACP's Ethical Framework 2026: risk-assess each tool, be competent with it, know where client data goes, tell clients honestly, get informed consent before any of their personal data goes in, and keep every clinical decision their own. Add insurer and supervisor checks and an exit plan.

Two premises to correct first. Older summaries describe the framework as "under review" and say BACP has no rule on AI. Both are out of date. BACP's Ethical Framework page says the 2026 framework was formally adopted on 3 November 2026 and that all members must work to it from that date, and the framework itself now names artificial intelligence in Core responsibility 2.1(e). There is still no separate AI standard: the duty sits inside the framework. The one carve-out on timing is for accreditation, where an application submitted before 4 May 2027 may still reference the 2018 framework.

Follow me on Instagram@sagnikteaches

What BACP has actually published, and what it has not

It helps to separate four kinds of material, because they carry different weight and counsellors often blur them.

Connect on LinkedInSagnik Bhattacharya

• The Ethical Framework. The binding document, and the reference point BACP's professional conduct panels use when they consider a complaint. The 2026 version was adopted on 3 November 2026. Its AI wording sits in Section C, Core responsibility 2.1(e), under keeping yourself and the people you work with safe. The clauses on privacy notices (3.1), records (3.2), insurance (4.2) and supervision (4.3) do not name AI, but they apply to every tool you use. Read Section C yourself in the PDF on the framework page.
• Ethics Essentials. BACP's short supporting resources. One is a 14-minute video, Ethics and the use of Artificial Intelligence, dated May 2026. The Ethics Essentials index needs a login, but the video's transcript is published on BACP's site. Its speakers say the virtue is to seek consent, to get active consent from clients, to test a tool before rolling it out, and to use AI for admin rather than to outsource ethical reasoning or reflective practice. The video, recorded during the review, calls the clause 2.1(a); in the adopted PDF it is 2.1(e).
• Events and slide decks. BACP hosts good-practice sessions on counselling and AI. One deck I read is a speaker's view, and it says so on its first slides ("Just because it is a BACP event does not mean it is correct"). It offers four things a counsellor can do. Useful, but not policy.
• News and consultation responses. In February 2026 BACP responded to the medicines regulator's consultation on regulating AI in healthcare, calling for transparency about which AI tools are regulated, shared accountability across developers, organisations and clinicians, ongoing training for professionals and monitoring of adverse incidents. That is BACP's lobbying position on regulation, not a rule for your room.

So the honest summary is: there is a binding framework with one AI clause, a supporting video, some good-practice thinking and a lobbying position, but no separate standard. The checklist below starts from the clause, then adds the framework's general duties and the published good-practice points. Where wording matters, I quote it and name where it comes from.

Subscribe on YouTube@sagnikteaches

Core responsibility 2.1(e), turned into evidence you can show

The clause asks members to assess the risk of any artificial intelligence tool, digital tool or online platform before using it, and to "be able to demonstrate" five things. The word that matters is demonstrate: each point needs something on file, not a good intention. Here is each one with the record that shows it.

A filled-in vendor sheet for point (ii) might read like this, for an illustrative dictation app used only for the counsellor's own post-session summary: storage region named in the vendor's privacy policy; audio deleted after transcription per the help page; transcripts kept until the user deletes them; content not used for model training on the paid plan, stated in the terms; vendor staff access only for support with the user's permission; delete-all option under account settings. Each line carries the page it came from. Where a line reads "not stated", that gap goes into the risk assessment, and often decides the matter.

Two other clauses catch people out. Clause 3.1(d) asks that clients know about and agree to foreseeable limits to confidentiality, including "the use of digital storage systems, platforms or tools that may monitor or collect data", so the privacy notice needs a line about the tool, not only the information sheet. Clause 3.2(c) says records include all communications, notes and other information held about clients, which suggests a transcript or AI draft you keep is part of the client record, with the same storage duties as your notes. If you are unsure how either applies to a particular tool, BACP's ethics service is the place to ask.

The four good-practice points, in the speaker's words

The slide deck, hosted on BACP's site, lists four things a counsellor can do. I quote the gist of each because they shape the checklist:

1. Do not use AI in therapeutic contexts unless you have training, insurance and informed client consent.
2. If you decide to use AI in therapeutic contexts, take responsibility for it: do a risk assessment and test it before using it with clients.
3. Even with a risk assessment, have a plan for what to do if it goes wrong, the client withdraws consent, or you can no longer use the tool.
4. Do not upload identifying client information to an AI system.

Those four lines are the skeleton of the dated plan that follows. Since 3 November 2026 the consent, testing and risk-assessment points are no longer only one speaker's advice: clause 2.1(e) makes them duties. The deck adds a test worth remembering on its consent slide: "If you can't undo it, probably don't do it."

A dated action plan for a practice owner

Dates below assume you are starting in the week of 9 November 2026. Shift them to suit. The point is that each action has an owner, a deadline and a record, so you can show your thinking if anyone asks.

Eight rows will not take long. The two that take effort are the insurer email and the consent wording, which is why they sit in the middle with time around them.

Sort every use into three groups before you decide anything

Most counsellors can use AI safely for something. The mistake is treating all uses as one decision. Use three groups.

Group 1: never touches a client

Drafting a website page, brainstorming a workshop title, tidying a policy you wrote, preparing a fee-change letter template. No client details, no session content. The risk is ordinary: accuracy and tone. Safe AI uses for counsellors that never touch client notes lists several worth trying first.

Group 2: client-adjacent

Anything that could identify or describe a client even loosely: an anonymised case for supervision, a reflective note about a difficult week, a reminder email with an appointment time. The question is whether the text could be traced back, including by the client themself. Default to not using AI here unless the tool, the contract and your consent wording have all been reviewed.

Group 3: client-facing or session-related

Recording or transcribing sessions, summarising notes, chatbots that talk to clients, anything a client sees or that holds what they said. This is where clause 2.1(e) and the four good-practice points bite hardest: competence, insurance, informed consent, a risk assessment and an exit plan. If you are not ready to do all of that, the answer for now is no. Read whether therapists can use ChatGPT for session notes for what goes wrong with general chat tools in this group.

Consent wording you can adapt

Informed consent means the client knows what the tool does, what data it holds, who can see it, how long it stays and what happens if they say no. Here is an illustrative information-sheet paragraph for a Group 3 use (a dictation tool that turns your spoken summary into a draft note, with no session audio recorded):

How I use AI tools in my practice

After a session I may dictate a short summary of my own, without your name, into a tool that turns my speech into a draft note. I check and edit the note myself. The tool [name] is used on [plan type], which does not use my content to train its models. The dictation is deleted within [x days] and your session is never recorded.

You can ask me not to use any AI tool for your notes at any time, with no effect on the service you receive. Please tell me if you would rather I did not. If I stop using the tool or change it, I will tell you.

Check every line against the real setup before you use it. The "[x days]" has to come from the vendor's documented retention setting, not from hope. The "no name" line only helps if you actually leave names out. If a client says no, record that in their file and make sure your workflow has a way to keep them out of the AI route. For the wider question of telling clients, should therapists tell clients they use AI goes through the options.

Insurance and supervision: two emails to send this month

The good-practice deck puts insurance next to consent, and for good reason. A policy written before generative AI may or may not treat an AI-assisted note as an ordinary part of practice. Do not guess. Send your insurer a short, specific email:

Subject: Question about AI tools and my professional cover

I am considering using [type of tool, e.g. a dictation tool for my own post-session summaries] with no client names and no session recordings. Does my current policy [policy number] cover professional activity that includes this use? If there are conditions, or exclusions for AI-assisted work, please set them out in writing.

Keep the reply with your register. Do the same with your supervisor, framed as a supervision topic: what are you planning, what are you worried about, and what would make them uneasy? A supervisor's concern is useful data, and the supervision notes are part of your record of taking responsibility.

Test and risk-assess a tool with invented material

The "test it before using it with your clients" point is easy to nod at and rarely done. A practical test takes under an hour.

1. Write a fictional session summary of 200 words with a fictional client, including two emotionally heavy lines and one risk-relevant statement.
2. Run it through the tool as you would in practice.
3. Check the output: did it drop the risk-relevant line, soften the heavy ones, add something you did not say, or label the client's feelings with certainty you did not have?
4. Check the settings: where does the text go, what is stored, how long, who at the vendor can see it, can you delete it?
5. Write the risk assessment as four short lines: what could go wrong, how likely, how bad, what you do about it.

A realistic result, illustrative: the tool summarised a risk statement as "the client expressed some low mood", losing the specific sentence. That one finding is enough to decide the tool may be used only for non-risk content, or not used, or used only with the counsellor checking every line against their own memory. The risk assessment says which. For tools that record, a confidentiality checklist for AI note tools adds the vendor questions to ask, and recording or dictation helps you choose the style.

The failure drill: three scenarios to write down

The third good-practice point asks for a plan for when it goes wrong, the client withdraws consent, or you can no longer use the tool. Write one paragraph for each, in the form "what I do, who I tell, what I record".

The tool goes down or changes

You arrive at a session and the dictation tool no longer works, or an update changes its behaviour. What you do: take notes by hand as you did before the tool. Who you tell: nobody unless it affects a client. What you record: the date, what changed, and whether any note was affected.

A client withdraws consent mid-way

What you do: stop using the tool for that client from that session. Delete what the tool holds for them if the vendor's settings allow it, and note when. Who you tell: the client, to confirm what has been done. What you record: the withdrawal and the deletion, with dates.

You suspect client information reached an AI tool by mistake

A typed name in a prompt, a pasted note, a transcript in the wrong account. What you do: stop, delete from the tool as far as you can, and note exactly what was entered and when. Who you tell: your supervisor, your insurer if you are unsure whether it is reportable, your data-protection contact or BACP's ethics service for guidance on whether it must be reported, and the client. Clause 2.2 of the framework asks you to explain fully and promptly what has happened, apologise to the people affected and act promptly to limit harm. What you record: everything above. Do not fix it quietly; a record of prompt action protects you and the client.

Tool settings to check at the six-month review

Settings drift. Vendors rename options and change defaults, and a setting that was off can become on after an update. At each review, open each tool's privacy settings and confirm three things: training on your content is off (on consumer plans it is usually a toggle in the privacy or data controls; business plans typically do not train on your content by default, but read the plan's terms), history and memory are set the way you intended, and you know how to delete what is stored. How to stop AI tools training on your business data goes through the common tools. If a screen has changed and you cannot find the setting, record that as a finding and ask the vendor.

A two-counsellor practice sorts its tools

Illustrative. Two counsellors share a rented room and a booking system. They run the register exercise on one afternoon and find seven uses of AI between them:

The surprise is row six. Neither counsellor had chosen to use AI there; the booking vendor added it. This is the commonest finding: the feature you did not install is the one that holds client text. Checking built-in features inside software the practice already uses is part of the register, and reading each one's privacy terms is part of the review, because clause 2.1(e)(ii) applies to a feature you did not choose as much as to one you did.

The outcome is not "no AI". Two Group 1 uses continue at once, three Group 3 uses are paused with named questions, and the practice has a one-page record dated this week. That is a defensible position: nothing risky is running, and the reasons are written down.

Three client questions and honest answers

Clients will ask. Have an answer ready, in your own words, for each of these.

• "Are you using AI on my sessions?" A plain yes or no, then the detail: what, where the data goes, how long. If the answer is no, say no, and say you will tell them before that changes.
• "Can I use AI between sessions and tell you what it says?" This is a client-side question, and BACP has written about clients turning to AI. Treat what the client brings with curiosity, not dismissal; the good-practice deck suggests starting from the client's own agency and doing a shared risk check if you think a suggestion could harm them.
• "Can you delete everything the tool has?" Know in advance whether you can. If a vendor cannot delete, that belongs in the risk assessment, and probably in the decision not to use it.

What to put on one page for your own file

If anyone ever asks how you approached AI, one page is better than a recollection. Keep this, updated each review:

• The date you read the Ethical Framework 2026 and the AI resource, and what you took from them.
• The register of tools and the three-group sort.
• The insurer reply and the supervision date.
• The consent wording with its version date.
• The test notes and risk assessment for each Group 3 tool, or a note that you do not use any.
• The three failure paragraphs.
• The date of your next review.

None of this requires a technical background. It is the same habit as keeping contracts, risk assessments and supervision notes, applied to a new kind of tool. If you want a low-risk place to practise first, pick admin that holds no client information at all, such as tracking your own CPD hours or insurance renewal dates. It carries little confidentiality risk and shows you how the tools behave before anything sensitive is involved.

Where the checklist will date, and what to watch

This subject is moving. The Ethical Framework 2026 was adopted on 3 November 2026, and BACP says it is supported by Ethics in practice and Ethics essentials resources, to which more may be added. BACP also continues to press for clearer regulation of AI in mental health. Expect changes: new Ethics Essentials, new guidance for particular settings, perhaps clarification from regulators. Check bacp.co.uk before each six-month review. If something published after November 2026 conflicts with this tutorial, BACP's page wins, and the dated plan should be updated, not defended.

Questions counsellors ask about BACP and AI

Does BACP have a standalone AI standard?

Not a separate one that I could find. The AI duty sits inside the Ethical Framework for the Counselling Professions 2026, adopted on 3 November 2026, at Core responsibility 2.1(e), with an Ethics Essentials video on AI published alongside it. Check bacp.co.uk before each review, because supporting resources are still being added.

Is the Ethical Framework 2026 mandatory?

Yes. BACP's framework page says all members must work to the 2026 framework from Tuesday 3 November 2026, and members accept that registration may be at risk if they fail its commitments. The one carve-out: an accreditation application submitted before 4 May 2027 may reference the 2018 framework; from that date it must use 2026.

Do I have to tell clients I use AI?

In practice, yes. Clause 2.1(e) asks you to be honest and transparent with people about your use of AI, including its benefits and risks, and to have their informed consent before inputting any of their personal data into AI or digital tools. Offer a no-AI route to anyone who declines and record their choice.

Can I use AI for admin that never touches client information?

Usually yes, with ordinary care: marketing text, a policy draft, a fee letter template. The checklist still applies if a prompt could identify a client, so keep names, details and session content out entirely. See the safe uses linked in this tutorial.

Further reads

• How to Use AI on SimplePractice: Note Taker or Care Aide? — How one practice platform's AI features are set up.
• Blueprint AI for Therapists: Notes, Cost and Alternatives to Check — Compare another notes product and its alternatives.
• AI Scribe Security: Audio, Retention and BAA Vendor Questions — Audio, retention and vendor questions to ask before buying.
• Is It Safe to Put Customer Data Into ChatGPT? — Why general chat tools are the wrong place for client details.
• Questions to Ask an AI Vendor Before You Sign Anything — Questions to put to any vendor before you sign.
• How to Set Up Human Review for AI Work Without Slowing Down — Keep a person checking AI output before it reaches a file.
• What to Check in an AI Tool's Privacy Policy and Terms — Read a tool's privacy terms before client text goes near it.
• Therapist Credentialing: Using AI to Track Applications and Renewals — A low-risk admin job to learn how AI tools behave.
• Jane App for Counsellors: AI Scribe Cost, Consent and UK Data Location — What Jane's AI Scribe costs a counsellor, where recordings are stored, what the Class I status means, and how to handle consent before each session.
• Starting a Private Counselling Practice in the UK: What AI Can Do — The admin a new UK private counsellor sets up first, which parts AI can draft safely, which it must never touch, and what first-year running costs look like.
• AI Note Generator for Counsellors: Turn Your Own Bullets Into Notes — Type your own session bullets, strip identifiers, and have AI draft the note you then edit and own. Prompts and examples included.
• Counselling GDPR: Can You Use AI With Client Information? — A counsellor's step-by-step vetting of any AI tool before client information goes in: basis, contract, location, retention and privacy notice.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: BACP website, all read on 4 November 2026: the Ethical Framework for the Counselling Professions 2026 page and PDF (formally adopted 3 November 2026), the ethics hub, the transcript of the Ethics Essentials video Ethics and the use of Artificial Intelligence (May 2026), the news item of 23 February 2026 on AI regulation, and a BACP-hosted good-practice slide deck on counselling and AI. This is a practical summary, not BACP's own wording or legal advice.