Let people refer themselves for talking therapy through a chat, at any hour
In England, many NHS Talking Therapies services use Limbic Access, an AI chatbot on their websites, to take self-referrals and prepare a report for the clinician. In a study of 28 services, referrals rose 15% over three months where it was used, against 6% in matched services.
Try it withClaudeChatGPT / OpenAI modelsGeminiGrok
Deployed; Class IIa UKCA medical device since January 2023, used by NHS Talking Therapies services in England; NICE says it can be used during a three-year evidence generation period, on conditions, Limbic, with NHS Talking Therapies services in England
Why it matters
NHS Talking Therapies is the NHS service in England that treats common problems such as anxiety and depression, and people can refer themselves to it. Services that do not use a chatbot collect self-referrals in other ways, such as web forms. Limbic built Limbic Access, a chatbot that takes the referral as a conversation instead, at any hour. In January 2023 it gained Class IIa UKCA medical device status, which Limbic describes as a first for an AI mental health chatbot. A study published in Nature Medicine in 2024 looked at data from 129,400 people referring themselves to 28 NHS Talking Therapies services; half used the chatbot. Over three months, referrals rose 15% in the 14 services that used it, against 6% in 14 matched services that did not, and the authors write that the increase was particularly pronounced in minorities, including 29% among people from ethnic minorities. Their analysis of free-text feedback from 42,332 people pointed to two potential drivers of the more diverse uptake: the chatbot's human-free nature, and people's own realisation that they needed treatment. NICE's early value assessment says Limbic Access, and one other tool, can be used in the NHS for people aged 16 and over while more evidence is gathered over three years. NICE puts its use at about 40% of NHS Talking Therapies services; Limbic says it has served 45% and delivered clinical assessments for over 650,000 patients.
How it works
A person visits their local NHS Talking Therapies website and starts the referral in a chat window instead of a form. NICE's description says the chatbot checks eligibility, takes contact details and demographics, asks about presenting symptoms and collects the standard PHQ-9 (depression), GAD-7 (anxiety) and WSAS (work and social adjustment) questionnaires, plus further screening questions. It then produces a clinical report summarising concerns, risk levels and clinical notes for the clinician who carries out the assessment, and it can suggest a problem descriptor, the label services use for a person's main difficulty; NICE left that function outside its assessment. Limbic says the tool links with the electronic record systems services already use and with NHS Spine to confirm eligibility. The decision on treatment stays with the clinician: Limbic describes the tool as augmenting the human-led clinical assessment. NICE lists a licence fee per referral of £4.20 to £6.59 including VAT, falling as volume rises, with no fixed costs. Its conditions are that the tool has regulatory approval, including NHS England's Digital Technology Assessment Criteria, and that the evidence NICE set out is collected, on matters including the quality of the information gathered, its effect on clinical decisions, administrative burden, service users' views and costs such as training, promotion and digital safety assurance. The committee also asked for comparisons with usual referral routes and for referral completion rates.
Who it's for
Mental health services, health systems and commissioners that take self-referrals online and want to reach more of the people who need help, and teams evaluating conversational intake tools.
The prompt
You are helping me plan a pilot of a conversational self-referral assistant for a mental health service. Do not ask me for any real person's name, contact details, symptoms or records. Our service: [for example, a public talking therapies service for adults in one region, 12,000 referrals a year, self-referral by web form today] Who we struggle to reach: [for example, men under 30, people from minority ethnic groups, people working night shifts] What we can use: [for example, an approved supplier or an AI model under a contract that keeps data in our country, and our electronic record system] 1. List the questions the assistant must ask to complete a referral, in plain language, including eligibility, contact details and the validated questionnaires our service already uses. Mark which answers are required and which are optional. 2. Write the safety rules: how the assistant recognises a risk to life or a crisis, the exact words it uses to direct the person to urgent help, who on our staff is alerted and how fast, and what it must never do (diagnose, promise treatment, give advice on medication). 3. Describe the report a clinician receives before the assessment: what it contains, how any suggested label is shown as a suggestion, and how the clinician corrects it. 4. Set the privacy rules: what is stored, where and for how long, who can see it, and whether any conversation may be used to train a model. 5. Plan for people the chat may exclude: other languages, low literacy, screen readers, and a phone route that stays open. 6. Design the evaluation: compare referrals, completion rates and the mix of people referring with the months before the pilot and with a similar service that does not use it; check a sample of reports against clinicians' own assessments; and collect feedback from service users and staff. Leave placeholders rather than invent rules, figures or targets, and mark everything a clinical lead, data protection officer or regulator must sign off before launch.
Open inClaude (opens in a new tab)ChatGPT (opens in a new tab)Gemini (opens in a new tab)Grok (opens in a new tab)
Sources
The Nature Medicine study is peer-reviewed but was funded by Limbic, and the authors named in its competing-interests statement are five Limbic employees and one paid consultant. NICE's assessment is independent of the company, though it notes that most of the evidence it saw came from company-affiliated sources. MIT Technology Review reported the study and sought an outside view. No independent study of the tool's accuracy or of treatment outcomes was found.
1. Nature Medicine (Habicht et al.): "this increase was particularly pronounced in minorities". Closing the accessibility gap to mental health treatment with a personalized self-referral chatbot (opens in a new tab)
2. MIT Technology Review (Rhiannon Williams): "The number of referrals from services using the Limbic chatbot rose by 15%". A chatbot helped more people access mental-health services (opens in a new tab)
3. NICE (HTE30, now HTG756): "a clinical report summarising concerns, risk levels, clinical notes". Digital front door technologies to gather service user information for NHS Talking Therapies: information about the technologies (opens in a new tab)
4. NICE: "can be used in the NHS during the evidence generation period". Digital front door technologies for NHS Talking Therapies: early value assessment (opens in a new tab)
5. Limbic: "augmenting the human-led clinical assessment". AI mental health chatbot that predicts disorders becomes first in world to gain Class IIa UKCA medical device status (opens in a new tab)
6. Limbic: "clinical assessments for over 650,000 patients". Limbic for Talking Therapies (opens in a new tab)