Free tool

The cost of therapy, in context

Weighing up private therapy against what the research says it gives back. Built by a practising psychologist, with the assumptions shown rather than hidden.

Looking for the actual fees?

What this tool does and does not do

The chance of improvement below comes from published research and applies to anyone. The money side does not: it models the impact of distress on paid work, so it only makes sense if you are working. It also describes costs that mostly fall on an employer rather than on your own pocket.

If you are retired, studying, out of work or caring for someone, ignore the money and look at the sessions figure. What therapy is worth to you was never a salary calculation anyway.

What brings you here?

Select one or more to personalise your cost estimate. Each adds evidence-based economic context to the calculation.

Your situation

Annual salary£42k

≈ £162 per working day

£20k£150k

Sick / lost days per year4 days

UK average 2024: 4.4 days per worker (ONS)

0 days30 days

Current stress level6 / 10

Modelled at 18% below your usual output (a modelling assumption, not a measurement)

1 / 1010 / 10

Therapy plan

Number of sessions12 sessions

Total cost at £80/session: £960

1 session26 sessions

Research note: ~50% of clients show meaningful improvement by session 8; ~75% by session 20–25. Saxon et al., 2017

Fees

Session	Length	Fee
Standard session and initial assessment	From 50 minutes	£80
Single session therapy	Up to 60 minutes	£100
Specialist treatment (EMDR, addiction)	Up to 90 minutes	£150
Reports and non-clinical work	Per hour	£100

Fees are in pounds sterling. In-person sessions at The Minster Centre cost more, because the room is hired on top of the session, so ask for the current figure before booking. The free 15-minute consultation has no cost and no obligation. Written therapy reports are quoted before the work starts.

Your numbers at a glance

Chance of meaningful improvement

58%

After 12 sessions, based on Howard et al. (1986), the study that established the dose-response curve in psychotherapy.

Estimated annual impact on work

£8,206

Employer-side. Sick days (£646) + reduced productivity (£7,560).

What therapy would cost

£960

12 sessions at £80/session.

What therapy could give back

£2,128

Therapy could recover ~58% of that estimated impact (£3,088). After paying for therapy, you're still ahead by £2,128.

Annual impact, estimated

Days off work

£646

4 days × £162/day

Working below your best

£7,560

18% reduced output

Estimated annual total£8,206

→ Therapy investment & return

12 sessions × £80

£960

Projected recovery value

£3,088

~58% improvement est.

Net return

£2,128

3.2× return on investment

Estimated annual impact£8,206

Days off plus reduced output

Therapy investment£960

Projected recovery value£3,088

Deloitte UK: avg £4–£5 return per £1 invested

Illustrative only , individual outcomes vary. Not a clinical or financial guarantee.

📈

Escalation risk: Moderate

Difficulty that persists tends to get harder to shift, and to spread into work and relationships. Starting earlier usually means a shorter piece of work.

83%

of systematic reviews report positive SST effects

415

clinical trials in the 2025 umbrella review

95%

client satisfaction in solution-focused SST trials

>50%

of people with mental health needs never access any treatment

What single-session therapy is , and isn't

Single-session therapy (SST) doesn't mean you only get one session. It means each session is designed to be complete and useful in its own right , so that if you only come once, you still leave with something meaningful.

This approach underpins how I work regardless of how many sessions we end up having together.

"If you spend the first session exclusively diagnosing, you've lost your opportunity to take advantage of the first and most powerful encounter." Hoyt et al., 2020 — Single Session Thinking

What 24 systematic reviews found

2025 umbrella review , Annual Reviews of Clinical Psychology (Schleider et al.)

AnxietyPositive evidence

DepressionPositive evidence

Substance use & alcohol misusePositive evidence

Eating difficultiesPositive evidence

Externalising problemsPositive evidence

Treatment engagement & uptakePositive evidence

Schleider et al. 2025 Kim et al. 2023

Effect sizes and what they mean

The pooled effect size across SST studies is SMD = −0.25 , a genuine, measurable improvement achieved in a single encounter. For depression and anxiety specifically, SST is most effective when embedded in community and primary care settings.

SST has positive evidence for:

Bereavement and complicated grief

Alcohol use and addiction

Trauma (brief, solution-focused formats)

Anxiety and phobias

Depression in adults and adolescents

Schleider et al. 2025 Kim et al. 2023 Talmon, 1990 — Single-Session Therapy

When one session is a starting point , not the whole story

SST works best when offered as a first step, with the option to continue. Research doesn't claim SST is always sufficient , particularly for complex, long-standing, or co-occurring difficulties.

"None of the reviews reported that SSIs showed a significant adverse effect on any mental health outcome , and several showed SSIs outperformed care-as-usual controls."

Schleider et al. 2025, Annual Reviews of Clinical Psychology

In my practice, we agree together on the right approach , whether that's a strategic one-off session, a focused six-session package, or a longer course. The evidence guides, but you decide.

Effect sizes: face-to-face vs. digital CBT for depression

Meta-analysis of 106 studies (N≈11,854). A standardised mean change (SMC) above 0.8 is considered large.

1.97

Face-to-face CBT

Standardised Mean Change

1.20

Digital / guided CBT

Standardised Mean Change

0.2–0.5

Typical app (unguided)

vs. waitlist , inflated

Face-to-face CBT completion82.4%

Digital CBT completion72.9%

Kambeitz-Ilankovic et al., 2022 — npj Digital Medicine Andersson et al., World Psychiatry 2014

The comparator problem

When platforms say "clinically proven," they rarely disclose the actual effect size , or that the app was compared to a waitlist, not a real therapist. This inflates apparent effectiveness significantly.

"Effective" can mean anything from lasting change to a small nudge that fades , and most apps sit at the small-nudge end.

Lecomte et al., 2020 — JMIR mHealth Abi Hana et al., 2024

Why apps are cheap , and what it costs you

Large platforms achieve low prices by cutting therapist pay , sometimes to gig-economy rates of £25–£55/session, requiring clinicians to see 40+ clients a week. High volume, low depth.

In independent practice, your fee pays for a fully present professional not under pressure to hit volume targets , exactly where research shows the benefit lies.

Clear Health Costs 2024 Harvard Gazette

The business model behind the apps

The question is not only whether an app works. It is what happens to therapy when it is delivered through a platform: therapists paid per session and rated like drivers, clients matched by algorithm, and the messy, uncomfortable part of the work quietly designed out.

Elizabeth Cotton, who worked as a psychotherapist in the NHS, calls this the uberisation of therapy and argues that the human therapist is the part that cannot be substituted. This practice is independent precisely for those reasons.

The framework I presented at the BPS conference for staying independent →

Elizabeth Cotton, UberTherapy: The New Business of Mental Health (Bristol University Press, 2025)

🤖

Why AI can't replace a therapist , and why that matters

I'm genuinely enthusiastic about AI , I use it in my research and stay current with the evidence. But there is a meaningful difference between AI as a useful tool and AI as therapy. Here's what the research actually says.

As of 2024, only 16% of LLM-based mental health chatbot studies had undergone clinical efficacy testing , meaning 84% haven't been tested to see if they actually reduce symptoms. The technology is moving fast; the clinical evidence is not keeping up.

Systematic review of 160 AI chatbot studies 2020–2024

🧠

AI cannot reliably detect crisis signals

In a Stanford study, multiple LLM therapy bots failed to recognise suicidal intent , one responded to a message about bridges and job loss by listing the heights of bridges. This is not an edge case; it is a core safety failure that human therapists are specifically trained to handle.

Moore et al., 2025 — arXiv:2504.18412

⚖️

AI shows bias toward certain conditions

The same Stanford research found AI chatbots expressed increased stigma toward users presenting with alcohol dependence and schizophrenia , even in models designed for therapy. This leads users to disengage from care precisely when they need it most.

Moore et al., 2025 — arXiv:2504.18412

📉

Effects are small and often don't last

Meta-analyses of therapy chatbots show small effects that are typically not sustained over time. Users also report frustration with responses that feel generic, repetitive, and disconnected from their actual experience.

Siddals et al., 2024 — generative AI chatbots in mental health (PMC)

🔒

No memory, no continuity, no real relationship

Meaningful change is built over time through a trusted relationship , what researchers call the therapeutic alliance. AI has no persistent memory, no ongoing knowledge of your situation, and no actual stake in your wellbeing. Each conversation starts from zero.

Pandya et al., 2023 — Frontiers in Human Dynamics (perspective)

🧩

AI cannot formulate or adapt

Effective therapy requires a clinical formulation , understanding why you feel what you feel, what maintains it, and what will actually shift it for you specifically. AI generates plausible-sounding responses based on patterns, not clinical reasoning about your individual situation.

Balan & Gumple, 2025 — ChatGPT clinical use in mental health: Scoping review (JMIR)

🔐

Privacy and data risks

Sharing sensitive mental health information with commercial AI platforms raises real data privacy concerns. Unlike therapy , which is legally confidential and regulated , AI conversations may be stored, used for model training, or exposed in a data breach.

Kalam et al., 2024 — ChatGPT and mental health: Friends or foes? (PMC)

🪞

AI tends to agree with you , and that can make things worse

Large language models are trained to produce responses that feel helpful and pleasant. In practice, this makes them sycophantic , they tend to validate what you say, affirm your perspective, and avoid saying things that might feel uncomfortable. This is a known, documented problem in AI systems.

In everyday use, this is mildly annoying. In a mental health context, it can be actively harmful. Many of the patterns that keep people stuck , avoidance, self-critical narratives, distorted thinking, unhelpful relationship dynamics , are exactly the things a good therapist will gently but persistently challenge. AI is structurally unlikely to do this well, because pushback feels less "helpful" to the user in the moment, even when it is the most therapeutic response.

✓Validates avoidance rather than exploring it

✓Agrees with self-critical thoughts instead of reframing them

✓Mirrors your framing rather than offering a new perspective

✓Avoids productive discomfort that drives real change

A real therapeutic relationship is built on honest, calibrated challenge. A skilled therapist knows when to validate, when to sit with discomfort, and when to gently push back , because growth rarely happens inside a comfort zone. That balance requires human judgement, clinical training, and a genuine relationship. It cannot be replicated by a system optimised to make you feel good in the moment.

Sharma et al. 2023: Sycophancy in LLMs Montemayor et al., 2024 — Can LLMs replace therapists? (JMIR AI)

What AI can usefully do , and where it fits

AI is not useless. It can be a helpful supplement for specific, bounded tasks. The evidence simply does not support it as a standalone substitute for a trained therapist , particularly for anything beyond mild, transient stress.

Where AI can help

Psychoeducation , learning about conditions

Journaling prompts and mood tracking

Practising coping skills between sessions

Finding and accessing information

Low-level stress and daily wellbeing support

Where AI falls short

Trauma processing and PTSD

Crisis support and suicidal ideation

Addiction and complex behavioural patterns

Grief and bereavement

Anything requiring ongoing formulation

Diagnosis and clinical assessment

My position: I encourage clients to use AI tools alongside therapy when they're helpful , for journaling, between-session reflection, or learning. But for the work itself, there is currently no credible evidence that a language model can replicate what happens in a genuine therapeutic relationship.

Ready to explore whether therapy is right for you?

A 15-minute free consultation , no commitment. We'll talk about what you're facing and whether working together makes sense.

Book a free 15-minute consultation →

This tool is for illustration only, based on published economic and clinical research. Not a clinical prediction or financial guarantee. Individual outcomes vary considerably.

Where these numbers come from

This is the part most calculators leave out. The return figure draws on Deloitte UK's work on employer mental health spending, which is about organisations rather than individuals, so treat it as background and not as a prediction about you. Session ranges follow the usual length of a course of CBT.

What this tool will not do is tell you what your own difficulty is costing you as a fact. That would depend on assumptions about your salary, your working pattern and your health that no web page can know. The figures here are a model, and the inputs are yours to change.

See fees