How UK charities are using AI to work more efficiently.
Four in five UK charities now use AI routinely. Almost none of their boards know it. This is what the adoption data actually says, where the time savings genuinely come from, which organisations have produced real numbers, and why governance is now the sector's biggest exposure.
Adoption is near universal. Strategy is rare.
AI has stopped being a question of "if" for UK charities. The 2026 Charity Digital Skills Report, the sector's largest annual survey with a record 807 responses, found 79% of charities now routinely using AI, up from 76% in 2025 and 61% in 2024. Small charities made up 71% of respondents, so this is not simply a story about big organisations with big budgets.
The gains are real. They are also narrower than the headline suggests. Time savings sit overwhelmingly in admin, drafting and grant applications, and only a small number of organisations have achieved anything genuinely transformative in service delivery.
Three years of adoption
Charities using AI. Source: Charity Digital Skills Report, 2024 to 2026.
The more useful question is not whether charities use AI but how deeply. Here the picture is sobering. CAST's own survey found that of organisations using AI, 50% were merely exploratory, 32% actively using tools, and just 10% using AI extensively and strategically. Charity Excellence's April 2026 Future Charity Report, drawing on surveys of 220+ charities and 205 grantmakers, found more than 60% still at early stages and fewer than 3% with AI fully embedded.
Put plainly, most "AI adoption" in the UK charity sector is one member of staff using ChatGPT or Microsoft Copilot at their desk to draft an email, tidy a report or summarise a meeting. That is not nothing, and it is where most of the aggregate time saving currently comes from, but it is a long way from organisational change.
"79% uptake in AI, but one in three unable to use it effectively."
Will Ranjan-Churchill, Technology Lead, Esmée Fairbairn Foundation, on why the sector would have acted by now if this were any other tool (Charity Digital Skills Report 2026)
Where the efficiency actually comes from.
Admin dominates. Fundraising is close behind and growing fastest. Service delivery remains the smallest category and the most consequential.
Admin and internal operations
The workhorse category and the source of most aggregate time saved: meeting transcription and minutes, email triage and drafting, document summarisation, case recording, and data extraction from paper registers into digital records.
Fundraising and grant applications
First-draft bid writing grounded in the charity's own outcome data, reshaping one dataset into multiple funder report formats, donor research and segmentation, personalised acknowledgements, and propensity modelling for legacies and major gifts.
Service delivery and advice
Historically the smallest category (and only 5% offered services built on AI tools), yet where the largest efficiency multipliers have been demonstrated. Adviser copilots, triage chatbots, eligibility checking and accessibility conversion.
Comms, content and research
Social posts, newsletters, appeal copy and website updates. Low risk, quick to adopt, and the most common entry point. The counter-current is AI-generated imagery, which many charities trialled and then abandoned.
Impact and evaluation
Analysing qualitative feedback at scale, generating case studies from recorded conversations with consent, and predictive demand planning. The prerequisite most small charities do not meet is reasonably clean historical data.
Research and science
Fact-checking at national scale, conservation imagery triage, and clinical prediction. Small in volume, but this is where a handful of charities are doing work that would be impossible by hand.
The same picture, with risk attached
| Function | Uptake | What it looks like in practice | Risk |
|---|---|---|---|
| Admin & project management | 63% | Transcription, summarisation, email, case notes, data entry | Low to medium |
| Grant fundraising | 45% | Bid drafting, funder reporting, donor research | Medium |
| Comms & content | High, informal | Social posts, newsletters, appeal copy | Low text, high imagery |
| Impact & evaluation | Common | Qualitative analysis, case studies, demand forecasting | Medium |
| Service delivery | ~15% | Adviser copilots, triage chatbots, accessibility conversion | High |
| Research & science | Niche | Fact-checking, conservation imagery, clinical prediction | Varies |
The organisations with real numbers.
Published UK charity case studies with credible efficiency figures are rarer than the volume of AI commentary suggests. These are the strongest currently in the public domain.
Built in Greater Manchester after a post-pandemic intake of trainee advisers overwhelmed the supervisors validating their answers. Caddy is restricted to GOV.UK, AdviserNet and the charity's own knowledge base, drafts an answer with citations, and a supervisor approves before the adviser speaks to the client. Clients always speak to a human. The trial covered 1,000+ interactions: 80% of responses were good enough to pass on unedited, advisers were twice as likely to feel confident and 1.5 times more likely to resolve the issue. Now live in 40+ offices with a 90% adviser approval rating and 70+ more waiting.
RNIB converts prescriptions, bank statements, mortgage applications and welfare information into braille, large print and audio. Every document type used to need its own workstream, custom code and error-fixing. Mailings converts documents into an internal markup format automatically, an RNIB employee checks accuracy, then it goes to publishing and braille embossing. Azure AI Neural Voice replaced the previous robotic audio. The freed capacity let RNIB take on more clients and earn more income.
Citizens Advice must record every client interaction as a case note, both for insurance and for continuity between advisers. CASORT piloted a second tool that transcribes the call and generates the note from the charity's own templates, so advisers review a draft instead of writing from scratch. The charity is now exploring monetising the model it built as an income stream, which is a rare example of a charity turning internal AI work into revenue.
Representing over 200 hospices, Hospice UK tested Copilot and CRM AI for eight months. The biggest win came from ordinary daily tasks: email replies, summarising bulk email after leave, and Teams recordings with action recaps. Clinical teams in member hospices report AI as invaluable for summarising documents and logging patient interaction notes. Business Systems Manager Phil Grace attributes the sector's lag to a knowledge gap and to ethical concerns, and concludes that rollout depends on role-specific relevance rather than generic training.
The West of England Centre for Inclusive Living, a user-led disabled people's organisation in Bristol, built a chatbot that behaves like an Easy Read document: short jargon-free sentences with clear images and visual navigation options. CEO Dominic Ellison notes it makes the whole website more navigable for people who are learning disabled or neurodivergent. This is AI deployed explicitly to remove access barriers, designed by an organisation led by the people it serves.
Camera traps generate vast image volumes, and up to 90% are false triggers with no animal in frame. Reviewing them at two seconds an image would take one person around four years. WWF worked with other conservation organisations and Google to build Wildlife Insights, which filters blanks, identifies species and shares images with researchers. After the 2020 Australian bushfires, ten scientists studied more than 28 million acres, identifying over 150 species among seven million images.
Three more worth knowing about
Full Fact
Uses a large language model to monitor news sites, live TV speech, podcasts and social media, identify checkable claims and match them against claims already checked. Human journalists verify offline. The AI does triage at a volume no team could cover manually; humans keep the judgement.
Prostate Cancer UK & Movember
Jointly funded £1.2 million towards UCL-led research using ArteraAI's multimodal AI, which reads multiple biopsy images together to predict the best treatment combination for a specific patient. One in eight men will get prostate cancer, a disease that had lagged others in precision medicine.
Great Ormond Street Hospital Charity
Piloting TORTUS, which listens to outpatient consultations and drafts clinic notes and letters so clinicians get more face-to-face time with patients. A separate partnership with Roche UK applies AI to routinely collected data to speed up treatment for rare and complex diseases.
What "efficiency" really means here.
Most charities are saving minutes rather than restructuring operations, and the savings are not always as durable as they look on a slide.
Research by the Joseph Rowntree Foundation and We and AI found that short-term efficiency gains are not always sustainable, because time is still needed to monitor outputs, manage compliance and trust issues, and adapt to new tools. A twenty-minute drafting saving can be eaten entirely by the review cycle it creates, particularly where accuracy is non-negotiable. The Charity Commission puts the same point more positively: where AI frees up time in manual and administrative processes, it may make more time available for direct work with service users. That conditional is doing real work in the sentence.
The binding constraint is no longer tooling. It is skills. 56% of charities cite limited skills as their biggest barrier and 45% point to a lack of training and support. Building staff and leadership capability is now the top priority for 59% of charities and 72% of larger ones, up from 43% a year earlier. Training tops the funding wishlist for 44%, and 60% say sector-wide AI training is essential to their future.
CAST's data on the support gap is starker. 47% said their organisation provides little or no AI training or support, rating it 0 or 1 out of 5, with just 4% providing full support. Only 18% felt there was sufficient AI upskilling support available to the sector as a whole. Meanwhile distrust is growing rather than fading: charities reporting a lack of trust in AI tools rose to 35%, up from 15% the previous year.
Access to digital funding from trusts and foundations has fallen to 17%, down from 30%, and it hits small charities hardest. This is happening at exactly the moment AI capability is becoming a competitive advantage in grant applications. Only 36% of small charities have an AI policy. The divide in the sector now runs through funding rather than enthusiasm.
The people charities exist to serve
One number in this research is harder to defend than any other. Only 6% of charities have researched how the people they serve actually use AI. CAST found 25% now hearing from beneficiaries about AI's effects, a big rise from 7.5% a year earlier, but still a minority. Reported concerns include job displacement, low confidence, safeguarding, and practical harm from AI-generated misinformation such as incorrect opening times.
The equity picture matters too. 52% of charities support beneficiaries with digital inclusion, often informally. Charities led by and for disabled, deaf, neurodivergent and global majority communities are driving the accessibility work while facing disproportionate barriers to funding and infrastructure. 22% say their digital services are not reaching diverse and marginalised communities, and 27% rarely co-design services with users.
The boards do not know.
This is the sector's defining AI problem, and the one most likely to turn into an incident.
Set the two numbers side by side. Between 76% and 79% of charities are using AI. The Charity Commission's Trust in Charities research found that just 3% of trustees said their charity was using AI at all, rising only to 8% among larger charities. Charity Excellence calls this "a significant gap between what is happening operationally and what boards believe is happening." Trustees remain legally responsible either way: the Commission has been explicit that they carry responsibility for AI-related risks including data protection, safeguarding, bias and decision-making.
- Strategic assessment of AI's impact on the organisation
- Clear trustee or committee responsibility for AI
- Organisation-wide training and compliance
42% use personal accounts
CAST's most uncomfortable finding. 42% of respondents use a personal AI account for work, and of those, 47% said AI use was either not permitted or they were unsure of the policy. That is potentially sensitive beneficiary data flowing through consumer accounts with no data processing agreement in place.
Written, then ignored
Coverage has improved: 52% of organisations have an AI policy and 30% are developing one, against 16% and 31% a year earlier. But only 27% of those with a policy refer to it regularly, 39% rarely do, and 5% do not find it helpful. A policy nobody opens is not a control.
The skills are not there
A third of charity boards and a quarter of CEOs are rated poor on AI skills, and almost half of organisations lack a trustee with relevant digital expertise. Data protection is the number one concern sector-wide: the biggest adoption barrier at 38% and the most pressing support need at 69%.
Permission is conditional, and the conditions are specific.
Charity Tracker's UK-wide survey of 3,000 adults found 36% feel positive about charities using AI, 27% negative and 37% unsure. Most people are still forming a view, which means the sector's behaviour now will shape it. The conditions attached to that permission are sharp. Trust is highest where AI protects charitable funds through fraud detection or improves back-office efficiency. It drops steeply when AI influences decisions about people: 38% said it was unacceptable for charities to use AI to decide who gets help. Only 13% are comfortable with sensitive personal data being used in AI systems.
The Charities Aid Foundation surveyed 6,102 people across ten countries with additional UK focus groups. Globally, 37% focused on the opportunities and 22% on the risks, a net positive outlook of +15%. The UK's net score was only +5%, among the most sceptical in the sample. Kenya scored +44%. Australia was the only net-negative country at −4%.
Generosity tracks positivity. Net positivity rises from +5% among non-donors to around +30% among high donors. The people who fund you are the people most comfortable with AI. They are also watching: only 13% said they would pay little or no attention to what a charity says publicly about its AI use, and 77% of high donors pay "a fair amount" or "a great deal" of attention.
The imagery trap
Both charities and the public are uneasy about AI-generated imagery. Charity Excellence found many charities tried it and then stopped, citing ethical, reputational or authenticity concerns. Research including work at the University of East Anglia points to strong public support for authentic imagery and much lower acceptance where AI images look realistic or emotionally manipulative, particularly in sensitive contexts. For a sector whose currency is credibility, synthetic imagery of people who do not exist is a poor trade.
What the regulators have said.
Nobody is banning anything. The consistent message across all four is accountability, proportionate human oversight, and transparency scaled to risk.
Charity Commission: trustee guidance
Charities and Artificial Intelligence confirms trustees remain legally responsible for AI-related risks, covering data protection, safeguarding, bias and decision-making.
National Lottery Community Fund
Confirms it will not reject AI-assisted applications, publishes 10 AI Principles and applicant guidance, and urges charities to personalise AI-drafted content so it reflects their own community.
Charity Governance Code refreshed
A new eighth principle brings, for the first time, an explicit recommendation that charities have a policy for the use of technology and AI tools.
Fundraising Regulator: first guidance of its kind
A life-cycle approach covering trustee accountability including for third-party AI use, proportionate human oversight to check accuracy, fairness and legality, and transparency scaled to the risk of misleading donors.
"The Fundraising Regulator is neutral on whether charities should use AI. What matters is that when AI is used it must be in a legal, open, honest and respectful way, in accordance with the code."
Kieron James, Board member, Fundraising Regulator, December 2025
What the sector is asking for.
CAST's open-response data has been consistent across two years, and it is not a request for shinier tools.
- Fund training and consultancy, including protected time away from duties to learn and experiment.
- Treat AI licences as legitimate operational cost and include them in full cost recovery.
- Fund dedicated roles and exploratory time, such as AI coordinators, in-house or shared across the sector.
- Share guidance and experience. Access to peers and case studies have been the top two support needs two years running.
- Be transparent about AI use, including funders' own policies on whether AI can be used in applications.
- Fund the foundations, not just the tech: data governance, secure systems and digital maturity.
Free infrastructure that already exists
Courses, hub and a sector-owned tool
Free self-serve AI courses, an AI Hub of guides and templates, a Library of 30+ documented AI experiments from real charities, and a Digital Leads Network of 500+. Worthwhile Chat, a sector-owned generative AI tool built with The Developer Society around real charity tasks, is in testing for a Summer 2026 release, with Deloitte funding 150 free six-month places for small and medium charities.
Trustee training and policies
Free trustee and management AI courses, 60+ downloadable policies updated for AI, and the AI Ready programme. If your board has never had the conversation, this is the cheapest possible starting point.
Open source, and reusable
Caddy is open source and available for other advice providers and helplines to build on. It uses retrieval-augmented generation, a standard and comparatively cheap technique. The prototype took about four weeks; the robust version took roughly twelve months, largely on volunteer time with Cabinet Office support.
Eight things that actually work.
Drawn from the case studies and survey data above rather than from vendor claims.
Start where work is repetitive and stakes are low
Transcription, meeting notes, case recording, data entry from paper. Every credible case study starts here. Visible time savings, manageable risk, and the tasks where AI is genuinely good rather than merely plausible.
Constrain the model before it goes near a beneficiary
Caddy is the reference implementation: retrieval over verified sources only, visible citations for every answer, a supervisor in the loop. This is standard technique, not bespoke development.
Keep humans making decisions about people
38% of the public find it unacceptable for charities to use AI to decide who receives support, and that is exactly where bias, safeguarding and regulatory exposure concentrate. Use AI to prepare information for a human decision, never to make it.
Close the shadow AI gap deliberately
Assume staff are already using AI, because 42% are using personal accounts. The fix is a short, usable, actively communicated policy plus a sanctioned tool. Prohibition just moves the activity out of sight.
Get the board genuinely engaged, not merely informed
Named trustee or committee responsibility, a strategic assessment, and organisation-wide training. Those are the three controls the sector is currently failing, and trustees are legally accountable for risks they largely do not know they are carrying.
Budget for the review cycle, not just the licence
Monitoring outputs, managing compliance and adapting tools consume the time AI appears to save. Measure net time saved after review, not gross. This is the single most common reason a pilot's numbers do not survive contact with business as usual.
Ask the people you serve
Only 6% of charities have researched how their communities use AI. This is cheap to fix and the most direct route to knowing whether a deployment will help or exclude the people it is meant to reach.
Be transparent, particularly about imagery
The Fundraising Regulator's test is proportionate: the greater the risk of misleading donors, the more transparent you should be. Synthetic imagery of beneficiaries fails that test in most sensitive contexts.
The barrier is not the technology
The step change happens only where organisations built a purpose-specific tool around a well-defined bottleneck, kept humans in the decision loop, and had the support to do it properly. Caddy and RNIB's Mailings both fit, and both produced roughly 50% to 95% reductions in process time. What is missing elsewhere is skills, funding and governance capacity.
Sources & further reading.
- Charity Digital Skills Report 2026, Zoe Amar Digital and Think Social Tech (807 responses; the 79%, 63%, 45%, 56% and 17% figures). See also UK Fundraising's summary.
- CAST's AI survey 2026: all the results (the 89%/92% adoption, 42% shadow AI, policy usage and support-gap figures).
- Charity Excellence, Future Charity Report: AI in the Charity Sector 2026 (embedded-AI, board-control Red ratings, imagery findings).
- Charity Commission, Charities and Artificial Intelligence: trustee guidance, April 2024, and Public Trust in Charities (the 3% trustee figure).
- Fundraising Regulator, Guidance on using artificial intelligence in fundraising, December 2025.
- Caddy, Incubator for AI (i.AI) and Charity Digital, AI and the future of service delivery (Caddy and RNIB Mailings efficiency figures).
- Charity Times, Lessons from Hospice UK's AI journey, August 2025.
- Charity Digital / DSC, How charities are using AI in service delivery (WECIL, Prostate Cancer UK & Movember, WWF, GOSH, Full Fact).
- Charities Aid Foundation, What the public think of charities using AI (6,102 respondents across ten countries), and Charity Tracker's UK survey of 3,000 adults.
- Joseph Rowntree Foundation and We and AI, Grassroots and non-profit perspectives on generative AI.
- CAST AI resources, Library of experiments and Worthwhile Chat, and The National Lottery Community Fund's £3m programme.
Working in the charity sector?
We help charities pick the use cases that genuinely save time, deploy them safely with beneficiary data, and give trustees the governance and evidence they are accountable for. Practical, proportionate, and sized for the sector's budgets.