AI Tools for Charity Speaker Screening: A Balanced Guide

Why speaker due diligence has become unavoidable for trustees

Charities that host public events, conferences, or community gatherings are increasingly expected to know who they are putting on a platform, and why. The Charity Commission for England and Wales has made this explicit in a series of recent regulatory actions. In May 2025, it directed the Islamic Centre of England, under Section 84 of the Charities Act 2011, to provide 'rigorous oversight of future speakers and online activity' following years of non-compliance after events eulogising a UK-sanctioned individual. In the same year it issued Official Warnings to the Mosque and Islamic Centre of Brent and the Central Oxford Mosque Society, finding that both charities lacked effective speaker-management policies, and that some speeches reviewed contained content the Commission judged 'inflammatory and divisive'.

These are not isolated examples. In 2016, the Commission's inquiry into Hindu Swayamsevak Sangh UK concluded that trustees had 'failed to follow their own procedures and had not properly screened speakers' after a documentary filmed objectionable remarks made to young beneficiaries at one of its camps. The common thread across these cases is not malice, but the absence of a documented, consistently applied process for assessing who is invited to speak, and why.

The Commission's own compliance guidance (Chapter 5 of the Protecting Charities from Harm toolkit) sets out what good practice looks like: a clear risk-assessment policy for inviting speakers, defined criteria for flagging concerns, checks against the Home Office list of proscribed organisations and the OFSI sanctions list, and a documented rationale for decisions, particularly where a speaker is controversial. This is the backdrop against which AI-assisted screening tools have started to find a role.

What AI-assisted screening can genuinely help with

Manually researching a speaker's public record, media coverage, and affiliations is time-consuming, and quality varies enormously depending on who does it and how much time they have. AI research tools can help by pulling together publicly available information quickly and consistently, giving trustees or event organisers a structured starting point rather than a blank page.

• Speed and consistency: an AI tool can search and summarise public reporting, statements, and affiliations in minutes, applying the same checklist to every speaker rather than an ad hoc effort that depends on who is doing the vetting that week.
• Structured output: a well-designed tool returns findings against defined categories (for example, extremist affiliations, sanctions exposure, financial or legal concerns, public statements) rather than an unstructured pile of search results.
• An audit trail: because the process is repeatable and produces a record, it directly supports the Commission's expectation that speaker decisions be documented and explainable after the fact.
• Reducing obvious oversights: a tool that checks a name against proscribed organisation lists and sanctions registers as a matter of routine reduces the risk of a basic check simply being forgotten under time pressure.

Where AI tools fall short

None of this makes AI research a substitute for judgement. Large language models generate research summaries by searching and synthesising public material, and that process has real limits that trustees should understand before relying on it.

• Coverage gaps: a speaker with a limited English-language public footprint, or whose activity is reported mainly in other languages or on platforms not well indexed by search, may return a falsely reassuring 'no concerns found'.
• Context and nuance: identifying that someone shared a platform with a controversial figure, or that a quote has been reported second-hand, requires judgement about materiality and context that automated summarisation does not reliably supply.
• Errors and fabrication: AI models can misattribute quotes, conflate two similarly named individuals, or state things with more confidence than the underlying sources support. Any finding that would materially affect a decision should be traced back to its original source before being relied upon.
• No accountability: a tool cannot be the 'decision-maker' of record. Charity law places the due diligence duty on trustees, and a regulator reviewing a decision will ask what the trustees considered and concluded, not what a piece of software output.

Data protection considerations

Screening a named individual, even using only public information, involves processing personal data, and in some cases special category data (for example, information touching on someone's religious or political views, or alleged criminal conduct) under the UK GDPR. Charities should be able to point to a lawful basis for the processing, typically legitimate interests given the safeguarding and reputational purpose, and should be able to justify that the checks are proportionate to the risk posed by the event and the speaker's role.

Practical steps worth building into a screening process include: retaining records only as long as needed to support the decision and any later review, being clear with trustees and staff about what data an AI tool sends to third-party providers and where it is processed, and avoiding the use of screening tools for purposes beyond the stated one (routine due diligence, not general monitoring of individuals). Where a charity uses an external platform to run these checks, its data processing terms and retention practices should be reviewed as part of the charity's own compliance obligations, in the same way any other supplier arrangement handling personal data would be.

Keeping a human in the loop

The regulatory cases above share a structural weakness: an absence of any documented process at all, rather than a bad outcome from a careful one. An AI tool addresses that gap by making a first-pass check routine and consistent. It does not address the second requirement implicit in the Commission's guidance, that trustees exercise judgement about what the findings mean for a specific event, audience, and charitable purpose.

A workable model is to treat the AI output as an input to a decision, not the decision itself. That means a named person reviews the findings, considers whether anything needs independent verification, weighs proportionality (a routine community talk carries different risk from a large public conference), and records the reasoning, especially where the speaker or topic is at all sensitive. Where a tool flags a serious concern, such as an apparent conviction, extremist affiliation, or sanctions match, that should trigger escalation to a trustee decision rather than being treated as a pass/fail result on its own.

A sensible way to bring AI into the process

Charities considering an AI-assisted approach are generally better served by treating it as one stage of a wider policy, alongside a written speaker-approval process, defined escalation routes for flagged concerns, and periodic review of how the policy is working in practice, rather than adopting a tool in isolation and assuming the due diligence obligation is thereby discharged.

Structured screening tools such as CharityScreen can support this process by giving trustees a consistent, documented starting point for speaker research, alongside, not instead of, their own judgement.