Avoid increasing risk when using AI in youth work

We recently ran a workshop with London youth workers about AI. What became clear very quickly is that frontline teams are stretched. When you’re dealing with safeguarding issues and limited funding, running a proper AI trial is rarely top of the list. AI use tends to happen informally – or more often not at all.

At the same time, the young people you work with are already using these new tools every day. They are asking questions, getting advice and sometimes treating chatbots like trusted adults. That puts safeguarding and efficiency at odds. AI can save time but it can also produce unreliable information and, in some cases, risky or harmful responses for vulnerable young people.

We don’t need an all-or-nothing approach here. Some AI use is sensible. Some of it is a safeguarding headache. The difference comes down to what you’re using it for and what data goes in. Let’s untangle that.

Like summarising notes, drafting messages, organising data, planning campaigns. That’s fine, but it comes with real security risks if it isn’t used correctly.

A lot of sector concerns around AI sit within misinformation and data breaches.

Charity staff worried about inaccurate outputs

75%

Charity staff worried about sensitive information being exposed (ThirdSector)

70%

The main thing to stick to is keeping sensitive information out of open tools. Notes about young people – case details, identifiable information. If it shouldn’t be public, it shouldn’t go in.

It’s important to decide collectively on how to use AI.

• Which tools are okay to use – ideally use general rules of selection rather than a fixed tools list. Don’t be too descriptive, at the current pace of development, be prepared for that to change every week!

TIP: don’t use any free tools. You always want to be able to disable ‘use your data for training’. These options are usually only available for paid accounts.

• What sort of information should never go in?

You don’t want to share any personal identifiable information. If you wouldn’t give the info anyone on the street, don’t put it in. Assume it could be made public.

• When is human input necessary?

Hint: always!

AI works best when it handles repetitive or complex tasks that can’t be easily automated with existing tools and don’t involve sensitive data. That way, your team gets time back without exposing case notes or breaking safeguarding rules.

Here are a few ways to use AI safely:

• Scheduling and reminders

AI can autofill session plans and track attendance. Your team still make the final calls, but AI cuts the tedious admin. However, when feeding it session info, ensure you omit all sensitive information.

• Summarising anonymous notes

Draft internal summaries for funders or reporting without exposing any personal details.

• Data checks and organisation

AI can spot duplicate records or inconsistencies, so staff don’t waste time chasing them. It’s also great at detecting patterns and normalising entries, e.g. if you have a long spreadsheet of entries that show relationships like ‘father’, ‘dad’, ‘parent’ – these can be unified easily with AI without exposing all your columns.

These tasks give staff more time to support young people directly and choosing the right tool for the job matters!

Many charities use digital tools that are simply referred to as “AI”. But not every tool does the same thing, and not every tool will help in the way they’re brought in for.

Generative tools – creating text, images or videos

Tools such as ChatGPT, Gemini, Claude or Canva – generative tools are typically considered the most common.

For youth workers, they can speed up drafting and formatting, especially reports, funder comms, campaign materials and social posts (but be careful with these, authenticity matters!).

Useful for saving time, but they shouldn’t be used to speak on your behalf. Generative AI tools do not always provide reliable information, and they can sometimes produce misleading or even dangerous results, particularly when vulnerable young people are involved.

Bear in mind that these chatbots don’t actually ‘know’ what they are saying. These tools are simply predicting the most likely next word. So make sure to check your output.

Classification tools – sorting information

Tools such as HubSpot AI, MonkeyLearn, AWS Comprehend and Zapier AI Classifier. These look at data and put it into categories.

For instance, messages or forms can be labelled automatically so staff can handle the important ones first (e.g. detect a safeguarding incident and alert the team via email or chat). They help organise work, but staff still act on it.

Search tools – finding meaning in information

Tools such as Algolia, Glean and Typesense.

These dig through documents or records to highlight what’s relevant. They save time hunting for details, but don’t decide what to do with that information or where to send it. That sits with human judgement.

Predictive tools – spotting patterns

Tools such as Salesforce Einstein, Microsoft Copilot, Blackbaud Target Analytics and HubSpot Predictive Lead Scoring.

These tools identify likely outcomes.

For example, one charity used a tool for a Christmas appeal to not send ‘more’ postcards, but predict which customer group ( based on location, previous donation history, income, etc.) is most likely to donate. They sent fewer but better targeted communications – doubling their fundraising income while saving money on reduced postage fees.

The insight from predictive tools means less guesswork when planning campaigns.

What’s important to consider is that the most value comes from tools that support decisions and highlight opportunities, not from those that simply generate words.

AI only earns its place when it’s used safely and under human oversight. In youth work, the repetitive admin is what often holds impact back.

That’s the drain you need to plug.

When those background tasks take less effort, your team can focus properly. More time and energy for young people. More headspace to stay present when it matters.