As AI transforms social impact measurement, governance is in short supply

The national network for practitioners will soon release AI use principles after survey respondents nominated ethical and responsible use guidance as their most urgent needs.

The 2026 Pulse Survey – produced in collaboration with consultancy Social Impax – drew 75 responses from the not-for-profit, consultancy and government sectors. Around half of all respondents used AI regularly, largely relying on general-purpose tools such as ChatGPT, Microsoft Copilot, Claude and Gemini.

Ninety per cent of AI users reported time saving as the major benefit, and the most popular uses were general research, recording and notetaking, and drafting reports and summaries,

The biggest barriers to adoption were related to data privacy and ethical concerns, above skills gaps and costs. The top risks nominated by respondents were inaccuracy and unreliability, followed by misuse and over-reliance, data security, and concerns about bias and unfairness.

Despite these concerns, the use of formal safeguards is not widespread. While 71 per cent of respondents said they checked AI outputs through human review, fewer had implemented AI policies (52 per cent), informed consent processes (30 per cent), risk assessments (11 per cent), or formal ethics reviews (9 per cent). Just over half of respondents (54 per cent) disclosed to clients that AI had been used in their work.

The study’s authors suggested that AI use presented an opportunity for practitioners to boost fairness and equity, rather than simply save time, and they listed priorities for practitioners as:

• validation of AI outputs
• informed consent processes
• privacy measures
• AI use disclosure
• consideration of broader AI use impacts, including social impact and society.

"The biggest theme area was about ethical and responsible use guidance in this space," said Rebecca Roebuck of Social Impax, who presented the findings at a SIMNA webinar this month.

The survey also showed that a significant proportion of practitioners lack confidence in using AI tools. Around 60 per cent of respondents described themselves as only somewhat confident using AI, despite adoption rates that suggest the tool is already embedded in practice.

Roebuck raised a longer-term concern: whether practitioners, particularly less experienced ones, are developing the judgement required to assess AI output.

"One of the things that worries me about AI use [is] whether they will have the judgement to be able to review an AI output that's technical and understand whether it's good or not good. I think that's a real risk area for future practice."

More information

Examine the study in full