When AI has an answer for everything, who’s left to disagree?

We can all see that Artificial Intelligence (AI) is useful. It can synthesise information, generate ideas, overcome the blank page and turn rough thinking into something coherent. For charities working with limited time and resources, that efficiency is understandably attractive.

But good thinking needs more than answers. It needs disagreement and people willing to challenge assumptions, interrogate evidence and offer a genuinely different perspective. If AI helps generate the idea, develop the strategy, write the copy and then critique the result, we risk outsourcing not just the work, but some of the critical thinking around it.

For charities, that raises another question as fundraising relies on trust, empathy and human connection. As AI becomes more involved in creating those communications, we need to think carefully about where it adds value, and where human judgement, scrutiny and disagreement are worth protecting.

 

Good thinking needs a critical friend

In strategy and creative work, there is enormous value in having a “critical friend”: someone who understands what you are trying to achieve and wants the work to succeed, but is independent enough to challenge how you are getting there. A critical friend questions assumptions, interrogates the evidence and introduces perspectives that may not have occurred to the people closest to the work.

Sometimes that means asking whether the evidence actually supports the strategy. Sometimes it means questioning whether the organisation is solving the right problem. And sometimes it means being willing to tell a room full of people who have spent days developing an idea that the idea simply isn’t very good.

Good strategy rarely emerges from uninterrupted agreement. Neither does genuinely interesting creative work. Ideas improve when they encounter different experiences, disciplines, interpretations and opinions. We defend them, modify them or abandon them altogether.

That is also part of the value of bringing outside perspectives into the process. Whether it comes from an agency, a colleague or someone with entirely different expertise, a critical friend brings something another iteration of the same thinking cannot: distance. They aren’t simply there to execute the work, but to question the brief, challenge the assumptions behind it and suggest when there might be a better way.

Disagreement, then, isn’t an unfortunate part of the creative process that technology might eventually eliminate. It is part of the process itself. And that matters when we start asking AI to play the role of critical friend, because AI has a documented problem with agreeableness.

 

AI doesn’t necessarily challenge the person asking the question

Researchers use the term “sycophancy” to describe the tendency of AI systems to produce responses that align with a user’s expressed beliefs or preferences rather than consistently prioritising accuracy. Research into sycophancy in language models found the behaviour across five AI assistants and multiple text-generation tasks. The researchers also identified a particularly human complication: people themselves were more likely to prefer responses that matched their own views.

This creates a problem when AI becomes not only the tool that helps develop an idea, but the tool we ask to assess it.

Give generative AI a strategic direction and it can help construct the case for it. Ask it to strengthen the argument and it can. But challenge the original premise and the same system may construct an equally persuasive argument for another direction. That flexibility makes AI enormously useful. But flexibility isn’t independence.

 

We risk mistaking iteration for interrogation

Consider how easily AI can now sit across an entire creative process. It can analyse research, identify themes, generate strategic territories, develop a proposition, draft campaign copy and edit the finished work. It can then review that work against the strategy and recommend further improvements.

There may be dozens of iterations within that process. Yet iteration and interrogation are not the same thing.

A 2025 Microsoft Research study into generative AI and critical thinking examined 319 knowledge workers and 936 real-world examples of AI use. It found that greater confidence in AI was associated with less critical thinking, while greater confidence in one’s own abilities was associated with more. The researchers also found that AI shifted human effort towards verifying information, integrating responses and overseeing tasks.

That doesn’t mean AI makes us less intelligent. It means that when technology increasingly produces the first answer, our role increasingly becomes deciding whether that answer deserves to be accepted. That requires judgement.

 

AI makes media literacy more important, not less

For years, media literacy has encouraged us to interrogate the information we consume. Who created this? Why? What evidence supports it? What perspective is missing? Can it be independently verified? Those questions belong in an AI-enabled workplace too.

AI-generated information can be articulate, confident and convincing. None of those qualities guarantees that it is accurate, insightful or appropriate. The danger is that fluency can make scrutiny feel less necessary. Recent Anthropic research into patterns of human reliance on AI examined approximately 1.5 million conversations. While potentially severe cases were rare, researchers observed interactions in which users repeatedly sought guidance and accepted AI outputs with minimal challenge. They concluded that user education is an important part of helping people recognise when they are ceding judgement to AI.

AI literacy therefore cannot simply mean knowing how to write a better prompt. It must also mean knowing when not to trust the answer: checking sources, recognising assumptions, seeking competing interpretations and distinguishing evidence from something that merely sounds authoritative. The better AI becomes at producing convincing answers, the more important our ability to question those answers becomes.

 

Fundraising makes the ethical question harder

For charities, there is another layer to this conversation. Fundraising is fundamentally relational. Behind an appeal is a person, family, community, researcher, frontline worker, animal, environment or cause. On the other side is another human being being asked to care enough to act.

The communication between them isn’t simply transmitting information. It is trying to establish trust, empathy and shared purpose. So what happens when that communication is increasingly generated by a machine?

There is an important distinction here. Using AI to transcribe an interview is not the same as asking AI to create the emotional framing of someone’s story. Using it to summarise research isn’t the same as asking it to determine which parts of someone’s lived experience will be most persuasive. Proofreading is different from creating a person’s voice.

Treating all AI use as ethically equivalent is too simplistic. But treating efficiency as the only consideration is equally inadequate. Imagine reading a fundraising appeal that moves you. Someone’s story feels sensitively told. The words feel personal. You understand why your help matters. Then you discover the appeal was generated by AI. Does knowing that change anything?

Perhaps not. But perhaps the emotional connection suddenly feels different. Perhaps you wonder how much of the story represents the person whose experience is being shared and how much has been optimised to make you give.

The point isn’t that there is one correct reaction. It is that the question exists at all. For charities, trust is an asset built over years. If the use of AI can alter how donors perceive the authenticity or humanity of an organisation’s communication, then its use isn’t simply an operational decision about saving time. It is a strategic and ethical decision too.

“Can AI do this?” is not a sufficient question. We also need to ask whether it should.

 

AI can be a voice in the room, not the room itself

None of this requires charities to reject AI. There are valuable applications for it across research, administration, analysis and ideation. It can help us get started, organise information and explore possibilities. But a starting point is different from an endpoint.

AI-generated information should still be interrogated. Important claims should be independently verified. Ideas should encounter people who weren’t involved in creating them. Lived experience should never become simply another input to optimise. Organisations should consider whether they would be comfortable explaining their use of AI to the people whose stories they tell and the donors whose support they seek.

Most importantly, we need to protect disagreement.

The biggest risk may not be that AI replaces the strategist, fundraiser, copywriter or creative. It may be that we gradually allow it to replace the friction between them — and outsource more of our judgement along the way. The value of a critical friend is not that they always have the right answer. It is that they give us another perspective against which to test our own. They challenge us to explain why we believe something, interrogate what we might have missed and sometimes change our minds.

AI can be incredibly useful in that process. But it shouldn’t become the process. Because if AI is increasingly giving us the answers, we need to make sure there are still people in the room willing to disagree with them, and with us.