If you changed the name, would anyone notice?

The temptation is real. Let AI do the heavy lifting and move on. But let me ask you this - if you changed the name of the programme participant or your organisation in your last AI-assisted communication, would anyone notice? 

In social impact communications, that question matters more than anywhere else.

Here are the pitfalls I have learned to avoid and the expertise I bring as a human communicator into the work.

Steer clear of the AI-processed polished slop 

The communities and individuals at the centre of social impact work have real stories, complexity and dignity. When their experiences get processed through AI into smooth, calibrated content, something important is lost. And increasingly, audiences can feel it even when they can't name it. AI-written drafts of real stories are often so polished that they become interchangeable. Change the name of the person or organisation, and nothing distinguishes one from another.

I have found that incorporating real conversations, reflective observations from the field, unpublished data, and in-depth work with communities into narratives creates the backbone of storytelling. AI cannot generate this. It can only sharpen it.

Start with specificity, not AI's generalisations

Instead of 'Skilling programmes transform lives', try 'When 21-year-old Maria was asked whether she could meet the target of 1 crore billing, she unabashedly responded, “I will do whatever it takes!”'

AI defaults to the general because it is trained on a wide range of data. 

I can bring in the advantage of specificity - a specific person, a time, a place or a feeling, a detail nobody else notices. The more particular the detail, the more the resonance, paradoxically. For this story, a conversation with Maria, a participant in the livelihood programme, brought out her unabashed confidence.

Never let AI's first draft define your voice

No matter how good I think my context-rich engineered prompt is, AI generates the most statistically probable version of whatever I asked for. It tends to use homogenised words repeatedly - transform, delve, robust, navigate, ecosystem, impactful, shaping.

I use the first AI draft to find holes in the argument it's making, suggest alternative structures, or generate options that I can react to. I give it access to sentence examples, words to avoid, and introduce organisation stances as a filter against each AI-assisted output. This allows each communication product to showcase the organisation’s genuine voice.

Never release unchecked AI summaries and generated framing

While AI does a good job of summarising complex research, it can also drift toward a pre-decided stance. In a partnership proposal draft, while aligning a non-gendered programme with a funder's focus on young boys and gender, AI gradually reframed the grantee organisation's universal work as a gender intervention - with boys as the primary programme participant.

I do the reading and thinking first - immerse myself in the primary research and the organisation's work before touching AI. Only then can I flag information taken out of context, challenge incorrect reframing, and prompt AI to reexamine its assumptions.

Lead with your point of view, not AI's

AI creates technically correct and structurally sound writing. Written by someone who reads a lot but lacks lived experience. A story that hits all the right beats but without emotional truth.

Before touching AI, I start jotting down bullet points of what I actually want to say, in my words first - my opinion and unique point of view. Then I ask AI to expand, restructure, and stress-test. It allows me to build upon and critically defend my view or genuinely explore another.

Question every piece of content before you prompt AI to write it

While AI can easily generate content from just about anything, questioning the need for it is the starting point. Why put out a 'happy girl child day' social media post when the organisation’s work has nothing substantial to show for it? 

Only when there is a genuine point of view to put out should you prompt AI to generate a draft. Then aggressively rewrite every paragraph from one's own knowledge and voice. Human judgment and editing of AI output are non-negotiable. 

The pace of change in AI keeps me questioning and learning. But I believe the social impact communications professionals who will be most effective are those who use AI as infrastructure while keeping human judgment, ethical care, and genuine voice at the centre.

Curious to know what AI has almost cost you in your communications work?