This is our primer on how, when and where we use AI at Narratives Work. We take you through a few things we believe, and a few things we don’t, about how AI can help companies like ours give clients better results – with guardrails.

AI is here to stay

Artificial intelligence tools are fast becoming part of modern work. Firms like ours, that advise partners, founders and CEOs, have an obligation to understand where the market is heading and how work is changing. That means finding ways to continue improving results for our clients.

So we do not have a blanket ‘no AI’ policy. But there are strict guardrails for its use – some of the strictest in our line of work, as far as we know. Here, we outline when and how we use AI in our work, and when we do not; which tools we use, and why we use them. 

The very best work is done by people, not machines. That will not change any time soon. But sometimes, new computing technologies can help us do great work faster and at a more competitive price. It is our duty to be clear about where we think AI can improve results for our clients, and where it can’t.

We believe that AI can be a helpful sparring partner

We use Anthropic’s Claude large language model (LLM) for a finite, pre-defined set of tasks during client projects.

In a content review, if there’s a sentence we want to improve, we might bounce a couple of ideas around with Claude, exploring alternatives and then, if Claude’s helped, using aspects of that conversation to create a better sentence. When we’re analysing data, we might upload a csv to Claude and ask it to highlight any patterns, anomalies or interesting findings.

We say ‘might’ because we will only ever do this with your permission. If you are unsure about the benefits of AI tools, or if you have concerns relating to data protection and privacy, we are very happy to carry out our work without these tools. If you wish for us to refrain from uploading certain forms of data – such as spreadsheets – to an LLM, just say the word.

Using LLMs in this way can speed up the time taken to deliver work, resulting in faster turnaround times on projects and, for some clients, lower fees. But LLMs can be inaccurate, and they can hallucinate. We do not take these risks lightly. We aim to use AI as a tool to generate early ideas quickly – not to ‘do work’ on its own, or to create client-ready deliverables.

So why Claude? Well, privacy is very important to us. We judge Claude to be the most transparent and privacy-oriented mainstream AI tool available right now. (Those two terms aren’t mutually exclusive. We think Anthropic has reasonable data privacy protections, such as excluding user conversations from its model training data by default, and that Anthropic does a decent job of being open about its data use and privacy policies. Read more about Anthropic’s approach to data privacy.)We preserve all artifacts* relating to our clients’ work until the end of an engagement, when all artifacts will be deleted. This is the default approach, but clients are free to ask us to delete artifacts, or to share with them any and all AI artifacts relating to our work with them, at any time. We will always carry out these requests.*Artifacts are Anthropic’s term for the outputs Claude provides to users as a response to a prompt. Read more about artifacts.

We believe that we are, and will always be, more creative than any AI

Every day, your current and potential customers wade through a mass of subpar content, much of it AI-generated slop. The reality is that your buyers are jaded. They are rarely sure that the stuff they read and watch will actually help them.So the bar is set low. And this is your opportunity. Great content and communications sparkles; it’s sublime, electric. It seeds trust. It changes people’s minds. LLMs can’t tap into the experience we’ve acquired having spent years delivering great work for clients just like you.

We back ourselves to give you results full of heart, wit, and ingenuity – the stuff AI can’t automate. We do this by understanding you and your business before jumping to recommendations. How do you want to speak? What tone of voice fits your needs best? What stories and trends excite you? Hearing in your words what you want to achieve gives us the colour and the context we need to do exceptional work.

What we don’t believe

We do not believe that quicker means better. Good, cheap, fast: you can have two, but never all three. Most of the time, using AI gets you OK results very quickly. We don’t think that’s the way to do exceptional work. When we need to take more time, we will say so. We will never use AI to rush work, or to hit a deadline, when the use of AI would deliver lower quality results.

But, we do not believe that using AI automatically means you’re lazy or sloppy. We do a lot of thinking about choice of language, tone and style, about how data and opinions mesh together, and about the distribution and marketing of the work we do together with our clients. As a resource and a companion in this thinking process, LLMs can expand the range of options we’re able to consider. This is helpful as long as people have the final say on all work, from the title to the very last punctuation mark.

These are our AI ethics – until they aren’t

As with all things, when the facts change, we reserve the right to change our minds. In a space that’s morphing and expanding so quickly, the most sensible thing to do is to keep close tabs on the tools we use, and our own methods of working, and pull back if there are substantive shifts that may impair our ability to deliver stellar results for you.