Leading from the front: Our generative AI guidelines for content creation

Insights

• Generative AI is increasingly part of daily workflows, and content creators are considering how to get the most out of this transformative technology while understanding the risks and challenges it presents.
• Generative AI can support human work with tasks such as summarizing data, drafting ideas for content, and carrying out administrative functions.
• As a thought leader in this space, the Infosys Knowledge Institute has developed a series of guidelines for our content creation.

Generative AI is taking the world by storm, raising hope for progress, yet fears for the many risks it presents as well. Infosys is a leader in enterprise AI, and as its research and thought leadership arm, the Infosys Knowledge Institute is also striving to lead in the practical application of AI for content creation in the enterprise. For the Knowledge Institute, this means walking the talk: not only must we create insightful content in writing about AI, we must also demonstrate effective use of AI in our own work.

What is generative AI?

Generative AI can be considered a collection of tools built on computer algorithms that generate content. This includes written content, images, video, music, and synthetic data.

At the most fundamental level, generative AI is predicting the next word, the next pixel, the next note, in a sequence. The Financial Times has published a clear explainer of how this technology – the transformer model – works.

Generative AI algorithms are trained on a wide range of content and can additionally be trained on specific data such as an organization’s reports.

Generative AI tools can be used to support and augment humans at work: These tools can perform tasks such as summarizing data, suggesting topics for further research, drafting content, as well as more specialized applications such as visualizing new protein structures for further investigation by scientists.

The challenge of generative AI for content creators

Content creators around the world are grappling with how to use generative AI effectively and responsibly. In our work at the Knowledge Institute, we are experimenting with how to get the best from generative AI with a view to understand what this transformative technology can – and cannot – do for us.

This is why we created a set of guidelines for our writers, editors, audio-visual specialists, and data analysts – and share with others to add to the body of knowledge.

These guidelines are a work in progress: they codify our thinking as of autumn 2024. We will return to these guidelines; we will iterate on them. We may drop some as the technology and our thinking evolve, and we will add to them as relevant.

Above all, we will be transparent about our thinking on our own use of generative AI, and we offer these as an informed and thoughtful addition to the discourse.

Foundational principles

At the Knowledge Institute, we use generative AI to support our work rather than as a creative tool in its own right.

We understand how generative AI works, and know where it can support and augment human work. We are experimenting with how to integrate generative AI into our workflows. These are our foundational principles for using generative AI:

• Generative AI is a tool that humans must understand and learn to use. Think of it as gaining a skill in the way you learned to use a keyboard.
• Generative AI is a source of input for your research, not a replacement for it.
• Generative AI is not a knowledge engine: It is fundamentally a giant autocomplete tool with no understanding of what it is outputting. We therefore always have in mind that hallucination is a feature, not a bug, of generative AI’s written output, and that its output should not be taken on trust and must always be checked.
• Generative AI is an iterative tool: We get the best out of it when we build things bit by bit, not all at once. This process of creation means checking its outputs, and validating and enhancing those outputs with our own research.

Working with these foundational principles, we have devised a set of guidelines in four areas.

1. Enhance your fact-checking and editing

We insist on our team not assuming its outputs are to be relied upon:

• Producers must verify the accuracy of AI-generated content, including fact-checking and updating references.
• Producers must refine any AI outputs, adding depth, insights, and a human touch.
• Producers should specify in prompts the sources for AI to draw upon but must also verify AI’s outputs are real by going to the sources it cites and by verifying AI-generated links.

2. Put the human in charge

This makes clear that we do not see generative AI as a shortcut to high-quality work. We expect our team to drive the outputs of generative AI, not for generative AI to drive our work.

• Producers must create the overall narrative structure and key takeaways of their article before starting with generative AI.
• AI outputs are not a draft: you must integrate quotes, anecdotes, and specific examples from industry experts and from our survey results in the process of writing drafts.
• Producers must write drafts themselves, infusing their voice and expertise into the final content.
• AI tends to produce a series of statements. Producers must make sure the content flows logically, checking that transitions between paragraphs, sentences, and concepts work, and that the narrative is internally coherent.

3. Keep improving

This underpins all our work: we strive every day to improve on our work, and we support our colleagues both formally and informally to grow in their work. This equally applies to our work with generative AI:

• We will provide training for producers on how to effectively use AI tools, emphasizing the importance of having a clear narrative structure and takeaway in mind, and using targeted generative AI prompts to fit to this.
• Producers must document positive and poor prompts.
• Producers must also document changes in time taken and where your efforts have been focused during the process.
• We will develop guidelines for the acceptable percentage of AI-generated content to ensure originality and authenticity.
• We will have fortnightly group knowledge sharing sessions.

4. Be transparent

We are clear what we expect of everyone who contributes to our work:

• We are transparent and explicit about when and why we use generative AI.
• We will publish guidelines that clearly lay out our usage.
• We require our stakeholders and suppliers to be transparent about their use of generative AI.
• We will regularly publish our lessons and guidance.

How we use generative AI

At present we are using generative AI to help shape written content by using a series of iterated prompts to discover and consider key points for inclusion in our articles. We are also using it to suggest headlines, and for SEO optimization.

Our team members are also using generative AI tools to summarize documents, and to transcribe and summarize Microsoft Teams meetings. We are also using generative AI tools to monitor social media trends, and to support web development coding.

Our studio team is using generative AI to transcribe interviews, recordings and conversations, as well as enhancing audio for our podcasts, and to generate first drafts of scripts for videos and podcasts. We are also using generative AI to suggest names for episodes, as well as to generate short clips from videos to use in promotional material.

Our use of generative AI will evolve, and we will share openly and transparently how our use of these tools changes. We are learning as we go, and we hope you find our guidelines and experiences with generative AI helpful, and that they will help you on your own AI journey.