When using generative AI in journalism, it’s essential to create the right prompts. The quality of the prompts will define the quality of the response.
Here we look at how journalists can create prompts that bring meaningful results. The following draws on how we use AI at Media Helping Media (MHM).
First, some rules that journalists and media organisations might want to consider adopting.
- Rule 1: AI to be used only to expand upon and support our own published content – never to generate original material.
- Rule 2: Check all AI-generated material carefully and edit it where necessary to ensure accuracy, coherence, and relevance.
- Rule 3: Use AI as a carefully supervised assistant and NOT as the driving force – which will always be human input and control.
So, with those rules in mind, let’s look at how to compose prompts.
Clarity
Be absolutely clear about what you want AI to produce. If you are not specific about what you want AI to do, it will struggle to deliver the results you require, and it might even create unhelpful results that lead to misinformation and confusion.
Give it a role
When you compose the prompt, tell AI the role you want it to fulfil. You might want it to act as a proof reader looking for spelling and grammatical mistakes in an article you have written.
If so, tell it what its role is and what you expect of it. You may be surprised to find that AI will attempt to emulate that role and tailor its answers accordingly; AI is keen to please.
Assign it responsibilities
Treat AI like you would a journalist in the team who you are instructing to produce a piece of journalism.
You would brief the journalist with exactly what you wanted. So do the same with AI.
Make it absolutely clear what you want it to produce. If you are vague the results will be vague.
Give it the ‘grunt’ work
There are many newsroom tasks that take journalists away from their main role of digging where others don’t, shining a light in dark places, scrutinising the executive, and holding leaders accountable.
So why not hand some of these to AI in order to free up journalistic resources?
If you have to write a social media tease based on an article you have written, why not ask AI to do it.
All you need to do is tell it the platform you want to post on, the character limits for that platform, how many hashtags you require, and the link you want added to the post. You might also want to stress the keywords you want added.
Then give it the content you have written and AI will do the rest.
You can prompt it to provide social media teases for all the platforms your news organisation shares content on and it will tailor its response to those needs.
And you are now free to work on more important journalistic work.
Get it involved in training
If you are involved in training your junior colleagues, you might want AI to create the structure for a four-hour training workshop based on an article you have written.
So you need to write a prompt to save you hours of work building a training course.
All you need to do is make sure you ask AI to create a timetable for the event, tell it when you want the breaks, that you want it to include a presentation (based on the content you give it), include activities for the participants to take part in, produce an outline of any discussions you would like, and tell it you want it to set an assignment (again based on the content you upload).
It’s also worth instructing AI not to add any content that has not been produced by you. You want this training material to be based only on the material you have written.
You will find that AI will produce in a matter of seconds what would have taken you several hours to write.
Set rules
Be absolutely clear about what you want and what you don’t want. You can’t be rude with AI. Blunt and focused is good; soft and vague is bad.
So set out exactly what you require and leave no room for misunderstanding.
Your prompts need to be direct, forceful, and specific – ambiguity will confuse AI and the results could be damaging.
Audience
It’s important to tell AI who you are writing for.
You will be well aware of your target audience and their information needs, but AI won’t have a clue – unless you tell it.
So your prompts need to make clear not only what you require and why you require it, but who it is for.
Tone and language
All news outlets have their own tone and language which is set out in the house style that all journalists are expected to follow.
You will be well versed in that, but you can’t expect generative AI to know what your media organisation’s chosen style is. So you need to tell it. And the prompt is the way to do this.
Repetitive tasks
Google Gemini has a tool called a Gem. This allows journalists to programme their own AI chatbot to perform tasks to achieve certain results.
At MHM we have a large collection of Gems to help us speed up the production of content and make it more accessible and more useful.
The benefit of these Gems is that you can set all the requirements listed above, and more, in the Gem’s ‘instructions’.
This means that you need to tell AI what to do just once and the same Gem can be used multiple times – another time-saving benefit.
MHM AI Gems
Our AI Gems currently deliver:
- SEO-friendly headlines to ensure articles perform well with search engines
- Social media share teasers to boost social sharing
- Quick guide checklists for journalists to remember the essential elements of journalism
- Self-study exercises for those starting off in journalism
- Workshops for group learning
- Lessons for day-long training courses
- Modules that cover six-weeks of lessons to be used by colleges and universities
- Refresher courses based on the modules and lessons for senior journalists
- Proofreading for spotting typos, ambiguity, or poor sentence construction
And all the Gems listed above have detailed instructions (prompts) built into them so that they perform to the exact standard required by the person building the Gems, in our case the MHM team.
If you are interested in building and prompting your own Gem try this helpful guide which takes you through the required steps.
To sum up
To return to the point made at the beginning of the piece, the importance of giving AI a well-crafted prompt is simple: “Quality in, quality out”.
And this is essential if journalists are to keep control of AI in the news production process.
If we don’t tell AI exactly what we want, how we want it, who it’s for, and what we intend to do with it, then it is likely to produce inadequate, and, even worse, harmful results.
But, if we are clear on what we want, and set that out in high-quality and detailed prompts, then we lessen the chance that AI will produce material that is not only unhelpful, but which could be potentially harmful.
