Prompts: How to speak with AI
Prompts are the requests we make to AI. How we formulate those questions is key. The quality of the answer we will get depends on the clarity and specificity of what we ask. To phrase it correctly, we need to be clear:
► Goal: Do we want AI to produce a text that we are going to use, or do we want to research, inform ourselves, explore? If it is the first case, we will have to specify details as to the tone and purpose of the text, and we may have to explain to the AI whose role it should play. If we want information, we will have to ask as specifically as possible what is needed.
► Conciseness: As AI is a natural language processing tool, it is best to be precise, avoiding words with double meanings or ironies. The clearer the description of the need, the better the response tends to be.
► Rephrasing: If the answer is not focused the way we wanted, we can rephrase the question to orient it towards what we are looking for, as many times as we need. The answers are endless.
► Adjustment: The generated response usually tends to be a good base but will need editing to adapt it to what we want.
► Creativity level: Some tools can blatantly invent what they don’t know. In some, such as Copilot or Gemini, you can regulate how much you want the AI to come up with. When we need information, more accuracy is better, but when we ask it for ideas, high creativity is better.
There are plenty of websites that offer prompts templates. It is best to experiment until you find the one that works best for you.
The AI tool of your choice should be set and customized further for the usage. ChatGPT for example allows to predefine a role and define how you would like to receive responses. Gemini only allows presetting by creating your own project through a Google Cloud API, while Bing/Copilot can be preset in corporate users. Having a registered user helps the tool learn our preferences.
So, you will soon find the right way of prompting according to your needs. To give just one example – if you want to generate a copy for social media, your prompt may look like this:
- You are the Community Manager of a company called [COMPANY NAME] specialized in [ACTIVITY]. Write a text for your profile in [SOCIAL MEDIA] aimed at [BUYER PERSONA] in order to [INFORM/INSPIRE/SELL] about [CONTENT THEME].
Best practices guide on conversational AI usage
In the end, we drew some basic conclusions for working with AI at evercom, which proved to be a kind of guide to best practice:
- Distrust: Don’t believe everything it tells you, it tends to make things up easily. Verify by using several sources.
- AI is always the beginning, never the end: It is a base to work from, an inspiration. Never accept what it presents you with as a final version. It moves freely in the realm of political correctness, it lacks wit.
- Improve it with experience, creativity, and critical spirit: Claim your input: Experience to apply your knowledge of the reality, the client, the context. Creativity to give it identity and relevance. Critical spirit to doubt and better substantiate what you write.
- Never upload confidential information: Documents’ confidentiality is not guaranteed. Do not copy and paste confidential texts or upload sensitive client information.
- Don’t deny that you use AI, but you don’t need to warn anyone: Never deny to a client that you have used AI if they ask you about it. We have nothing to hide, just as we don’t think it’s necessary to put a disclaimer when we use it, since we won’t accept its output as a final deliverable.
And here are a few additional tool tips
The debate about the origin of content will certainly intensify in future. So, it’s good to know some tools to check if something is generated with AI:
- Español Detector de contenido de IA – Smodin
- Detector de IA para GPT, GPT2, GPT3 y GPT4 (plagiarismdetector.net)
The other way round, there are also some tools that help make the texts generated by conversational AI sound more natural:
- AI Content Detector, Checker and Humanizer | Undetectable AI
- Humanizador de IA: hacer que la tecnología se sienta como humanidad – TextFlip
The more widespread the use of AI tools in agencies becomes, the more efficiently they will work. And that means that the use of the customer budget will become an issue. Agencies would do well to prepare themselves for the debate about what value the creation of texts or concepts will still have in the future, if they are apparently so easy to generate with AI. It is a big task to find a fee model that compensates for all the expenses incurred by the agency long before the perfect result is delivered.
Tilo Timmermann, Co-founder and Managing Director of the technology PR agency TDUB Kommunikationsberatung in Hamburg, Germany.
Ander Serrano, Partner and Director Tech & B2B Communications of evercom in Madrid, Spain.