Marketers have spent the last two years studying how to encourage AI systems by tossing ideas at them and seeing what sticks. This approach has resulted in a few practical techniques, such as step-by-step algorithms for condensing prompt information and galleries, including various text templates.
However, sometimes the black box of AI demands surprising tactics to achieve desired results. For example, when Jesse Suchmann is confronted with a large amount of complex material, he may occasionally ask a chatbot to break it down like an intoxicated friend at a bar.
“That usually brings me directly to the point,” said Suchmann, Edelman’s senior VP and executive creative director. According to Suchmann, if the bot uses harsh language, the summaries will be more remembered.
This is one of the techniques that marketers are adopting to improve their AI prompting game for chatbots like ChatGPT, Gemini, and Claude. Users query AI systems using prompts, which are often written as bits of text. However, prompting, like any other type of writing, is an art form that should be deliberately steered to generate the greatest outcomes.
Ad Age has developed a list of these methods from agency staff who work with AI regularly. Here’s what the pros recommend, whether you’re starting with a prompt or adding the final touches to your project.
Don’t be Stingy with Details
“I think the essential thing to remember is that generation AI requires and deserves the same input as a human copywriter to generate strong copy,” said Bas Korsten, VML’s global chief creative officer of innovation and chief creative officer for Europe, the Middle East, and Africa, via email. “The more context (e.g., client brief, creative brief, important insights, brand guidelines, tone of voice), specificity, and direction, the better.”
Jessica Berger, senior VP of innovation at Publicis Media, gives detailed information from the start.
“Typically, I offer a brief idea description, followed by copy prompts geared to certain audiences or moods,” Berger stated. “If we have visuals, such as photographs or movies, the prompts should correspond to those storyboards to ensure consistency.
Berger uses many prompting structures to prepare the AI for certain tasks, such as asking it to “act as [blank]” to assign it a part to play. Berger also provides context for the chatbot, such as asking it to “write for” a certain circumstance, as well as guidelines, such as telling it to use a specific slogan or avoid jargon.
These specific instructions allow Berger to avoid general prompts, which are a quick way to squander time.
Berger described a too broad and unfocused question as, “Tell me everything about developing an ad for a new soft drink.” “Compare this to: ‘Create a slogan and brief ad copy for a new soft drink made just for gamers. Concentrate on its strong caffeine content and stylish look. Include two versions, one for competitive esports fans and one for leisure players.”
Some Context is Useless
“Overloading the AI with too many queries or complicated context frequently results in ambiguous or confused conclusions,” Berger stated. “It is preferable to focus on one activity at a time to keep outputs clean and concise.”
When chatbots are presented with unclear context, they may hallucinate, thus Berger additionally promotes factual correctness in her questions by noting “Refer only to genuine persons or well-documented incidents,” Berger said.
AI Isn’t Don Draper
Regardless of how hard a marketer tries, existing AI chatbots will not produce amazing advertising material on their own. Some platforms, notably ChatGPT, are still working out the intricacies of basic marketing principles, according to Emma LeMay, a copywriter with Pereira O’Dell. Chatbots, for example, may be unable to distinguish between a “headline” and a “tagline,” according to LeMay.
According to Peter Steiner, senior director of creative technology at Code and Theory, asking a computer to write advertisements will result in only the most obvious concepts. This is because AI is fundamentally predictive, surfacing predicted material based on previous data. According to Steiner, models like ChatGPT are more valuable as interpretative AI rather than generative AI.
“The technique works by recontextualizing information,” Steiner explained. So it doesn’t truly produce anything. At best, it uncovers previously unknown links.”
When in Doubt, Start Again
“Continue to start afresh,” Steiner urged. “If you’ve been working on the prompt for a while and don’t know what to do next, press the reset button. Begin a fresh conversation with your improved prompt. From a technical standpoint, you must clean the memory so that the model is not confused or distracted by any previous concepts that you have abandoned. Begin quickly and fail quicker.”
Some people may dislike the thought of failing rapidly, but there are methods to depersonalize the process and move it along even faster. For example, Steiner frequently requests that the model compose the question for him.
“That will rapidly get you over the blank page, and it frequently provides you with a fantastic structure to work with,” Steiner said.
Even if the results are initially poor, they may be valuable.
“Start a fresh discussion and give the AI the prompt and the negative response,” Steiner said. “Tell it what you don’t like and ask it to assist you improve the prompt.”
“I also like asking [the bot], ‘What am I missing?’ or ‘What am I not thinking of?'” Steiner said. “The AI may surprise you.”
Try on Different Audiences
“Always try to fine-tune the tone, energy, and call-to-action based on the audience’s attitude, ensuring the final result is personalized and connects,” Publicis’ Berger explained. Berger provided an example: “Rewrite the following content about a technology device in a way that piques your interest and excitement. Then build a version that appeals to tech skeptics by emphasizing trust and dependability.”
Pereira O’Dell’s LeMay employs a similar method, feeding the AI pre-existing work and demanding tone changes, such as in the voice of an eight-year-old, according to LeMay.
Korsten at VML likes to create various AI personalities with varied perspectives, allowing for input that differs from persona to persona. Google and other providers have begun to make persona development easier by introducing single-purpose AI helpers.
Remove the Excess Fat
“Character count is a constant fight for copywriters,” LeMay remarked. Fortunately, LeMay has discovered that ChatGPT works well for adapting headlines or content she has written to new character lengths.
Edelman’s Suchmann only uses the bot’s trimming skills when the copy is ready to go.
Source- Adage
English
French
German
Russian
Spanish