Introduction
This article follows on from the introductory How Do I Create Feed Variations. It aims to explore in more depth how AI can be used within the Feed Variation table, focusing on Copy and Translation.
AI within Feed Variation
To briefly recap, once you have setup your Feed Variation table with headers, mapped those that need mapping to elements/layers of the creative, and input any fixed, known or reference data in the respective columns, you are in a position to 'Ask AI' to help you generate content in the cells of your choice.
Example: Godi Coffee
In this simple worked example we will setup a simple sheet with a simple table for a brand Godi Coffee looking to capture attention with some new creative with a short, simple Call To Action (CTA), which in this case is mapped to the Creative layer 'Headline':
The variables we have are Drink, Audience and Language, which I have pre-filled with the following data:
I will be using 'Ask AI' to generate CTA copy that references the Drink in column A, the Audience Persona in column B and expresses the CTA in the Language given in column D.
Iteration 1
In this simple example, through a process of iteration, and refinement of the prompt, we will look to maximise the quality of the CTA - both in British English and in Italian.
We will start with a very basic prompt to highlight the structure:
Tell the AI model what we want: 'Generate a CTA'
Tell the AI model we want it to aim the CTA at the Audience in $B2. Similarly to Excel or Sheets, prefixing column B with a $ sign will 'fix' column B as the reference column for Audience when we come to applying this same prompt in the cells below.
Tell the AI model that the CTA is to buy (the Drink) in cell $A2
Tell the AI model that the CTA should be in the language specified in $D2
Clarify for the AI model that $A2 (Drink) is the noun in this CTA
Tell the AI model that it should generate a CTA with a maximum of 8 words
Below are the results of this first prompt:
The CTAs are fairly basic - both the British English and the Italian. They could both be improved and so I will work on the next iteration of the prompt:
Iteration 2
“Generate a CTA that starts with an imperative verb followed by a luxurious sounding adjective to describe $A2, which is the noun, keeping in mind Audience $B2, and language given in $D2. Max 8 words"
Here, we see repetition in the British English CTAs and possibly a slightly awkward use of the word 'Experience'.
The CTAs generated in Italian have a few issues:
'Mocha' is not a common term in Italian for the drink so would likely need some context.
The Mocha CTA comes across somewhat awkwardly and overwritten - with the use of a double embellishment “Delizioso” and “lussuoso” a bit excessive and unnatural for Italian marketing copy, and the adjective order feels clunky.
The Flat White CTA is a little better, but still quite literal as opposed to natural sounding.
'Flat White' is loaned from the English in this context, and "Assaggia" suggests that you are sampling a spoonful. There are likely more idiomatic choices for a drink.
Final Iteration
We want to improve our prompt to encourage more natural, idiomatic language, and encourage the AI model to use a greater variety of adjectives.
“Generate a CTA that begins with an imperative verb and includes a fresh, luxurious-sounding adjective to describe $A2 (the drink), tailored to $B2 (the audience). Ensure the CTA feels natural in the language specified in $D2. Avoid mechanical phrasing or overused language. Prioritise fluidity and idiomatic tone. Use a distinctive adjective and verb not already used in other CTAs. Prioritise vivid, elegant vocabulary. Keep to a maximum of 8 words."
The results show some significant improvements in the British English CTAs although in this simple and fairly limited example they might want further human improvement, or choosing to define specific brand-aligned words for the AI model to choose from.
The Italian CTAs are the best iteration yet and while they could be re-generated over and over with the same prompt, it's unlikely that consistent further improvements can be made now with the Prompt itself.
“Assapora il tuo vigoroso Long Black oggi!" - Assapora is a good verb here - evocative and elegant. Vigoroso works from a technical standpoint but can feel slightly mechanical or medical in Italian. 'Deciso' or 'Intenso' might be more natural sounding
“Sorseggia un Flat White vellutato e raffinato.”: This is a good CTA, combining a luxurious trio of words: “Sorseggia”, “vellutato”, and “raffinato”
Conclusion
Although the simple nature of the eight-word CTA task here doesn't always give the AI model a great deal of space for creativity, it does highlight the importance of prompting well both in terms of adding descriptors to improve the choice of words, as well as being clear about what you do and don't want from translation - in terms of style, tone, and types of vocabulary.