Variables and Placeholders in your ChatGPT prompts

1. Home
2. Insights
3. Variables and Placeholders in your...

Get more out of ChatGPT with smart and functional prompts

Since the launch of OpenAI’s ChatGPT in November 2022, the day-to-day life of marketing teams has changed fundamentally. Now, after almost a year, it is the right time to report on the benefits of the new way of working as well as everyday challenges.

The need for a prompt repository

As part of our daily agency work, we use AI to create engaging content for our clients’ advertising campaigns, websites, newsletters or visuals. Through continuous optimization and excessive testing of the commands to the AI, we have been able to keep improving it and thus produce the best possible content for our clients’ brands. The result is an ever-growing list of commands (AI queries) that we use as a template to create different types of content we need.

Our team quickly needed a repository for recurring AI queries so that we didn’t have to rewrite already optimized AI queries for different content types over and over again. An example of this is the creation of recurring social media posts (e.g. sweepstakes), which we implement for a client’s brand on a monthly basis.

An example of such an AI prompt is the following:

The content generated by the AI then looks like this:

By the execution of A/B tests or other learnings from the business, these AI prompts are very often extended and further optimized to achieve the best possible outcome with the AI. In addition, queries are often customized to the product or brand to be advertised in order to best match the brand voice.

The need for an easy accessible prompt repository is thereby essential to keep track of all customized AI prompts. In addition, it ensures that learnings are not lost and are already included in upcoming AI prompts.

The use of placeholders (variables) for recurring AI queries

After creating an AI repository, the wish for placeholders (variables), which can be freely placed in individual AI prompts, quickly arises. For example, in our previous AI prompt, the briefing could be parameterized as shown below:

The execution of the AI prompt could be easily triggered by just entering all needed parameters (variables) in an input mask and the expected outcome will return:

The integration of external content (or platforms)

Another level of automation is the connection of external content or platforms. A simple example is the automatic creation of teaser texts or the summary of content from external sources or websites. Instead of manually copying the content for each article teaser of a newsletter from the respective website and integrating it into our AI queries to write a summary, we thought about how to automate this process.

Therefore, we use a placeholder type, which automatically crawls the content of an external URL and combines it with our given AI prompt:

The above AI prompt is then automatically completed by simply entering the number of characters of the title and introduction and specifying the URL of the article for which a teaser text will be generated. When doing so, we enclose the external content of the [{url}] placeholder intentionally in triple quotes and follow OpenAI’s best practices for placeholders. The final AI prompt would then be composed as follows:

Exclude irrelevant content from the external URL

As a further addition, we thought about the exact selection of content areas of an external URL, especially to exclude footer content or other not relevant content, that have nothing to do with the actual article itself. The easiest way to solve this problem is to select a specific HTML element or container to which the external content is restricted.

This can be done the following way by including the CSS-Selector.

Only include the content of the element with the ID “article“:

Only include the content of the element with the Class “content“:

Combine AI prompts in templates

Grouping various AI prompts into templates allows our team to create comprehensive content structures with just a few clicks. For example, a complete newsletter with all required elements can be generated from the AI by simply specifying the overarching topic. Article teasers can be created in any number and generated fully automatically by simply entering the URL of the article.

Instead of writing full newsletters, the AI now generates multiple options for all required content types (such as subject line or a newsletter introduction). Our team focuses purely on the selection and optimization of that content to achieve the best possible business outcome.

A quick insight into what we are currently working on

Now that we have also created the possibility to group different AI prompts into templates, the next step is to integrate more platforms to accelerate content distribution and automation (e.g. a fully automated delivery of AI-produced, pre-approved and slightly modified newsletter content to common campaign solutions).

Try Formula8.ai for free to create fully automated marketing content and let us know your suggestions for improvement.