Whether you've played with other types of generative AI or have arrived fresh on the scene, it's important to understand how generative models work and how Phrasee is different from other AI systems.

Firstly, putting AI aside it’s important to remember that with any copywriting task there are two parts to consider:

If you’ve used other AI content generation platforms, you’ll have seen the word "prompt" used to refer to user inputs. A prompt is the (natural language) input to a generative language model, usually an instruction for how to complete a task.

Importantly, a prompt goes directly into the model with no intermediate processing. This is why in many platforms you need to include instruction not just about Step 1 (what to say) but also Step 2 (how to say it).

With Phrasee’s Content Generation, the situation is different. We have abstracted away the need for you to instruct the system about Step 2 (how to say it).

Take a product description for example. We have already written a boilerplate prompt with instructions on how to write a product description. All we need from you are details about what to say: the product name, its features, etc. These inputs then become one section in a much longer, more complex prompt (see Figure 1 below).

To review, Phrasee takes care of instructing the model on Step 2 (how to say it), so you only need to provide Step 1 (what to say). Consider these two user inputs:

BAD: We are sending a re-engagement email to our womenswear segment offering 50% off everything for Black Friday

GOOD: 50% off everything for Black Friday

The first is not ideal for the Phrasee system. Why? The first part of the input, "We are sending a re-engagement email to our womenswear segment," is not necessary. We have already covered this information in the prompts we construct behind the scenes.

The second is ideal because it only specifies the message you want included in your content.

Adding information about how to write the content (e.g. "write it under 50 characters" or "write as if you’re talking to Gen Z") is not expected, and including it will result in poor content as the Phrasee system was not designed to process these kinds of inputs.

Review your input. Is it written literally and unambiguously? Have you unintentionally added instructions for how to write the content instead of focusing only on the message of the content?

Review your settings in the Brand Voice page. We will soon be releasing a new feature that allows you to configure the tones Phrasee should use when writing your content, giving you even more control.