Using generative AI for content management

How to improve your customers' digital experience with generative AI

How to improve your customers' digital experience with generative AI

Meeting the demand for persona-based content can be a challenge for content marketing teams, and it seems that artificial intelligence (AI) is the answer to this problem. At Magnolia, we've been working to make AI work for content teams, and with the release of our AI Accelerator, we've found a solution that allows editors to increase the speed and scale of content creation with proven generative AI tools of their choice - without leaving the application interface and workflow they're used to with Magnolia.

But with so many generative AI tools to choose from, it's not always easy to decide which one is right for you.

Text generation
Image generation and editing
Translation

We also look at use cases (including how we use some of these tools at Magnolia), potential legal or ethical concerns and the role of AI detectors.

Are you ready? Then let's get started.

Understanding the importance of artificial intelligence for content creation

Topics such as computational linguistics and algorithmic language processing are so often in the headlines these days that it is difficult to grasp the specific impact of each AI model. Before we look at the different tools, let's define what we mean by generative models in the context of marketing or content management.

With the exception of virtual assistants and productivity tools, you will mainly find examples of generative AI in the marketing world rather than a general-purpose neural network. When we talk about "generative AI models", we are referring to the machine learning model's ability to create new content. The term is therefore often used to distinguish models with a narrow focus from general-purpose AI systems.

In order to use generative AI models effectively, understanding the data science behind them is crucial. Without this understanding, it becomes difficult to compare and evaluate options.

Many companies advertise the synthetic pre-processing of their data or the number of tokens, as they have recognized their influence on the quality of the results. In order for a large language model (LLM) to make good use of its training data, engineers must standardize and encode the information for the algorithm. Limited data sets and incorrect data cleansing lead to inaccurate results or a discriminatory AI algorithm.

Similarly, the tokens of each generative AI application determine the use cases it can handle. You may have heard of a hallucinating AI chatbot. The tokens determine the framework in which a model can interpret a prompt before it starts hallucinating. ChatGPT-4 is capable of processing 8,000 to 32,000 tokens or 6,000 to 24,000 words. Another language model may have different token limits or may try to compensate for them by using integrated web access to accept new data.

If providers do not provide information, tests such as the ROUGE or BLEU score offer metrics for AI translated or summarized texts.

Regardless of which tool you choose, you should always consider the ethical and legal implications of incorporating these tools into your marketing strategies. Using generative AI models requires you to share at least some details, and entering confidential company data could impact your contractual obligations.

Nevertheless, integrating a generative model into your content management efforts enables optimized content production and can save valuable resources.

Choosing the right AI tool for text creation

Given the abundance of tools and features, it's impossible to recommend a single AI tool for your text needs. We'll start with some use cases, followed by our recommendations.

Drafting marketing copy: While applications will differ in their user interface (UI) and output, most can be used to generate headlines, meta descriptions, alt texts, or fleshed-out blog and landing page copy.
Summarizing videos or website data: Generative AI tools like ChatGPT and Copy.ai scrape publicly available data to enhance their text-generation capabilities. You could use those to summarize a podcast’s talking points or draft a TLDR for a blog post.
Develop content briefs and outlines: Some applications offer templates aimed at marketing teams, whereas others require you to prompt the language models to deliver one, either based on your company profile or competitors’ branding choices.
Brainstorm ideas: Every marketer has gone through a creative block. Especially in remote teams where a colleague is not always available, bouncing ideas off AI technology can be a great starting point.
Repurposing copy for social media: Simply feed your blog copy into generative AI algorithms and ask for a LinkedIn post or Instagram caption based on your content.
Analyze competitors: Whether you’re analyzing a competitor’s word choices in headlines, their use of technical jargon, or their entire stylistic preferences, a machine learning model makes en-mass analysis easy.

The natural language processing solutions presented here lead to different results within different user interfaces. To accelerate the controlled training process, content generation and optimization in a scalable way, a composable digital experience platform like Magnolia offers a combination of tools in unified user interfaces and editorial workflows.

Remember this when assessing generative AI technology because some integrate with platforms like ours, giving you more flexibility and helping you avoid the tedious and error-prone copy-paste, as well as the time-consuming context switching between multiple tools.

ChatGPT: While possible, using ChatGPT alone necessitates copying and pasting results into your draft. Paired with a platform like Magnolia’s, you can generate landing page content, meta tags, and image descriptions right where you need them, in the authoring interface. The thing to remember with this tool is that the quality of your output hinges on well-designed prompts.
Jasper: With a narrower focus on marketing teams, Jasper complements the open playground ChatGPT delivers. It relies on Zapier for most integrations beyond Google Docs and Grammarly, but it can make up for that with dedicated brand voice features and SEO functionality.
copy.ai: Similar to Jasper, copy.ai focuses on email marketing, social media copywriting, and blog content. copy.ai offers no integrations but instead doubles down on templates for workflows like webinar follow-ups or competitor analysis reports. This makes it a suitable choice for those dipping their toes into AI waters.
Writesonic: Writesonic's tools cater to diverse content generation needs, such as article writing, paraphrasing, summarizing, story-generating, and landing page generation. Its unique value proposition comes from combining those with image- and audio-generating tools.
Writer AI: This application appeals to healthcare and financial sectors with a self-hosted solution that won’t use customers’ prompts for training. To round off that niche offer, its workflows are tailored toward clinical services or equity research.
Amazon Comprehend: An ideal solution for ongoing research and project management and an integral part of Magnolia's text classification module, where you can use it to analyze and tag your text content for better searchability.
Google Bard: After Microsoft integrated ChatGPT into its search engine Bing, Google responded with Bard. That makes it not so much a content generator as a research assistant, the advantage being that it links to external sources, which some free tools can’t provide.
Poe: Think of Poe as a registry for bots and AI tools, including StableDiffusion and Claude. While it can be used to try out and compare various options side by side, it’s not aimed at marketers but rather designed like a social network where users can develop their own tools and share results.
Perplexity: This chat bot is similar to Google Bard in that it references sources. That can make it a viable research tool, but it’s still mostly a question-and-answer machine, even though the paid version does provide API access.
You: You, or YouChat, is one of the AI platforms running on ChatGPT. It’s not just a differently themed skin for OpenAI’s product, though. It enhances results with videos, images, and diagrams, and it also references sources.
QuillBot: This application stands closer to editing tools like Grammarly or ProWritingAid. It can be used to rewrite existing copy or adapt to a different style. Unlike ProWritingAid, it’s available in many languages.

So how can such text creation programs help with day-to-day editorial work? Take a look at how we have integrated ChatGPT and Amazon Comprehend into Magnolia to support the following use cases:

Automatic generation of components (text, images, or a combination of these) and even entire pages.
Automatic generation of component variants for personalization.
Automatic generation of metadata and image description (SEO and OG), to optimize content before you hit publish.
Automatic generation of image descriptions based on image tags.
Automatic tagging and text classification for better searchability.

Generative AI tools for images and logos

Image generation tools face some of the same challenges as other AI models, be it legal considerations or the extent to which the process relies on prompts rather than templates. However, applications that offer AI capabilities for image generation tend to have slightly more issues with providing consistent branding across all content.

Nevertheless, they can be helpful in the content creation process because they serve different needs than graphic designers. Sometimes they can simply help a client get a rough idea across or fill in graphic patterns to a level of detail that would otherwise not be feasible for many. Here are some tools to consider for your image editing needs:

DALL-E: OpenAI’s AI model can help you generate visual content based on text prompts, which makes it a useful tool for product visualization, content ideation, and creation. Although images will mostly require a graphic designer to maintain a uniform style, this gives you not only a great productivity boost in terms of content output but also a helpful aid in supporting feedback rounds with design teams. We’ve integrated DALL-E into our DXP so you can easily generate and edit images and directly save them in the Magnolia DAM.
Shutterstock: When the photo stock marketplace entered the AI realm, it addressed an issue totally unrelated to creativity or output, namely licensing. With their shift toward AI, Shutterstock introduced a Contributor Fund to compensate artists and an AI-specific license, which lets you use generated images on commercial pages.
Midjourney: If DALL-E excels at generating images based on text prompts, Midjourney has a stronger focus on image transformation through filters and modifiers. That can make it a better choice to add visual effects, whereas DALL-E tends to deliver more accurate images in one go.
PhotoRoom: Free tools like PhotoRoom don’t try to impress you with endless features, but they do one thing incredibly well. In this case, you can easily remove backgrounds on your product pictures.
Runway and Playground: While Runway’s or PlayGround’s features may not be unique, they could make it easier for some users to get started through a limited number of choices. Where you would need to be familiar with dedicated operators to expand an image or adjust the style, they give you a selection menu or let you use text input for photo-editing features, similar to what you know from tools like Adobe Photoshop.
Hotpot: Another example of a tool that doesn’t necessarily stand out based on features but curation. Rather than overwhelming you with an open playground, Hotpot is organized into use cases like logo generation, profile picture editing, or game copywriting. It may not match everyone’s needs, but it certainly serves as an indicator that templates and examples can make AI’s limitless opportunities less intimidating.
Stable Diffusion: While realistic images haven’t been one of this model’s strengths in the past, it does offer greater customization options than Midjourney, including the ability to run it locally on your machine.
Leonardo: This tool tries to appeal to the creator community and artists looking for more creative control. Marketing teams could benefit from more advanced editing features, although Leonardo’s Discord community seems to suggest they’re addressing individuals and smaller teams.
Cutout.Pro: Cutout.Pro supplies dedicated AI tools for background and object removal in images, color correction, and art generation. Other than comparable tools, it provides those for both image and video editing and offers an API for further integration.
Looka: Where other vendors curate single solutions based on common edits, Looka focuses on a marketing team’s need for consistent branding, giving you a complete toolkit for brand guidelines, social media posts, newsletters, and more.
Amazon Rekognition: As the image and video counterpart to Comprehend, Amazon Rekognition lets you analyze multimedia content to identify objects, patterns, and people. This can be useful in many business applications, from user identification and violent content detection to logo detection for your social listening efforts, which is why we’ve integrated it into our platform’s Image Recognition module.

At Magnolia, we have integrated DALL-E and Amazon Rekognition to support the lifecycle of your content:

Automatic generation of new images during operation
Modification of existing images
Automatic recognition and indexing of images for better searchability

AI translation tools

Although we have already looked at applications such as ChatGPT or Copy.ai that have certain translation features, we believe that this category deserves its own section for different features that would otherwise not be used. In our comparison, we leave out some obvious features and focus on those that are most important for content production and marketing teams.

DeepL: DeepL overtook Google Translate among users a long time ago because it recognizes text nuances better and reacts flexibly to desired variations. This is why it is our AI translation tool of choice for integration into Magnolia. It has also been trained on legal texts
Google Cloud Translation: Not to be confused with Google Translate, this tool appeals to larger content production teams. You can set up several user profiles, upload resources, and manage translation tasks within Google’s dashboard. It also allows you to brand the UI, should that be helpful for client-facing drafts. However, translation memory and post-editing functionality are reserved for its higher price tiers.
Microsoft Translator: While it tends to fall behind DeepL and Google in many reviews when it comes to accuracy, Microsoft’s translator offers a few unique solutions, like the option to translate into multiple languages at once or to auto-detect the language sent to the API. It also provides transliteration capabilities and a bilingual dictionary, which may not necessarily help with content translation but streamline communication in an international team.
Translate.com: What makes Translate.com unique is its acknowledgment that machine or AI translation has its flaws. That’s why they’ve embedded options for localization and human translation into their service offering, which can be especially useful for medical or technical industries.
Lokalise AI: Lokalise AI takes a different approach that’s more tailored to marketers. You can upload style guides and glossaries and customize your translations, even for SEO optimization. The model constantly evaluates its own output through reports and loops in language professionals who fix the most critical errors for you in the background.

AI translation gives you access to markets that would otherwise be out of reach, but it's important to prioritize your needs to find the right solution for your strategy. Certainly, the quality of translation is important to everyone. However, you should assess whether a more accurate translation is worth the investment. Perhaps start with a rough translation to understand the demographics and invest in fine-tuned localization after you've gathered some data.

These considerations are important because it's easy to get hung up on language support and quality alone. While it's nice to have a tool that can theoretically translate into 30 other languages, you may want to choose a different solution because it integrates better with another tool or offers the workflow your team needs, be it a customizable user interface or SEO features.

With Magnolia, you can choose between Google, Microsoft, Translate.com and DeepL in our translation module. We also offer integration with AT Language Solutions, a translation and digital marketing agency, for those who prefer to stick with human translation.

How reliable are AI detectors?

Finally, we need to address the big issue of recognizing AI. It is only natural that some people are hesitant to adopt AI immediately, and we should all ask questions to understand how this new wave of technology fits into our lives. We are not here to solve the ethical problem that arises from the use of AI, but merely to discuss how some are trying to address it.

Whether you want to perform quality control or find out if your AI tool is unknowingly plagiarizing on your behalf and putting your brand's reputation at risk, tools like ZeroGPT or Smodin's plagiarism checker offer a solution. The only question is whether they can deliver what they promise.

In most cases, unfortunately, you will be disappointed, and the reason is not the providers. Their intentions may even be good. However, due to the variety of AI models, detecting plagiarized or AI-generated text is not as easy as you would want it to be. That's why you often see promises of high accuracy rates alongside references to the possibility of false positives.

In recent years, we have all evolved to recognize blatantly mechanical language use that obviously betrays language models. It is also true that some models have adopted and plagiarized entire sections of text, which is where the model's awareness of ethical principles comes into play.

It's important for you to know that language models do more than copy and paste text, and that both production and analytics tools require us to adapt in an unusually short amount of time. If you have legal concerns or feel uncomfortable using AI, there's no shame in that. However, you should stay informed about the latest developments and at least be aware of the new possibilities in order to understand what your competition is up to.

So should you use AI in content marketing?

It almost seems as if the question is not whether AI tools should be used, but rather which ones. The choice is endless and will continue to grow over time. The biggest challenge for those just starting out is finding a tool that fits their needs and use cases.

However, in addition to their content requirements, some companies also need to consider their legal environment and partnerships. While there is no doubt that AI models can generate everything from blog texts to LinkedIn carousels. But you should at least consider the technical differences between the various products, be it in terms of the number of tokens or privacy policies. In any case, you need to be aware that the data you feed a model with can potentially be used to train it and derive rules that protect your company, its partners and customers.

All in all, AI models offer an incredible productivity boost for content production teams. If you like the idea of eliciting ideas from a bot or automating certain administrative tasks but can't decide on one, that's understandable. You should also keep in mind that in the coming years, the pace at which new AI tools and services will emerge will be even faster, so it's important to remain flexible to integrate and utilize new tools. In this case, a DXP like Magnolia's can give you the flexibility to switch between different tools. This offers the convenience of a familiar interface and a secure software environment, as well as the ability to switch to another tool later if it better suits your needs.

In this 3-minute video you can see how AI text generation, image generation and translation work in Magnolia.