Sunday, February 26, 2023

Using AI for Content Marketing: Early Thoughts and Findings

I write for a living.

I am not now, nor ever will be, any kind of an expert in Artificial Intelligence (AI).

I write ad copy, technical documentation, and scripts for explainer content. I create infographics, edit videos, produce podcasts, and run the full content marketing gamut from social media to PPC advertising. I've been doing this for over 25 years, and I currently work in-house for a well know cybersecurity provider.

Since November last year, my output has changed dramatically; I’ve been using AI tools to support our content marketing efforts.

However, using AI is a bit like editing sound: You put shit in, you get shit out.


AI has come a long way in the last few months. No longer just words of caution from the pen of Heinlein or Gibson, AI can produce high-quality content virtually indistinguishable from that of flesh and blood writers. This is down to how one uses AI and how we create prompts (instructions) rather than a natural ability for AI to write in an approachable way. I hear a lot of people say that AI (I use the new ChatGPTPlus) is instantly recognizable because it talks in a passive voice. There’s an answer to this, as with pretty much all current gripes about AI textual output. Tell it not to.

Here’s an example of the most basic and first prompt I use in ChatGPT for creating text content for work:

“Please act as a B2B copywriter. You will write in a positive and active voice. Your copy will be <brand trait>, <brand trait>, and <brand trait>. It will also be original and innovative. It will be targeted at experienced <target persona> and <target persona>, and should be of genuine value to this audience. It will use terminology with which they are accustomed. Each article or post you write should contain sub-headings of major points, and bulleted lists of sub-points or actions. Please expand on any content with real-world examples and/or citation (as appropriate), providing URLs to further reading if possible. Please write in American English, and finish your writing with a fleuron.”

One can even afford a few affectations: Notice how I always say “please,” just in case it’ll spare me during the AI/robo uprising? I also ask for an indicator that the AI has finished writing because, in longer-form content, it’s necessary to type “please continue” to get past character limits, so I like to clearly know when it's come to an end.

I’d consider this the very basic of prompts and start every new chat with something along these lines. Then I go into what I want the AI to do after giving it its “mode of operation” and voice. “Create me an 800-word blog post worthy of Mordor,” etc. Usually, this will include additional instructions like SEO keywords I want it to cover, topics or solutions I want it to touch on or consider, specific industries I want it to research, or whatever. Yes, I do edit any output - pretty thoroughly. I insert more brand voice, add product value to blog posts on general topics, sprinkle in a bit of appropriate levity, add hyperlinks, further relevant information based on my general topic knowledge, I might play with the tense, add a bit more SEO glitter (if needed), and often whole paragraphs of additional content where the AI has inspired me to do so. I sometimes turn back to ChatGPT to “please expand on <blah>,” “please explain <technical term/concept>,” or “please rewrite this in <blah> number of characters,” and I’ll always fact-check any references, technical output, or legal standards that it provides. AI has become my collaborator and writing partner, and (right now) that's definitely the way to get the best results. If you’d like to see more prompts like the one above, I heartily recommend the splendid work of Maximilian Vogel, and there's tonnes of stuff on GitHub (including prompts for writing code, etc.). Remember, any copywriter won’t be able to give you what you want if the brief is poor, so learn how to prompt before you expect usable results. You can dip into anything written on the TrueFort blog since January to see some examples.

Also, try things like “Optimize <sentence> to 75 characters to use as a LinkedIn ad description,” "Please give me some ideas for blog posts targeted at <persona>, discussing their most important professional concerns,” or once it already has your traits and targets (and maybe even existing sample messaging) “Please create six attention-grabbing call-to-action messages, and buttons, for a landing page on <blah> where we want visitors to <action>." The functionality is only limited by your input. Even if it’s not immediately what you’re looking for, I guarantee some of the output will be a catalyst for other ideas. Seriously - go play - the basics are free, and my new “Plus” plan is only $20pcm. Plus is available during peak traffic (which seems to be all the time of late), has a much faster response speed, and I get priority access to new versions and features.

I’ve started using it in other ways to help my day-to-day, such as asking it to summarize a page of web content or “Please rewrite <text, or even a URL> in a more accessible or more targeted way.” I’d estimate that using ChatGPT is already saving me around two hours every day, and that’s valuable time I need to get my job done. As anyone in content marketing will testify, short of time travel or cloning, there can never be enough hours in the day. If the ROI of using a tool like this has a productivity saving of two or more hours on a writing day, and you show that ROI and the optimized output to your CMO, I guarantee they’ll find 20 bucks a month in their budget. Mine did.

In support of any output, I also run everything through Grammarly. Mainly to check I’m using American spelling (which doesn’t always come naturally) as I go along, and to check for duplicate online content for SEO purposes.

Already, I’m more inclined to use ChatGPT as a point of reference rather than Google, as I can ask it specific questions about, for example, controlling lateral movement and adopting zero trust best practices, rather than rely on Google pulling back other people's marketing content or, more annoyingly, my own. I’ve actually added a simple web link to my iPhone home screen.


Again, check out our blog. Every single header image (since November) has been generated with MidJourney

This text-to-picture artificial intelligence service empowers users to create art, from (virtually) photo-realistic to the work of Hieronymus Bosch, based on text descriptions. It’s far from intuitive to use, with a plethora of optional fields and seemingly zero instruction manual. It also uses Discord as an interface, so it’s all new unless you’re fourteen years old or an online gamer (which, thankfully, I am). I dipped into MidJourney a while ago to generate images for our online D&D games but soon ran out of the credits needed to make more artwork. A small investment, however, has meant my VTT players now have the perfect graphical representation of who or what I’m thinking.

More recently, I've begun to use this for work.

Putting this in the context of ROI again, MidJourney is $8 a month and takes me 30 seconds to get what I need (content, aspect ratio, angle, depth of field, etc.). In contrast, Shutterstock is $19 a month, takes at least 20 minutes to find what I need, and that will invariably be a compromise. UnSplash and Pixabay have limited assets - and are even worse time vampires. MidJourney is the very definition of disruptive innovation.

For the imagery on our blog, I augment the results with Photoshop to get that distinctive duotone that is our ongoing brand style, but otherwise, the output is perfect. We use a lot of perspectives and patterns in our imagery, and it replicates and topic in this style seamlessly. I’ve also used it in other ways, such as asking for suggestions for convention badges and custom art for mood boards (after uploading our logo and existing brand imagery) or asking it to suggest layouts for infographics.

Learning how to prompt isn’t immediately intuitive, and at first, I had to trawl around forums and communities, hoping someone would throw me a bone. I found the work of Lars Nielsen, Kris Kashtanova, and Guy Parsons very helpful. MidJourney will also test your visual chops above and beyond the subject matter itself, and I've found that my experience with photography, graphic design, and film work has helped a lot in creating prompts. It responds best to camera directions (symmetrical, low-angle, full-body shot, cinematic still shot, etc.), art movements (minimalist, film-noir, brutalist, pop art, etc.), F/stops, lighting types, and styles of photography (landscape, underwater, tilt-shift, still life, etc.). It even responds to prompts detailing the different types of camera you want to replicate (Nikon D850, disposable camera, Polaroid, Canon EOS R, etc.) with added lens types (telephoto, wide-angle, 85mm, neutral density filter, etc.). You also try directors in the prompt, such as “in the style of Sergio Leone” or “in the style of Wong Kar Wai” for that elusive mise-en-scène, so that quinquennial paying off loans from film school might have been worth it after all.


AI video is still in its infancy, but it’ll probably be as early as 2024 when we start to see script-to-explainer video content hit the market. Text to AI-voice and AI audio leveling/editing are already commonplace. I expect us to see a lot of content support-specific AIs turning up in the next few months as the industry realizes the potential and the developers who were victims in the recent spate of Silicon Valley redundancies get to grips with the associated APIs. Expect a crop of low-budget filler, but watch out for the unexpected coming from home developers - building their own Jarvis on any old skip-scrounged kit that’ll run Python. 

Also, expect some serious low-quality copy and a fall in news standards in the form of “black hat journalism.” It’s a simple matter of linking a few APIs to automatically write and publish magazine-style content to a cobbled-together portal for skimming off those referral link dollars. Without good prompting and some appropriate manual editing, we’re going to see some mediocre bollocks that’ll put a black mark on AI's journalistic copybook - and the real press will (justifiably) rush to point fingers and scoff.

Quality content is king. Google has already made a statement about this, but clearly says it has no problem with AI if the copy is of genuine value and will “reward high-quality content, however it is produced.”

Nation-state bad actors and hacktivist are going to ride ChatGPT like a pony, all the way down the misinformation highway to QAnonville. Now anyone can write copy and code, at volume, and the code side of AI functionality is bound to grow even faster as code writes codes. Brace for automated factual disruption on a Brobdingnagian scale. Throw in a couple of Deepfakes, and the 2024 US election is going to be a disinformation masterclass.

Obligatory call to action

Honestly, I think our industry needs to get on board, or we’ll be left on the dock.

I’m not saying human-generated content is dead, far from it, but a human/AI collaboration in a business environment feels productive, and creative, and it’s already happening.

We should look at AI now and consider how it can make what we do more efficient. It’s the copywriters, PR peeps, and content creators who don’t take an interest and don’t integrate AI into their workloads who will find themselves on the sidelines in favor of the people who do. I have a couple of extra hours of additional capacity a day because I know how to use the tools; why would you give my job to someone who doesn’t? Have we reached the point where we should be sharing the by-line with an AI? No, not yet, but I bet we're only a year away from Elon trying to charge us for a tick to prove we're flesh and blood. Expect "prompting" (or whatever they'll call it) to become a part of media courses in the very near future, but those joining University courses in '23 may well be training for jobs that won't exist when they graduate. Things are moving fast, the ball is already rolling, and we can’t stand by like print journalism once did, denying the inevitable.