There is currently a lot of excitement mixed with doom and gloom about AI in the world. Though the next decade will be as monumental as the modern internet, as this new technology displaces jobs in all industries and levels it will also enable those with passion and skills to do jobs they weren’t fully equipped to do before. This is what we should be excited about.
A few weeks ago, my daughter and I created a completely fabricated hotel brand called MRLEE Hotels. We made a golf course on an asteroid, a restaurant in a volcano, and an eco-retreat that had breathing walls. We accomplished this and created a 27-page hotel brochure in less than 7 hours with the assistance of AI.
This project was fun and exciting to create. We enjoyed imagining, dreaming, and working on this project together. You can view it here.
Today, I want to share my next creation,
which goes in a slightly different direction. It's a fully functioning travel site (an alternative to TripAdvisor) focused on activities for families in 237 cities across the world.
That's 237 different cities, 10 activities per city, making 2,370 activities. Each activity has a 1,000-word article describing the activity, making 237,000 words in total across the site. Plus, each activity and each city has a photo, making 2,607 images. There are also over 300 customer reviews that give positive experiences of the activities.
All 237,000 words, 400+ reviews, and 2,607 photos are generated by AI and cost me $53 to produce.
You can also watch the video of how I created kidventuresAI and the prompts to generate the AI content below.
Is this the future of content?
Let me break down how I got here and how content sites need to adapt to what is coming in the next few years.
When my family and I travelled, we would use different services to find things to do. We would search sites such as TripAdvisor or Google Local. The problem was that these sites were mainly used by tourists from western countries. As we were travelling near the end of the global pandemic, these sites had no reviews. We would go to hotels in Thailand or Sri Lanka, and the hotels would beg us to leave reviews on TripAdvisor after our stay, even if we had not had the best time. They just wanted the world to know that tourists were visiting.
So, if being listed on TripAdvisor or Google Local was so important to these hotels and restaurants, why didn’t they just create fake reviews and list them? The simple answer was that many people thought, "that's wrong," while others said, "if we get caught doing that, we disappear off the internet (being on TripAdvisor is the internet to a restaurant in Kandy, Sri Lanka), and then nobody can find us."
If TripAdvisor and similar sites had no reviews during this time, this opened a gap in the market. But what to create?
My first thought was around things to do in cities as a family, as we could never find great family activity lists on these review websites. I thought this was a gap in the market and maybe a new business venture, but how do you collate the 10 top activities for each city in the world? How do you know those activities are interesting to children? We had visited 20 cities, so that was a start but with thousands of cities across the world what do you do. We could use ads to drive revenue, and then we could share revenue with families who write articles when they visit different cities and write about what activities are available for kids. After thinking about it for a few hours, I put the idea to bed as my solution wasn't that much better than what was there already and not feasible since it was too human-intensive.
And this little service called ChatGPT launched when we were heading to Seoul, South Korea. I checked TripAdvisor for things we could do in Seoul. The first few pages returned nothing interesting, so I asked ChatGPT, "what are 10 things you can do with children in Seoul?" Number one was the Children's Museum of Seoul (surprisingly, this wasn't listed on TripAdvisor). This was probably the best thing we did in Seoul after building a snowman in the car park of the DMZ.
Fast forward to March 2023, and for $53 (and my time, of course), I have created a global information site with over 2,370 pages of activities for families wanting to travel with their kids and do interesting things in 237 cities around the world.
The simple question here is:
does that mean it's over for information/review sites if something like this can be generated without human writers?
Now that I have this setup, I could create websites for the top 10 hotels, restaurants, museums, parks, coffee shops, conferences, business centres, etc., for each city in the world. It's 2pm right now, in the next 24 hours, I could have 10 websites, each with over 2,370 pages of SEO content on these topics, and all for less than $600. Make that $6,000, and I could have each site in nine other languages too.
$6,000 for 23.7 million words of pretty good multilingual content 🤯
I appreciate that I have a skillset that means I can create this/use the API of the internet known as ChatGPT. But it won’t be long before John or Sara from the social media department for Visit [insert country] can write in natural language a series of prompts to create the same output I have shown here.
Survival in the AI age
It may not be a surprise to hear, but the internet is not full of high-value content. There is apprehension about this technology currently labelled AI. Some fear it and believe it to be nearly sentient, while others believe the majority of the output is predictable and low quality.
I sit somewhere in the middle. I think we have mislabeled the technology as AI, which creates visuals of Hollywood films and the Terminator. A more fitting branding could be machine learning or data science, though they don’t sound as sexy to the public.
The one thing I do know is that with each day, AI-generated content is getting better. The models are getting better, and more importantly, we humans are getting better at prompt-ing them.
But I don't think the game is over for content-based businesses and here are four Areas Where content sites can grow in an age of AI:
LLMs have a known issue around factual relevance, also known as hallucinations. These models have no intelligence per se. This will likely improve as the technology develops with the inclusion of citations. But the generative AI space is exactly that. It is generating the content, not summarising what is already there based on knowledge of the truth.
A win for content sites is human-written content or human editorial content becomes attached to the individual writer and their credibility. Combine this with an immutable ledger (i.e., blockchain), and you can hold writers accountable for what they publish. They can also benefit from a decentralised system of validation for their work.
I can see in the near term where an article you read has a badge/stamp that reads "human-written," "human-edited," or "human-fact-checked" with a digital fingerprint of who and when that was done.
Personalisation, to me, is the cherry on the cake. Imagine being able to create a space where individuals can come and share their preferences, much more than a browser cookie can hold, and where content is generated following your brand's data model to generate relevant and personalised content. Why, when I visit a news website, is it filled with doom and gloom about things happening on the other side of the world? Why can't I default to dark mode when I visit any website? Imagine visiting KidventuresAI, and instead of showing you 237 cities with 10 activities for families, it shows you three cities: one that you have visited before and enjoyed, one that you have said you would like to explore, and another that matches your other preferences. Speaking more broadly, websites in general are going to shift from displaying generic content to generating personalised content and information, but again, that is for another article.
Content sites have an audience, and as content becomes more generic or easily generated, you will no longer need to go to the BBC website to find out what happened in X city on Y date. Instead, you can ask your AI. What your content site currently has is an audience of people who share a common passion. I see this with a podcast called Beautiful Anonymous, where the host Chris Gethard talks to a random individual for an hour every week. It has been going on for seven years, and I have listened to every episode. I have also attended two of the live shows in London, and if I was in the USA in May, I would be attending the Beautiful Anonymous conference in Brooklyn. How can you capture your audience to bring them together in new and interesting ways?
AI hasn't taken over the world, and Skynet isn't around the corner yet, so there are plenty of ways content sites can be more creative. Using AI-based technology is, of course, the first step. Expanding to different mediums is another option. Why not turn your written news site about the legal field into a podcast or video stream? Having more time to focus on what your audience wants, why not create courses, training, or templates? Or build products to solve the problems of your audience? Many tech companies, once they hit their stride, create media entities as a way to get more customers. Why not do it the other way around and ask your audience what problems they need to solve?
There are many areas content sites can embrace in this AI age. They just have to want to and not become Blockbuster.
Lessons from Using DALLE and ChatGPT
I want to finish off this article by sharing a few interesting things that occurred during the development of KidventuresAI using DALLE and ChatGPT.
Firstly, the API would error a lot. If the API had been continuous, I could have generated all this content in less than 5 hours with some parallel processing. But I kept hitting the "API is overwhelmed" error as global demand continues to climb.
Secondly, I got to see what's under the hood of OpenAI and how it blocks requests through its trust and safety layer.
For text, I was able to generate all the activity content, run some secondary manual checks, and so far removed one sentence from one activity article that wasn't appropriate. But there could be more, which I will discover or be made aware of over time.
Images via DALLE were more interesting. I was unable to generate images for the following prompts:
Juba, South Sudan
Cockburn Town, Turks and Caicos Islands
Road Town, Virgin Islands, British
Charlotte Amalie, Virgin Islands, U.S.
Prompts 3, 5, and 6 are understandable why they were blocked. Kyiv and Sudan you can make some assumptions as to why they have been added. But we can see that OpenAI is being on the side of caution with image generation, understandably. Some of the images, mainly those around people, weren't great. There is even one where a hand has six fingers.
So now I ask myself, do I add a site map, point to Google Webmaster Tools, and wait for SEO to do its thing? Could my $53 investment become a competitor to the likes of TripAdvisor or Google Local? If I did decide to monetize it, am I cheating?
What do you think/feel about AI-generated content? please let me know.
I am trying something new for the age of AI and outlining how I am using AI to assist my content creation.
This article went as follows:
GPT3.5 gramma checked
Human read-through and human alteration