So, AI is taking over… or is it?
Well, one thing’s for sure, based on the fact that covering this topic has been highly requested – the hype is real.
- The content may not be factually accurate.
- The content may not be original.
- But… at least we know the hype is real.
Disclaimer: This post contains strong opinions from our CEO (also the person writing this disclaimer, Alex Panagis, as well as other members of our team). We’d love to hear what you think about the state of AI and how you’re approaching it.
Feel free to connect with us by joining our email list (form at the bottom of this page and our homepage) or by Tweeting @alexjpanagis…
So, without further ado – let’s dive in…
The Problem With Using AI
AI text generation is quite simply a language model. This comes with advantages, namely:
- It has “consumed” more resources than a human would be able to even in multiple lifetimes.
- It can generate text quickly.
- It doesn’t have much downtime (well, at least in theory), meaning it doesn’t suffer from mental blocks, etc.
But, as a byproduct of this, it also has some fundamental weaknesses:
- It passes off fiction presented as pure fact.
- It fails to understand context – how something fits into a bigger picture, what someone cares about, why, etc.
- The type of content it can generate is often rambling instead of focusing on helping readers achieve a specific objective, and getting points across as efficiently as possible.
- And last but certainly not least: it isn’t capable of going out into the world and actually doing things (say, for example, deploying Docker, building web applications, optimizing WordPress performance, testing something), so it cannot have real-world experience, conduct research, or take a stance.
That said, it is very good at pretending to do all of these.
Let’s Get Specific With An Example: The Future of WordPress
Instead of just speculating, let’s lead with a head-to-head comparison. Here’s what AI text generation says about the future of WordPress:
Now ask yourself if this is what you would expect from a person who’s worked in WordPress (or web development and design generally) for years, and has personal experience with both WordPress and other solutions.
I think we can all agree that the AI-generated text is bland, unresearched, lacks references, sources, and is just generally quite a vague impression of what the future of WordPress is. In short, it’s clearly rambling the same way you’d expect an inexperienced developer in an interview to talk their way out of a question they don’t know the answer to.
I mean, particularly with lines like:
- For example, WordPress may incorporate more responsive design features and tools for creating voice-activated interfaces – As if to suggest that it isn’t possible to design responsive websites with WordPress. 🤡
- In recent years, WordPress has been expanding its capabilities beyond traditional blogging and content management – WordPress has long not “just” been a blogging and content management platform. 🤡
If you find this suitable and/or usable for your website, that is great, and there are likely quite a few companies that will benefit from using AI to replace what was previously possible to do with the help of inexperienced writers.
Now, if we take a look at what we would expect from a writer at an absolute minimum (i.e., an introduction for a thought leadership piece on the future of WordPress so, consider that this realistically only begins to scratch the surface):
Even Human-Written Content Struggled With Originality
There are many who maintain a firm stance on preserving the integrity of human writing.
In fact, many suggest these are greater in numbers than those who support AI, but these are just the quieter ones (presumably because they aren’t trying to sell you an AI product/service or course on how to use AI 😉).
But, taking the opposite stance isn’t the right approach either. I believe hedging bets against AI would likely put you in a losing position.
And there are a few reasons why I can confidently say this.
The main one is that even the majority of human-written content struggles with the same problems that AI has. Approximately 0.0003% of writers who have applied to work with us (or are sourced through various recruiting efforts) have actually progressed to the next stage of our vetting process.
Very few people bring true talent to the table.
Yet this level of talent, experience and high level of thinking (all of which are characteristics we look out for when hiring) are not always necessary, depending on the nature of the content you produce.
Creating content for organic social posts, spinning out tons of ad variations and programmatically testing them, and writing simple product descriptions – all of these are things that we don’t have people on our team tackling. This is purely out to protect their schedule, because they are such low ROI activities.
And these are all great examples of when a human is now able to augment their work by using one of the many, highly promoted AI tools. It does make sense to automate such tasks, allowing them to be completed faster. Because doing this allows the human writers to spend less time on work where their input realistically has very little impact, and more time on tasks where it does (and where AI simply can’t compete).
Focus Human Energy On The Highest-Value Tasks
At least for now, this is the right approach. Depending on the nature of the content, AI either has no place at all, or a limited place. When it has a limited place, that is generally in the form of the writer using them as an aid to start getting content onto the page.
Using AI merely as an aid in this way allows the human both energy and time to be devoted entirely to the highest-leverage parts. This includes the structure, original thinking, introductions/hooks, outros, and whatever else makes the specific piece of content better than anything else available. It’s the human creativity and original thinking that will make people want to read their specific article on a topic vs. absolutely anything else out there.
A Side Note About Using AI As A Force Multiplier
This is a side note that I wasn’t planning to include initially because it seems like a very harsh, biased take, but it remains valid for the work that we do.
The effort of trying to use AI tools to improve your workflow often outweighs the reward.
Maybe I’m just a better writer (or just worse at using the tools?) than most people, but I find that working with AI tools as an aid tends to hinder me more than help. It definitely depends on the type of content, but I don’t find it that difficult to get my thoughts into a Google Doc (when I manage to find the time to write for our blog alongside all the other work I do!).
That being said, I guess this is compensated for by the time AI can save on other types of content. By cutting down the amount of time needed for previously unskilled work to be completed, I can focus on writing posts like the one you’re reading right now, which is essentially a stream of my thoughts on AI documented over the course of a relatively long period of time.
Uses Of AI We Can Endorse
Some other uses of AI that we would endorse are:
- Summarizing
- Creating lists
- Creating tables
- Putting together structured data
- Automating simple, repetitive tasks
- Creating complicated Excel formulas
- Certain types of content ***
*** There are very strong reasons for our views on this:
1) We have lots of data from sites that were using AI-generated content, and which quickly began suffering as a result of search engine algorithm updates. This is definitely something we can see continuing to be the case.
2) Don’t assume that because it worked before, that it will continue to work. This is a fundamental principle of operation for how we approach the things we do as a company, not just content. And, as it stands, AI content is essentially a new form of article spinning – in that it leads to lower quality content that is less likely to land any links, and less likely to produce any usable outcome since it contains no original thinking. None of its output can be considered truly original. Yet this doesn’t mean it has no uses. As mentioned earlier, it just means there are select uses for it.
Make The Effort To Stand Out
Our mission remains to make the internet a more helpful, informative, and enjoyable place for humans. And, in my view, this intent is stronger than ever.
AI tools use a language model. This means that it doesn’t have any idea whether what it’s writing is factually correct. It still falsifies citations and statistics, and attributes quotes to people who don’t exist!
And while this may change, the current issue is something that a great many site owners are already observing, and will continue to see for some time yet. Because with AI commoditizing cheap, half-decent content as something that can now be done instantly (at virtually no cost) and with little involvement from a human – Google is doubling down on EEAT (Experience, Expertise, Authoritativeness, and Trustworthiness).
And don’t imagine that it’s going to be as easy as having content that just appears to have a “little extra” work on top compared to a piece of content that could have been entirely AI-generated.
Everybody can produce average content now.
This places the work that we do (and others that really put in the work) in a stronger position than ever. We’ve always operated on a completely different playing field to the majority in our industry, who have played into the mass market that is producing large volumes of content, all of which are quite generic, basic, and lack original thinking.
And more than ever, this sets apart what we do from what most people can do. People are starting to realize – as Joost (founder of Yoast) is saying – how difficult it is to truly do 1% better.
Content that is just marginally better is already difficult to produce.
Content that contains original thinking, is written by subject matter experts, and that people actually want to read is more valuable than ever.
It’s also something we’re seeing companies valuing more than ever too.
And it’s easy to see why, because this type of content is quickly becoming the only type that will be capable of generating results in search listings. Not to mention performing strongly in all these other areas:
- Our content brings us new customers. Our articles rank well in Google, and generate a consistent stream of relevant “leads” for both us, and our clients. These leads are in the form of people joining our email list, community, etc. And as long as we promote the functionality of the product within those articles – this inevitably leads to conversions.
- Our content improves the retention of existing customers. This is particularly applicable if the product being sold is somewhat complex, and has a learning curve (higher time to value). By publishing lots of educational content we are helping our users figure it out, achieve their goals, and stick with us for longer.
- Content geared towards helping our customers (and potential customers) run a better business of their own. Even if our content is not directly associated with the product, if it helps our audience run more successful businesses of their own, then a highly probable byproduct of this is the increased likelihood that they remain a customer of ours for longer.
- Our content helps reactivate past customers. The common re-activation technique is to send your churned users an email list of all the new features added since they left. We showcase our product wherever relevant in our own content, so this mechanism is built-in there.
- Our content helps with revenue expansion. By reading our articles, our users might discover some functionality that is not available on their plan (and upgrade), or by helping them grow, we teach them to scale their operation, leading them to upgrade to a larger plan.
- Our content fuels the Word Of Mouth growth. When we teach our users something cool that helps them be more successful and achieve their goals, we’re giving them a “talking point” to share with their friends.
- The traffic that we get to our content leads to the “Mere Exposure Effect”. The more eyeballs we get on our articles, and the more we mention our product within them, the more “familiarity” we create in the minds of prospective customers.
- Our content helps us build a solid reputation in our field. People want to buy from someone who knows what they’re doing. By publicly sharing our knowledge, we promote ourselves as credible experts that you can trust (and buy from).
- Our content fuels our paid acquisition. Many pages are essentially sales pages for our product (this is more applicable to landing pages).
- Our content helps to remove strain from our customer support team. The more educational content you create, the fewer support requests we’re going to get.
- Our content helps us “get featured”. The chances of people talking about what we’re working on if we aren’t mentioning it on our own site are a lot lower than they would otherwise be. Create a compelling piece of content, and there’s a good chance that it will get picked up. Our content should get people talking about what we’re working on, and our product.
Other Ways In Which We See AI Playing An Important Role
1 – It turns average content into a commodity
And…gives people the false sense that average content is going to be good enough to cut it. Something that may be true in the short term (until Google eventually categorizes a site’s content as categorically unhelpful, at which point its traffic will dwindle).
This is already something we have seen first-hand, because we have had people come to us asking for help recovering sites when their traffic tanked.
2 – Their classifiers are actually quite good ways of making sure that content contains original thinking
The fact that AI is available has made us question our own human-written content.
Looking inwards to ask whether just about anyone could have written this, or if this is something that we can truly and proudly consider entirely original is an essential part of our philosophy.
The outcome of this is net positive. We are making sure that everything we are producing is worth producing.
It’s worth noting that, in our experience of using a range of AI detectors, the proportion of false positives is very high. We’ve run a number of tests on content we know has been written by AI, and content we know has definitely been written by a human. We even tested content written two years ago – well before GPT-3 was available.
While we found the classifiers reliable in picking up on content that was written by AI tools, there were a significant number of instances where they thought human-written content was AI generated. In particular, we found this to be the case for content that provided definitions of concepts, procedures, and lists of steps or examples.
Our feeling is that if these classifiers are, as yet, unable to be relied upon entirely for their accuracy, it is preferable for them to err on the side of caution and tend towards suspecting AI content. This tendency encourages us to strive even harder to create content that is truly unique, and doesn’t in any way read as though someone else could possibly have written it.
3 – It undeniably makes for some great jokes…
4 – AI has the potential to be a force multiplier for great writers
While we accept that AI is certainly making some astonishing steps forward, we still feel it’s too early to say that it is truly an asset to a good writer’s toolkit – at least in terms of the use cases that we currently have for it.
But one thing’s for certain, and that is that it definitely looks as though it will be at some stage. For now, we are cautiously adopting it into our workflow in slow, thoughtful, and, most importantly, measured ways.
This is both in the work that we do for ourselves as well as those select customers who are interested in us experimenting.
The reason for our caution, and our careful measurement of every step, is that in most cases where people claim that they have been able to rank AI-generated content, the results are actually very short lived. What works for the next 1-2 months isn’t necessarily going to continue to be possible. Our commitment remains to produce original content and research, with ideas matched with genuine experience – content that we know readers truly want to read from us.
Long-Term, Will Search Behavior Change?
I’d suggest taking a look at Google’s latest announcements relating to AI, what they intend to do themselves within the AI space, as well as their stance on AI-generated content.
I’ll add to this to say that it is interesting to see this space develop. And, although I can currently say for certain that I don’t prefer the experience of using AI chatbot-style experiences to Google Search (as many others don’t either), this remains something we have to keep an eye on. As would any new tool or development that has the potential to change consumer behavior, and where our customers (whether B2C or B2B) spend their time on the internet.
For now, I often jokingly liken AI-generated content to Facebook Messenger chatbots from 2016 – because of the hype and attention that they get for what they’re actually capable of doing.
That being said, it’s quite clear that this is different, and continuing to develop at a rapid pace. So whether, over time, our search behavior will tend more towards looking for AI snippet style answers vs. searching and clicking straight to the source is unclear.
However, even if search behavior does change so drastically, the reality is that people turn to search engines to find resources, not to get direct answers to every question they ask. Sure, some people expect direct answers from Google, like asking who an actor is, who plays in which movie, etc. But for more complex searches people want to find resources, often opening and reading multiple results to make their own judgment. Take, for example, a decision such as choosing between MongoDB and MySQL. I may want to read a post from Slack, Atlassian, and Figma to understand why each of them uses the solution they do.
Not from AI text-generated content – but based on experience from their CTO.
And yes, I’m sure some are thinking, “but surely AI (such as ChatGPT) is going to base its answer to your question on all the information it has from all the CTOs at dozens of major tech companies?”.
This is something that may very well be the case eventually. But the problem with this is that you lose the context you would otherwise get when you read a post such as Trello’s blog back in 2012 (followed by further updates to it in 2016 and beyond) about their tech stack – as well as other such articles. This forms a key part of your research, allowing you to make your own judgment.
You are instead trusting a computer to make the judgment for you, despite it not knowing your use case, without knowing the slightest thing about what you’re building, and it giving you no idea at all about where it is sourcing its ‘facts’ or ‘opinions’ from.
To me, if I’m being entirely honest, I find it hard to picture ever using this approach, based on what I use search for today. That’s simply because I like to know the true creators of what I’m reading. I mean, many argued that voice search would be dominating right about now too, and that’s also far from the case.
And, in a scenario where this was to happen, people would still want to read content from specific sources, i.e., going to Ahrefs to read about SEO (and hopefully our guides! 😉) because we’re trusted sources.
So at least short-term, even if “average content” has been commoditized, it isn’t as if thought leadership, style content, original thinking, ideas, and true experience can ever really be copied.
In the same way, people still (and will continue to):
- Buy books from specific people.
- Hire consultants for guidance because they want their specific input.
- Pay for courses from specific people because they want to know how they did (and continue to do) things.
All of which are also content – just packaged up in different forms.
If people will continue to have an interest in the higher-leverage content that they need to pay for, why would the interest in free content fade, unless it’s a response to a lack of original free content?
Services We’re Keeping An Eye On
I hope you enjoyed this post. As mentioned towards the start of this post, we’d love to hear what you think about the state of AI – and how you’re approaching it.
Further reading: NY Times Opinion: The False Promise of ChatGPT
Feel free to connect with us by joining our email list (form at the bottom of this page and our homepage) or by Tweeting @alexjpanagis!