The Expert Machine Blog
What we use AI for in content generation
There are two main AI-related concerns you’ll have as a potential customer of ours:
“Why use a content agency at all when we can use ChatGPT?” and “Will this company write everything with ChatGPT and ruin my Google rankings?”
For the first question, you already know the answer, otherwise you wouldn’t be reading this in the first place. You’ve already searched for an agency because you tried to use ChatGPT (or Surfer, Jasper etc.) and found the results to be lacking.
For the second question, the short answer is ‘no’. The long answer is below.
How we use generative AI
We use generative AI tools for:
- Creating article structures
- Drafting first rounds of some creative work depending on the topic suitability
- Idea generation for going further in-depth
- Quality checking based on certain criteria (in-house custom GPTs are sometimes used, but aren’t 100% reliable yet)
- Generating keyword cluster ideas as a first step in feasibility research
It’s rare that we’d publish a fully AI-generated article, because it’s rare that the quality is good enough for our goals. In theory, it might be possible. The tech isn’t there yet (and I don’t actually think all of the roadblocks can be solved by better tech).
The difference between you hiring us as an agency and you using ChatGPT to do it yourself is:
- We use massive custom multi-step prompt chains (bespoke for each client) to get LLMs to give the most useful outputs
- We structure each article with reader intent in mind, informed by both SEO research and topical knowledge
- We line edit, proofread, structural edit, and fact-check everything we publish
- We find original quotes from academic and industry sources that bring real human and scientific insight to each article
- We always push for information gain – adding new knowledge to the world and the web, rather than just remixing
All of these are key ingredients for making readable content that humans respond to. They also take aim at Google’s requirements for helpful content – not something your average marketing exec can check off a list in five minutes via ChatGPT.
Why not generate everything with AI?
Auto-generating every word of an article just isn’t very smart. Readers are getting bored of it. And so am I.
Consider this sentence, from a competitor’s website:
“Collaborating with influencers can be a powerful method to expand your blog’s audience and enhance its appeal. By carefully selecting and working with influencers, you can leverage their influence to grow your blog, all while providing valuable and engaging content to your readers.”
This post gets approximately 0 traffic, and you can see why.
It’s blindingly obvious that an LLM generated this – almost certainly GPT 3.5 or 4. Sales copy always turns out the same from this engine; sanitised, corporate, overly wordy, and horrible to read.
The almighty tell is the word ‘by’; “by doing X, you’ll see result Y” is a common cliché that ChatGPT spits out to conclude any idea it’s trying to explain.
In our AI editing workflows, this cliché is one of the cardinal sins for a writer. It’s removed every time.
Writing is a vector for idea transmission, AI doesn’t have ideas
At the risk of getting philosophical, words are carriers of ideas. Not creators of them.
LLMs do have emergent properties of creativity, but it’s my belief that they spark ideas in the minds of humans. They don’t actually have ideas themselves. Anything ‘new’ coming out of an LLM is still just a statistics machine mashing two strings of text together. It’s not a mind.
AI-generated text isn’t really saying anything at all. But it’s easy for humans to think it is.
If AI-generated text helps someone – which it absolutely can – then it’s worthy of existence. If the process of helping someone results in them becoming a customer of whoever publishes the text, I don’t think that’s morally wrong.
Bulk AI can sometimes work
We are experimenting with some programmatic AI content on Expert Machine’s website because SEO isn’t actually a great lead generation channel for our type of agency. LinkedIn, outreach emails, networking and referrals give a much better ROI as a starting point for our sales process.
So there’s not much point investing in human writers, when we could get more immediate results from other channels. Not every SEO agency has to have great SEO for their own website (only for their clients, of course).
Tons of agencies waste a LOT of time putting out dull posts like “how to edit a thought leadership article”. I’m just not interested in doing this.
In my opinion, if you’re searching for topics like:
- how to work with an SEO agency
- what to look for in a content agency
- content writing agency
- content writing agency uk
It means you’re already on your way to a purchase. You’re unlikely to read an entire blog post – you’re probably going to click the top-left logo, head to the homepage, and check out the offer. This is why I believe TOFU (top of funnel) content isn’t really worth investing in for our company type – at least while other channels are performing better.
We can’t claim to be the best content agency out there. That would be a bit silly. But there’s not actually that much differentiation between them. And most customers don’t need the very best – they want one that has a good track record of results, is available right away, doesn’t cost too much, and has friendly people. That’s it.
So, seeing as we match those criteria for many clients, it makes sense to simply show up before others. The sales process shouldn’t be massive; we’re not looking at million-dollar deals with huge conglomerates. We don’t have to spend hours of internal resource churning out ‘thought leadership’ trying to appear 3% more attractive than the competition. In many cases, we just have to show up and deliver.
So with that said, we can experiment with bulk content that’s 80-90% AI-generated. We’re not doing fully programmatic SEO; there’s not enough depth in the basic topics to warrant doing it in that much detail. But I think some automation to plant a flag on certain keywords does make sense, as long as we’re respecting the reader’s attention and not wasting their time.
The important thing when generating AI content is to make sure it’s edited by humans and make sure it’s representative of your brand voice.
I’d only consider publishing AI content if it’s:
- backed by proper SEO strategy
- genuinely informative
- prompted thoroughly
- not packed full of ChatGPT clichés
It’s not in Google’s interest to ‘clamp down on AI content’
Anyone working in SEO recently will have seen the spectacular downfall of Jake Ward’s ‘heist’. As Ryan Law points out on the Ahrefs blog, Ward essentially cloned a competitor’s sitemap, rewrote all the content with an AI tool, and gained a huge amount of traffic. His viral tweet boasting about it got tons of attention, and the next month, Google whacked a manual penalty on the site and its traffic disappeared in an instant.
In my opinion, this wouldn’t have happened without the tweet. Google’s reps were embarrassed that someone had gamed their system so easily, and shut it down to save face. They don’t want everyone on the web doing the same. Fine.
But then again…
Why would Google want a web free of AI-generated content, when they make one of the world’s most capable AI content generators themselves?
First known as Bard, then Gemini, Google is going in hard on consumer generative AI. Your entire Chrome and Docs experience will soon have an LLM tacked on to every task. Gmail can ‘write’ emails for you now (it’s been suggesting idiotic reply templates for at least five years, but we’re supposed to trust it now, because AI).
They’ve tried baking it in to search, with the SGE (Search Generative Experience), but it seems stuck in Labs Limbo, potentially facing the chop like so many other failed experiments. Halfway through 2023, the emergence of SGE threatened the entire business model of an ad-supported internet, and we all felt a little uneasy. Then many of us realised that by doing this, Google would cut off their own revenue streams. A pointless exercise.
And it turns out that consumers don’t really want to trust Google for the ‘truth’ about absolutely everything – they want to follow links out to publishers, and judge the provenance of information themselves. But Google still tells us AI is their future.
In short: Google is a provider of generative AI tools. Attempting to ban their use is in direct opposition to this. SGE is useless, dying, and won’t replace publishers of SEO content.
Oh, and do I even need to mention that AI content detectors don’t work? Even if they had the incentive to do so, how would Google actually enforce the rules? What if you generate an article with AI, then edit it? What if you only generate 50% of it?
We follow the Helpful Content Guidelines – the provenance of raw text doesn’t matter as much as its helpfulness.
So AI-generated content won’t inherently harm your Google rankings – but readers will prefer your competitors’ content if it’s done by humans. And that means they’ll beat you in the search positions.
We spent ten minutes trying to get DALL-E to generate a feature picture for this article, but the results were just so cringeworthy we went with a photo a human took instead. Props to Petr Slováček on Unsplash