You’ve seen these atrocities, lurking in dark corners of the web. “Salt Lake City” recast as “Sodium Lake Pond”. Pages of faux Shakespearian text generated by a Markov model. Scraped gibberish that drunkenly lurches between “finding the perfect girlfriend” to “essential oils” (um, just don’t ask…).
Mechanically generated drivel. For Free. A bargain, eh? The typical ‘content generator’ didn’t inspire much respect in the early days.
People laughed at the first steam engines. They were cranky and dangerous, generally considered more troubled than they were worth. A good solid worker could outproduce them all day long. Horses were even better.
Except over time, the steam engine gets better and the worker gets tired. The steam engine won. There’s a Honda in my driveway, not a horse.
While we were laughing at them, the robots are sneaking up on us. One of my personal side projects is tracking the evolution of artificially generated content. It turns out we’re starting to make some real progress in this area.
What Content Actually Matters?
I’ll start with the obvious – AI generated text isn’t even close to generalized artificial intelligence. You can’t point and demand generate articles worthy of Noam Chomsky. The Economist has nothing to fear.
But then again, we don’t actually need that level of quality to make money. High end opinion pieces and sales letters are maybe 1% of the real world web. The other 99% is… padding. The bottom 90% is pretty fluffy padding.
I need high quality web content in a few areas:
- Opinion and advocacy pieces
- “Top of Page” customer guidance
- Landing page copy
- Product descriptions
In short, the stuff that everyone reads and uses to make decisions. The content which drives clicks, sales, engagement, and attracts new fans.
Except that we’ve managed to create a digital eco-system where websites need to post loads of irrelevant text to be successful. Hundreds of words of drivel to convince a robot that a web page is about a topic. Consider..
- Google’s search engine crawlers use text content to establish the relevance of a web page to a particular search query.
- In many cases, Google also uses the degree to which a web page comprehensively addresses a topic to determine the rankings
- Many display advertising buyers are wary of buying inventory where the visible page is dominated by ads; adding garbage text helps this…
This strategy comes into full glory with “long tail SEO” strategies, in which a business attempts to rank higher by creating micro-targeted pages for a tiny little topic, each optimized to be the most relevant and comprehensive cat in that particular jungle. All of which drive traffic to a single lead gen form.
This explains why we have websites that iterate through epic topics like:
- Roof Repair Scranton
- Lost Shingle Scranton
- Who Gave Me Shingles Scranton
Dwight Schrute would be proud…
Newsflash: Nobody Actually Reads
I’ll go a step further. Your carefully written web page is more or less irrelevant to human audiences. A complete waste of time and money.
To crush the spirit of the few English majors who haven’t rage quit yet, I’ll cite SCIENCE (from our UX friends). We’ve known this for well over 20 years. Here’s a study from the nn group which measured this back in 1997.
- 79% of users merely scan text
- Text needs to be concise
- Bulleted lists are good
This has only gotten worse over the past twenty years, between audiences learning how to adapt to the web and the growth of mobile devices.
This follow up study only confirms it. Human readers are going to cherry pick you content, focusing on tables and items of note. Worse, certain features (block quote boxes) have a habit of disrupting engagement.
So – how can we use this to our advantage?
Introducing The Minimum Viable Article
Snotty comments aside, startups should consider playing the long tail SEO game. Ranking multiple specific pages for content related to your problem space is an efficient way to generate a stream of relevant end user contacts.
So what should this look like?
First, you’ve got to nail the key elements – the stuff humans actually read:
- The Headline
- First paragraph copy – your big point
- Bullet point value proposition
- Chart that illustrates key point
- Call to Action
If I sound like I’m trying to design a 300 x 250 pixel ad, that’s not really far from the truth. You’ve got about the same amount of attention to work with.
The other item I’d invest some effort into is the legitimate FAQ for the offer, addressing genuine questions and barriers to sale. Drop a call to action at the bottom of the FAQ as a last attempt to convert the reader
A resource list at the bottom of the page also tends to get some engagement.
After that? It’s all fluff. Written for robots.
Most of your copy after that point is basically written for search engine purposes. Or to pad out the page and sell additional advertising space.
So what if we… automate generating this additional copy?
What Does Good Fluff Look Like?
Independent of any SEO considerations, writing good fluff is basically just a matter of running out the literary clock. Most college graduates have fond memories of taking a term paper that could be distilled into 500 words and spinning it into an eight page epic.
I’ve been doing to this to you, by the way. There’s a reason I love bulleted lists and bold text. That’s the only content most of you will actually read.
So how can AI content software help us write better fluff?
Originally, it couldn’t. The early article spinner technology was horrific, really only useful for tier 2 content strategies. Duplicate content was an issue, especially for widely used tools. The other option, bulk articles, had many of the same issues. Your entire SEO empire could come crashing down in a bad afternoon. Most of these article spinner tools used some version of the following:
- Random iteration through Mad Libs style “spintax” templates
- Scraping the top N results from Google and pasting them together
- Replacing words in existing text via a spintax process
All of these results were quickly spotted by a human reader, who usually stopped reading at that point. The words didn’t look right. Topics bounced around wildly. It was obvious you were reading auto-generated crap. This stuff got ignored on social media. Google was able to spot it too – most article marketing sites came to a swift end a decade ago.
Amateurs. As any over-caffeinated college senior knows, the key to writing good fluff is two fold:
- The text needs to look like legitimate English
- You need to stay consistent with the topic
- There needs to be just enough substance to keep a reader engaged
The most recent generation of tools has made good progress in solving these issues. By switching from basic techniques (nested spintax, scraping) to a deep understanding algorithm and native-quality US English language generators, we’re getting articles that can pass as written by a human.
I’m using the term “pass” loosely. It’s a bit like watching a newly minted MBA consultant talk about “unleashing synergy”. The intellectual rigor of the content may be open to debate. But it sounds pretty. It passes the first scan. And that’s enough to make this work.
Secrets of A Content Mill Writer
I did a tour as a content mill writer a while back, getting paid peanuts for writing content for various publishers and SEO agencies. We were paid by the word, which taught us how to crank out nice looking blog post filler on autopilot. That taught me some tricks of the trade that can help us here.
Editing is almost always faster than writing. Having a big lovely block of well written text to play with lets you work in the world of ideas and phrasing, rather than pounding out your first draft.
This is especially true if the text you’re looking at is already well written English, since you’re not getting distracted by grammatical errors.
Know what the slow part is?
First drafts. Research. The time consuming part of the work, where you need to come up with a vision of the article and pound out the copy. Especially if the key points are already been decided by my SEO manager. (By the way, this is a good thing – with the right analytics, your SEO team can help writers develop content with a far better chance of ranking)
Seriously, I had a cut & paste article bank which I would recombine and plunder to make epic journalism. A nice batch of curated material. It works just fine. The content marketing clients loved it. It took me three hours to write my first article; by the end of my tour as an article writer, I was cranking out similar pieces (of higher quality) in about an hour.
For new topics, I’ll happily use an article generator to blast out a first draft, even if the copy isn’t perfect. Because I’m generally able to trim at least half of the resulting copy into a solid piece. Every once in a while you get a bad draft, which you can fix by merely rerunning the article generator again.
From my perspective, we’ve set the bar for AI generated content at a silly level. We don’t need the computer to do everything. We merely need it to handle all the boring stuff (research, rough draft), which allows the human writer to focus their expertise on polishing the text and finding visuals that can spice up the message for an audience.
Speaking as a publisher and content writer, being able to crank out content 3 X – 5 X faster has a huge impact on my own earnings.
Show Me the Money!
By the way, all that copy costs money to write. The fully loaded cost of using human writers in a Western language ranges between $75 and $300 per 1000 words published including editing, layout, and project management.
Article writing robots are cheaper. We’ll profile a couple of services later in this article, but the cost can be as low as $1 per article (usually as part of some kind of monthly package plan). Many of them can be directly integrated into a WordPress blog, which eliminates time spent on editing and formatting.
For my fellow misanthropes, this also spares you the effort required to recruit hordes of writers, negotiate rates, coach them, and otherwise engage in various acts of “management”. And the cost associated with that work.
Instead of spending money on filler content, redirect that spend into good editing, design, and graphics. You know, stuff people actually look at…
Better yet, many of those elements can be approached at a platform level. Design can be done as WordPress templates. Graphics can often be placed on similar pages. Once you have the key ideas of an article, it isn’t very hard to spin a few different version of an article’s headlines and calls to action.
Executed properly, this is a way to reduce your cost per page of published content by a factor of 4 to 10. Without losing quality or engagement.
You’ve got a couple of options for implementing this. We’re going to look at couple of the top contenders below….
Note: Several product profiles include affiliate links to free trial versions that you can experiment with. If you keep the product, we may earn a commission.
Competitor: Market Muse
Originally designed as software to help you create authoritative content to rank higher in Google’s search results, Market Muse’s claim to fame is being able to show you what topics Google (probably) wants to see for a query.
MarketMuse looks at the top results, identifies the key themes they write about, and generates a report that guides you through the topics to cover.
They recently adapted this process to compete in the market for computer generated text, with an auto-generated “first draft” service. These are bought via credit system – cost for a brief is around $100 per page. Their full featured plans start at around $500 per month and agency plans if you are going to support multiple writers / websites can easily run $1500 or more.
It’s a good tool, particular on the SEO side, but their machine generated text is expensive for broad use. It’s a good data point on the top of the market.
Competitor: Article Forge
One product that we’ve been watching is an article generation software known as Article Forge (older article forge review here). They’ve made a string of improvements in their product over the past year, addressing a lot of the issues we mentioned above. This includes better English language text generation and cleaning up the research process so the content does a better job of staying on topic.
The newest release takes another step forward, adding features to help SEO efforts and article length. The current articles are getting good enough to use on a money site with some editing. You can target a keyword and – with the most recent version, Article Forge 2.6 – the article will integrate LSI (Latent Semantic Indexing) keywords into the text. This adds authority for search engine indexing and SEO efforts.
The tool is getting better at generating longer articles as well, using a new set of content creation features to ensure longer articles have better structure and flow, are not repetitive, and have better overall topicality. The generated content will have a relevant title and the new article can be automatically posted to your WordPress site.
Cost is pretty affordable. They’ve got an unlimited plan which costs $27 per month with an annual subscription ($57 month to month). For a free 5 day trial of their software to compare their quality with a bulk content writing service, click here.
(With GPT-3 there are several article forge alternative candidates in the pipeline. This will change publishing.)
Going From Theory To Practice
Here’s three ways you can put this into practice with your existing blog site or projects. Some of these are no-brainers (content spinner? We can do better). Others support a more strategic approach to building your audience. Either way AI assisted writing tools can unlock new options for expanding your digital content footprint.
Old School Tier 2 SEO Websites: Setting aside the question of whether this stuff actually works (not going to critique someone else’s racket), bringing AI generated content into your network is a big step up from the tools you used to build these sites to begin with (article spinners, cheap content writing agencies). Your site will look more credible, enjoy a lower bounce rate, and carry a lower risk of Google Penalties. From a cost perspective, AI generated content will fall between spinners and a cheap content agency.
Enhanced Content Curation: Using article forge to supplement a material curation effort or short high quality article. I’m particularly fond of this approach – you stick the business end of the article at the top of the page (where it gets read) and put a good call to action once you’ve said your piece. Drop a lovely image below that. And then, finally, below that you use Article Forge and a little editing to pad this out into a magnificent 2000 word beast of an article (all unique content). A veritable stallion for search engine adoration. Go ahead, pay premium rates for a 300 word blurb at the top and infographic. Polish the call to action until it shines. The rest of the piece costs less than a buck.
Grinding Out Long Tail Pages: There’s a solid business case to create a bunch of micro-targeted pages pitching different aspect of your services. For a startup, the issue comes down to simple pragmatism: you’re the new player in town and trying to outrank established competitors and authorities on your topic.
And Google’s ranking system isn’t made to do that, at least not for the larger search terms where the traffic is. Their fundamental goal is to drive new businesses onto their advertising platform, where you pay to play. You’ve got to niche down in terms of creating content, since a new website won’t have the authority to rank for a big search. This approach helps you do that. Use AI assisted content to generate heftier articles and cover more ground.
Final Thought:
Amazon’s first commercial use of warehouse robots didn’t completely replace human pickers. They focused their efforts on the pieces of the process which could be well addressed by AI and used humans for the balance of the effort. This gave them immediate benefit (for traction), which will likely expand over time (as we build and deploy more robots).
A similar situation will likely occur as AI starts to enter the world of knowledge work. Be wary of setting high bars for performance. Sometimes you can make a lot of progress (in terms of productivity) with some very targeted efforts to address and improve high cost parts of the process.
The future isn’t some AI writing Shakespeare. It’s probably closer to using AI content platforms to quickly generate and customize the ultimate guide to “plumbers in Plano”.