What's your take on AI generated content?

[dennis_max]

I've been an early adopter of Copy.ai, Jarvis and a myriad of tools that generated good short-form content. Copy.ai saved me a TON of time at one of my jobs - absolutely great tool for generating thousands of FB and Google Ads.

However, all of these tool were lacking the long-form content. I've recently worked with a few developers who toyed with the idea of writing 1.5-2k articles with the AI, and the results seems promising so far. I want to build a test website to play with this idea, however, I'm concerned about Google's algorithms.

I remember reading somewhere that Google recently announced that it can detect and penalize GPT-3 type content quite easily, however, I'm having a hard time understanding how it can realistically detect that.

Overall, what's your take on GPT-3 and have you considered using it for long-form content on your projects?

 

[freshpeppermint]

Works good as a writing assistant, you can't rely on it to write most of it for you though. Also yes there's sites getting millions of traffic but at the end, the one which wins is quality.

Here's what's gonna happen: You produce fewer articles, you get slower traffic but it's quality article. So it will rank higher up, you will build more backlinks your domain will get more juiced up. So in the future when you put out more articles you can outrank other people. You are less at risk of Google's algorithms if you have the BEST content for that keyword assuming it's low-medium comp.

So it's like imagine making 100 then 1000 then 10000 then 100,000. VS making 1000 sooner and then penality. Time wasted. Although then you can argue that if you did make money sooner and pumped it into a quality project maybe it would do better. But I still think the quality approach is better.

 

[hotdog11]

The worst thing about writing is writer's block. With AI, you never get writer's block. Content just flows, but you have to edit. And fact check.

Like freshpeppermint says, it's a great writing assistant. But if you use it to spit out auto-generated text, I'd say the output is really just some readable garbage. Of course it's much better than those spin-text articles we used to generate in the past, but still low quality stuff.

If you're editing whatever the AI writes, and you're improving the quality of its writing, it's essentially human-written, isn't it? Google can't tell it's GPT-3 assisted.

 

[zelch]

Google can't tell it's GPT-3 assisted.

this is my concern, we don't really know this.
and we certainly don't know if they will eventually be able to detect it (and algorithmically penalize it)

 

[Boy]

I like it.

Not great for pumping out long-form content without a solid outline (depending on the content/context), but is fantastic for doing things like expanding content, turning features into benefits, whipping up summaries and product descriptions, and creating shorter-form content like emails, social posts, adverts & CTAs, etc.

 

[Philip J. Fry]

Been out of this solo-entrepreneur loop for a few years. GPT3 sounds awesome and AI-Assisted writing tools like Frase seems great. Hopefully this will cut down on writing costs, writing time, and increase quality. For some queries, such as "Where is the Mona Lisa in the Louvre?" The information that satisfies that query is only one sentence "Second story, in the Italy section, room 202" for example. The rest is fluff. GPT3 should help make better fluff, easier

 

[4potato]

I think you'll spend more time editing and trying to force good content out of it then you think. I tried Jasper and one of the other ones this year and it was sort of a letdown for me. Especially if you want to write anything remotely technical, it spews out nonsense with no fact checking.

Good for intro paragraphs, ad headlines, maybe extremely basic how to content. I tried some of the newer article builders where it first brainstorms some outlines, and you select an AI created outline, then it fills in the body content section by section..you have to supervise paragraph by paragraph and delete a large amount of bad output. I just found it wasn't worth it for me. Still a month of most subscriptions costs less than getting a few articles written for you. Not a huge expense to buy a month and put 5-10 hours of your own time in, but I think they're more hype then quality. A recent Authority Hacker podcast discussed this topic as well.

I think the staggering future breakthrough which will break the Internet is multi-document summarization. I follow this computer science subject loosely, its deep level PhD stuff that's over my head but the goal is spinning accurate content by taking dozens / hundreds / thousands of pieces of content, as input, reading and understanding them algorithmically and producing one pristine document as output. That tech is still being worked on by profs and brilliant students, it's nowhere near "product level" but keep an eye for that over 5 years.

At a basic level you take each document, apply natural language machine learning rules like tokenizing sentences, using CS similarity algorithms on tokenized sentences, throw out duplicates, group similiar sentences into paragraphs and rebuild one document as output from X amount of input documents. Not that this matters for right now. Buy the hype in the current tools if you wish, try for a month, see if you remain a loyal subscriber after a month. If the hype is accurate, good content at scale, the cost is trivial (if you could get 10, 20, 50 good articles a month out of tools like Jasper) that would cost you nothing on a per article basis. I think we all know when something is too good to be true.

 

[bernard]

It definitely has its uses, but I agree with most people who say it won't rank on its own.

What is it good at though? Summarizing widely available content into guest post content for one. That's what I use it for.

I've tested it to write "buyers guide" in roundup reviews, but too early to tell if it will rank.

 

[Boy]

The more I think about it, GPT-3 would be absolutely GREAT for

• BuzzFeed-style, social sharing sites with clickbait topics
• Daily horoscope sites
• Conspiracy sites

Anything that requires NO facts and isn't waiting for Google to rank.

I'm sure it would be fantastic for elementary to middle school fiction books as well. Some real Captain Underpants-type trash.

 

[Potatoe]

Some real Captain Underpants-type trash.

*cracks knuckles* Alright, let's step into the thunderdome... Say what you want about anything else, but when you bring into question the literary genius of Dav Pilkey, it's on sight.

Nah nah, but, interesting topic. It might be neat to poke around some of the SEO/expired domain selling folks and seeing if any of their niches match up with something that would also be really easy to generate content for, like some of the stuff you mentioned above.

With niche selection I'm not just thinking about how profitable it might be, or what the competition is like, or what domain I can find to build on, or all the other points, but also how easy it would be to write for and that last one is quickly soaring up the list of priorities.

I'd be willing to trade a fair chunk of earning potential, or difficulty, or whatever for niche that is like, 10x easier to create content for than something else. That ease of content gives so much breathing room and so many more lines to cast into the water or so many more snowballs rolling down a hill, or what have you.

I would be interested in finding a niche that's super easy to create content for, and is at least okay in all the other factors one might look at, with a solid domain available to build upon. I don't know much about this new fangled programmatic AI mumbo jumbo, but I think you're right on the money about finding niches that suit it best.

 

[bernard]

The thing is, we shouldn't be focused on what it does now, but what it will do soon.

And it will be better and probably sooner than we'd like.

I mean, things are progressing very quickly. This other day I was reading a mainstream article about Deepfake porn, deepnudes, if you will. For research sake I found Kate Beckinsale performing some acts and it was very convincing to the subconscious processes.

Of course the focus on the article here was on deepnudes made by scraping Instagram for your crush and hows that not cool to star in porn movies against your wish.

Now consider that with VR and Meta and all that, then pretty soon you can "have sex" with every person in the world. And it WILL be convincing.

How's that for dystopia?

Anyway, that's to say, we need to think ahead.

And what is it that AI does well, it uses data to approximate a human. So we feed it data.

I believe it was @Boy who had a thing with auto-creating comparison pages. I also frequently have data tables of unique research. I would like to see what an AI could make of that. Feed it a large dataset and give me pages for each of the rows.

 

[Magician]

The only way I know people using it is to generate content for Guest Posts.
I want to start experimenting with using AI. I heard people prefer to use the openAI playground over tools like jarvis. But I still have to figure out why.

 

[Ljc2424]

It's a plague on the SERPs at the moment as far as I'm concerned, unfortunately.

 

[hotdog11]

this is my concern, we don't really know this.
and we certainly don't know if they will eventually be able to detect it (and algorithmically penalize it)

You don't need to know. Just pass the organized article through an edit and it'll become an original. Not surprising, there'll be lots to edit and the final copy can be very different. And unique in both content and voice.

In the end, GPT-3 can only ASSIST in writing. It can spit out words that make sense, but it can't fact-check nor produce work with consistent voice. Therefore it's not a writer, only a writing tool.

What Google is against are content that's totally written by GPT-3 without any edit whatsoever. Because this doesn't serve their needs.

 

[dennis_max]

Been out of this solo-entrepreneur loop for a few years. GPT3 sounds awesome and AI-Assisted writing tools like Frase seems great. Hopefully this will cut down on writing costs, writing time, and increase quality. For some queries, such as "Where is the Mona Lisa in the Louvre?" The information that satisfies that query is only one sentence "Second story, in the Italy section, room 202" for example. The rest is fluff. GPT3 should help make better fluff, easier

Philip, actually, I really want to find a way to contact you and discuss some ideas for travel related sites. I've read your story about your former employer - let's find a way for a revenge? =) I have a very similar situation, and put a ton of cash aside waiting for this to happen - would love to talk if you're interested.

As for the website, here is a sample that I've just produced. I'm using a very different tool from Jarvis, copy.ai - and all the SaaS for short-form content. Looks VERY good to my eyes, what to you guys think?

10 Fun Things to Do in Denver This Weekend
​

If you're looking for something fun to do in Denver this weekend, look no further!

Here are 10 great ideas, from exploring the city's many museums to hiking in the Rocky Mountains. Whether you're a local or just visiting, you're sure to find something to enjoy.

1. Visit one of Denver's many museums: The Denver Art Museum, the History Colorado Museum, or the National Western Stock Show and Rodeo Museum are all great options.

2. Take a hike in the Rocky Mountains: There are many trails to choose from in the Rockies, from easy hikes near downtown to more challenging trails in the wilderness.

3. Go to a Museum: Denver has a wide variety of museums, from art to history to natural history, there is something for everyone in the family.

4. Explore the Denver skyline: Take a walk or bike ride up Broadway or take a ride on the free light rail system to see the many architectural gems in the city.

5. Play some golf: There are many great courses in the Denver area, from public courses to championship venues.

6. Visit a local brewery: There are many great breweries in the Denver area, from small boutique breweries to national brands.

7. Visit a botanical garden: There are many great botanical gardens in the Denver area, from large garden complexes to small gardens in residential neighborhoods.

8. Take a food tour: There are many great food tours in the Denver area, from farm-to-table tours to city tours.

9. Visit the Denver Art Museum. There are many different types of art at the Denver Art Museum, from western art to aboriginal art.

10. Visit the Museum of Contemporary Art. There are many different types of modern art at the Museum of Contemporary Art.

Bonus tip: Visit the Denver Museum of Nature and Science. There are many different types of exhibits at the Denver Museum of Nature and Science, from dinosaurs to space."

 

[bernard]

Jasper says:

I believe that AI-generated content has a lot of potential. However, I think it is important to be careful about how it is used and to make sure that the content is clearly labeled as being generated by AI. If not, there is a risk of people being misled or misinformed.

That said, I think AI-generated content can be a valuable tool, particularly in terms of helping to generate ideas or providing information that people might not otherwise have access to. For example, it could be used to generate summaries of long articles or to provide translations of text into other languages. Overall, I think AI-generated content has the potential to be useful, but it is important to use it in a thoughtful way.

Wow, Jasper is so good at getting your postcount up with inane drivel, so you can spam your links.

 

[Ljc2424]

In one word: Crap!
It's plaguing the SERPs and making it easy for black hat marketers to make a quick buck. Buy an expired domain, use AI content to flood the site with content, quickly rank, make some money then (maybe) eventually get found out by Google. Rinse and repeat.

I think it depends on where you set your moral compass ultimately and whether you care about what it is you're putting out on the internet.

 

[sciencestudent]

Is anyone watching known AI generated sites for this May 2022 update to see if they are going to get removed? I haven't done a good job keeping up with what the hot AI users are.

Looks like askinglot is finally getting some downward pressure

 

[DeadJoe]

A couple people shared their sites that got hit, and while they all say they have great content, they're using 80% AI generated stuff. It's really easy to catch if not edited properly

 

[Sassy]

I bought the Appsumo lifetime deals for 3 tools this past year: Niche$$$, WordHero, and Rytr. I use them mostly for social media posts, Pinterest descriptions, and post titles, outlines and intros.

It's great for writer's block like @hotdog11 mentioned and short-form content, but fact-checking and editing are essential, especially for longer articles.

I also used Jasper and Copy.ai for 2 months and must say that I liked Copy.ai a lot better than Jasper both in quality and usability.

All said, I will not fire my writer just yet, but for short posts, social media, descriptions, etc it's great to just pop in a sentence and quickly get some content.

 

[DannyDonchev]

Hey guys,

Here is my take on AI content. If you hire a VA/writer with good English to improve the copy you are good. Testing this on 2 websites for a few months and we are going up despite the recent Google update.

IMHO Jasper becomes too costly for what it is and there are other tools that with LTD can almost replace it. Right now we like bramework.com and ontentbot.ai most.

We also test Wordai and Articleforge. Whatever tools you use just make sure they do not add junk code to the text. For example, Wordai does this and this is a footprint.

For plagirisum checker use Quetext.

Junk AI content sinks fast. Examples - midogguide.com and serp.com

 

[Ashk881]

I remember reading somewhere that Google recently announced that it can detect and penalize GPT-3 type content quite easily, however, I'm having a hard time understanding how it can realistically detect that.

Classic supervised learning, that's how.

Basically, this is what they do: they get a GPT-3 or a model like it (Google has an even bigger one btw, they are calling it LaMDA. It's about 3 times as bigger as GPT-3's biggest model). They then make the model spit gargantuan amounts of text.

They then scrape gargantuan amounts of text from the internet that they can confidently say was written by a human.

A neural network is then deployed that "learns"on these datasets. It's told to find patterns and differentiate which one is written by a human and which one by ai.

With enough learning (that is, with large enough datasets) it can perform better than humans. It basically finds hidden patterns that we humans can't notice.

 

[UFO]

I remember reading somewhere that Google recently announced that it can detect and penalize GPT-3 type content quite easily, however, I'm having a hard time understanding how it can realistically detect that.

If this Open-AI detector can do it, you can damn well bet that Google can too. I have several AI content tools and anytime I throw in some unedited text it gets flagged as fake with almost 100% accuracy.

 

[Sassy]

If this Open-AI detector can do it, you can damn well bet that Google can too. I have several AI content tools and anytime I throw in some unedited text it gets flagged as fake with almost 100% accuracy.

I think the magic word here is unedited text. I just tried some of my ai generated content that I slightly edited and cleaned up with Grammarly and it gets flagged as 90%+ real.

 

[UFO]

I think the magic word here is unedited text. I just tried some of my ai generated content that I slightly edited and cleaned up with Grammarly and it gets flagged as 90%+ real.

Oh definitely. Huge difference between letting the AI do all the heavy lifting and using it as an aide to fill in gaps, find other angles, and flesh out ideas. I see the same as you with edits, or even using the Wordtune plugin to rewrite sections and sentences, 90%+ flagged as real.

Part of the problem is that some of the AI companies market in a way that can be misleading to n00bs. Claims like "wRiTe A 2,000 wOrD bLoG pOsT iN 15 mInUtEs" so people are blasting out content with little to no editing or regard to quality. Tons of automated trash out there too.

Will there be a day of reckoning? Who knows, but with Google's track record of targeting content quality, I wouldn't want to be caught with my pants down. If/when that happens we'll see a lot of pure AI loyalists screaming "muh site go down, wut wrong?!" That being said, I am testing high quantity AI content with very few edits on a local business lead gen site, if it gets nuked then no biggie. I wouldn't do this on any site of importance.

As a techie, I find the technology fascinating and see a lot of valid use cases for it already like social media posts, emails, ebooks, etc. Even though I am running tests on the mass generation side of the coin, I still firmly believe you shouldn't let a tool be a major replacement for your own brain.

Just my 2 cents.