59917
•
20-minute read
•
These days, AI tools have become routine in SEO. Google and Bing are competing on which AI is more powerful. Besides, independent AI writing tools appear and improve their technology.
At the same time, the SEO community is still trying to figure out how Google actually treats AI content. Is it a go or a no? Will Google penalize or derank pages if it detects machine-written texts?
In this article, I’ll dig deeper into the world of AI content writing tools to reveal how good they are. And, sure thing, try to understand how Google actually treats machine-generated content.
Here are the AI writing tools I’ve tested during my research:
It may seem that Google does not have any trouble with AI-generated content these days. What’s more, with Search Generative Experience under development, Google’s AI restrictions would seem really weird.
If we compare Google's AI content policies of, say, 2021 with what we have today, we’ll really see that they have become more loyal.
Here’s what Google said in 2021:
And here’s what the abstract looks like today:
Now, more attention is given to spammy content, no matter if it is human-written or AI-generated. But Google still emphasizes that machine content needs human revision; otherwise, it may be considered spammy. From my own experience, this assumption is pretty true for copies created totally with the help of artificial intelligence — they are often watery.
All in all, Google’s policies towards AI content do not differ much from those for any content at all — content should bring value to users, no matter who’s the creator.
As Google does not mind AI content, then you can use AI writers to make your content creation process faster and easier. Or… Or not? Machine-written texts still need a human review, which can take even more time than if a human author writes the copy from scratch.
To finally end the debates about AI writers’ capability, I've collected 9 popular tools and tested them to see what they can offer and how much effort it takes to produce a piece of content.
Yes, I know that tools like ChatGPT can write very good texts, but achieving this requires deep AI tools training. So, to make the competition honest, I decided to “feed” AI tools with the same input (or almost the same, depending on the workflow) to generate a blog post.
The topic I’ve asked them to cover is pretty evergreen — how to cope with burnout at work. Then, as I was testing tools to write content for SEO, I asked them to use the following keywords:
how to cope with burnout at work, cope with burnout at work, mental health, work-life balance, coping mechanism, avoid stress, deal with stress
The text length I needed was about 1,000 words, and, of course, the texts were meant to be well-structured and look natural, as if no AI assistants were used.
To evaluate what I’ve actually got with my research, I checked these articles for grammar mistakes, difficulty of readability, and keyword optimization. Additionally, I read the copy to see if it actually makes sense to a human.
I tried to interfere with the text generation process as little as possible. Just because my goal was not to create a super-cool article with the help of AI but to test what AI can do without any human assistance.
So, here’s what I’ve got.
ChatGPT needs no further description these days — everyone knows what it is. You may adore it or hate it, but you cannot ignore it — ChatGPT is here, it is widely used, and, if used wisely (not only widely), it can become a great assistant in many SEO-related tasks.
But what about writing? Here’s what I’ve got.
At first, I asked ChatGPT-4 to generate an outline for my post:
And the outline was really nice. Then, I asked ChatGPT to write a 1,500-word article based on the outline and gave him more instructions to follow:
If you think that I’ve instantly got my copy, you’re wrong — ChatGPT had a big problem with understanding the length I needed. At first, it only gave me 600 words, and it took me five or six tries to get a copy of 1,000 words. I decided to stop here without torturing the poor tool further.
To evaluate this 1,000-word masterpiece, I used Hemingway Editor and WebSite Auditor’s Content Editor module.
Hemingway Editor, which is aimed to improve readability, was kind of shocked by the amount of super-hard sentences, and scored the copy with 13-grade difficulty. Which is bad, as the maximum difficulty for an average human to understand easily is 9. In the perfect scenario, it should be 6.
As for search optimization check with WebSite Auditor, the score was 77%. Not the worst thing possible. Plus, WebSite Auditor evaluates the SEO score of a page as a whole, including alt texts for images, headings optimization, and meta elements. So, probably the final score would be even higher.
BARD is a Google-created rival of ChatGPT. It is also being actively developed, aiming to “slay” Chat with every new update. Actually, Google’s BARD works in a pretty similar way — you type your query and then communicate with the tool in the form of a casual conversation.
Recently, Google announced that BARD now operates on Gemini AI, a new AI model many times more powerful than ChatGPT. Gemini AI boasts increased multimodality, so it is meant to better understand the implications behind the queries. This, in turn, means that BARD's answers will become more human-like. Is this true? Let’s check.
The input was the same as I gave to ChatGPT. The only difference is that I was too lazy to generate a new outline and used the one by ChatGPT:
BARD turned out to be much more understanding in terms of length and gave me 1,300 words with a single reminder from my side. Probably, it needed a “punch” just because it ran out of space.
I thought that BARD would come up with an easy text — it boasted Gemini AI’s neural capacities too much — but the result was quite a surprise. Not the good one. A difficulty score of 14 is even worse than ChatGPT did.
WebSite Auditor also evaluated BARD’s copy lower and gave it only 73.5% optimization rate. Probably, because fewer of the recommended keywords got into the text.
Unlike general-purpose AI tools that focus on broad tasks, SE Ranking's AI Writer is built specifically for content creation. The tool is powered by the cutting-edge GPT-4O technology boosted by SE Ranking’s own algorithms. AI Writer promises to simplify the process of generating content ideas, building in-depth outlines, and creating drafts of entire articles optimized for search engines. Sounds impressive. Now, let's put it to the test.
AI Writer lets you generate a whole article in one click or take it step-by-step with your input along the way. I’ve decided to use the second option, as I want to have more control over the final result. Speaking of control, the tool allows you to specify user intent, target audience, tone of voice, target keywords, as well as use AI-generated text structure or specify your own outline settings.
The tool produced text closest to my desired length, generating a total of 1489 words.
Since AI Writer is built into the SE Ranking’s Content Editor tool, you also get SERP-based suggestions on your content's length, structure, target keywords, image use, and more. Plus, Content Editor helps you polish your writing with grammar, punctuation, and plagiarism checks.
SE Ranking has a built-in readability score, which assessed generated content in 68 points that correspond to “plain language.” However, for consistency with our analysis of other tools, let's also evaluate AI Writer's content using the Hemingway app.
Hemingway rated its difficulty as a 9, which is much better compared to the previous options. Yet, to make the text more understandable for a wider audience, you’ll need to simplify the vocabulary, shorten some sentences, and possibly adjust the structure flow.
While SE Ranking's Content Score gives the content an impressive 80 (well above the average 65 for similar pieces), let's again maintain consistency and assess optimization using WebSite Auditor.
WebSite Auditor delivers an even stronger result, giving the copy an 89.2% optimization rate. This is, by the way, the highest optimization score out of all content pieces I’ve tested for this review.
Frase is another AI-powered writing tool that promises to ease the SEO content creation workflow. The peculiarity of Frase is that it uses its own AI called Frase NLG (Natural Language Generation), not GPT-3. The website descriptions promise that you’ll forget about low-ranking content and optimization routine, as the texts generated with Frase are SEO-friendly by default. Let’s see if it’s true.
To create an article with Frase, you need to specify the language, enter the title, and choose the creativity level. That’s it.
The tool will then generate a copy of approximately 500 words long — you cannot regulate the length.
As for SEO routine, Frase has a dedicated SEO Tools module for that, which is, alas, not capable of writing long texts. However, it is still good for optimizing existing texts and even analyzing SERP competitors.
I’ve managed to make Frase generate two pieces of text, each 500 words long. Actually, it is quite obvious that the texts are two different pieces of content on the same topic, not one complete article. Nevertheless, let’s see how good Frase’s texts are in terms of difficulty and optimization.
The Hemingway app assigned it a difficulty score of 10, which is a pretty nice result compared with the tools I tried before. Still, this result is anyway far from the 6 points that ensure your texts’ availability for the widest audience.
WebSite Auditor’s score for search optimization is 78.1% — yet the best score of the tested tools, so there might be something to Frase’s focus on SEO:
Rytr is a pretty popular AI writing assistant aimed to help you create anything from YouTube video descriptions to social media bios. The tool has a blog writing feature, which I have tested for my experiment. Rytr operates on GPT-3, so it is actually a ChatGPT of early 2023, packed with some specific features.
Creating a text in Rytr is easy (it is actually easy across all the AI writing tools). You specify options like language, tone of voice, and creativity. Besides, you give Rytr the title and feed the tool with a set of keywords you want to be included in your copy.
The peculiarity of Rytr’s Blog Section Writing is that it can only write one section but not the complete article. So you’d need to come up with sections on your own and change the input every time you start a new section of your post. Otherwise, the tool will just generate endless variants of the same section.
Rytr produced a shorter (approx. 850 words) yet pretty coherent copy that was really easy to perceive. The sentences were actually not as long as they were in, say, BARD’s masterpiece. I’d be still lying if I said that the copy did not need any human supervision at all, but Rytr’s one needed the least of it. Which was actually a surprise.
As for the difficulty estimation, the Hemingway app gave the texts a score of 9 and said “Good”. This was obvious even at the stage of reading.
As for the WebSite Auditor’s search optimization score, it is also solid — the tool gave the copy a 77.7% optimization rate.
Perplexity is not an AI writing tool but “an alternative to traditional search engines, where you can directly pose your questions and receive concise, accurate answers backed up by a curated set of sources”.
By default, the tool operates on GPT-3.5 but can be upgraded to a Pro version, where users can choose the AI model between GPT-4, Anthropic’s Claude 2.1, Google’s Gemini Pro, or Perplexity’s Experimental 70b.
In reality, Perplexity works very similarly to ChatGPT and BARD — you ask it whatever you want and get the answer. Everything seems simple, so let’s try this beast for content writing.
Just as I did with the previous AI writers, I asked for a 1,500-word article with the title I needed. Sure thing, I also entered the outline (that by ChatGPT) and specified the keywords to add to the text.
Perplexity had some problems generating the required 1,500 words just like its big brother ChatGPT. So, after several tries to regenerate the article, I decided to stop when I got a text of slightly less than 1,200 words.
As for Hemingway’s difficulty score, it turned out to be impressive (in a bad way) — the copy got 14 and the verdict “Poor”. I should say I totally agree, as the text is really very hard to understand. Obviously, this is not what readers want to see.
WebSite Auditor was not really happy about the copy either. The optimization score the text got was 73.6% — not a tragedy, but a bit much on the keyword stuffing.
Writesonic is one more GPT-powered writing assistant that can create and rewrite copy for various purposes, from Instagram posts to SEO-optimized copies.
The peculiarity of this tool is that it has several writing modes that can produce long-form texts but need different inputs.
For example, you can use the most advanced version (GPT-4) to create an article with the help of references — you provide a link to a post you want to use as a tone of voice sample. Or you can use the standard AI writer (GPT-3.5) to generate an outline and a post with the help of keywords and the topic of your choice.
Besides, there’s a simplified version that is similar to how ChatGPT works — you choose a prompt and topic to get a text.
I decided to proceed with the standard version, which needs no samples but can generate outlines. So here’s what I’ve got.
I’ve managed to produce an article of approximately 900 words. What’s more, Writesonic is the only tool that provided the copy with an image (the image did not suit but nevertheless). Upon reading, the copy seems pretty well-structured, but in general looks like an average AI-written post, with buggy sentences and a bit hard for seamless understanding.
Now, let’s check the difficulty and optimization scores. The Hemingway app shows a score of 11 — poor readability but anyway, better than many of the tools I tested here.
The WebSite Auditor shows a score of 78% — actually, one of the top results in terms of this research.
Anyword is an AI writing platform that promises to cover any content marketing and SEO needs. The tool uses their custom AI model that is “the only one trained on billions of real marketing data points”. Well, the description is pretty promising.
The writing process of Anyword is similar to what I’ve seen when working with Writesonic. You have a Blog Wizard with a predefined set of steps where you provide a short intro and keywords, and then select the tone of voice. Then, the tool generates a couple of titles to choose from, creates an outline, and, finally, writes the article. The article is promised to be 1,500 words long.
In addition, the tool has a Research feature that is very similar to Google’s People Also Ask section. Here, you can see a set of extra ideas on your topic and ask the tool to dwell on the matter. As a result, you get a bunch of links to the relevant resources and can even add those links to your text.
Besides, Anyword has a built-in assessment feature that evaluates the readability and plagiarism rate of the generated text. Looks like the tool is a really nice choice for SEO content.
The article I got from Anyword was 1,265 words long. Not the promised 1,500, but still a nice try. Besides, I also got a bunch of useful links, which were really up-to-date.
But what about readability? The in-built readability checker of Anyword said the readability is 40, which can be interpreted as “Difficult to read. Likely to be understood by a reader who has at least a college education”. Hemingway's difficulty score actually agrees on this matter — 12 is a sign of a pretty difficult text.
WebSite Auditor’s keyword optimization score is 77.4%, which is relatively good in comparison with other tested tools. Among the leaders, I should say.
Copy.ai is an AI platform designed to automate marketing and sales tasks as much as possible. And, of course, it is capable of writing long-form pieces of content, such as blog posts, articles, and press releases.
The platform is said to operate on several AI models, including GPT-4, Anthropic, and Azure. Besides, Copy.ai boasts a great set of predefined prompts that can help you create SEO content, emails, ads, social media posts, blog outlines, etc. within seconds.
Putting the prompts aside, Copy.ai works pretty similar to ChatGPT — you give it instructions and get your text. And that was just what I did. I took the prompt I used with ChatGPT and BARD and gave it to Copy.ai.
Looks like it got me right and gave a well-structured post, but let’s see how it actually performed.
Spoiler: As usual, I asked for 1,500 words. But the tool managed to come up only with 980. Probably, this is a common disease for all AI content writers.
Without any doubt, Copy.ai gave the worst result compared to all the tested tools.
Hemingway's difficulty score is 15 — it is the poorest result I’ve seen when working with AI content during this experiment. The text itself is poorly written — many sentences are repetitive and wordy, which sets the text far from easy perception. And, to be honest, from any perception at all.
Nevertheless, the copy's structure is actually pretty nice. This is probably what the tool should be used for.
WebSite Auditor’s content optimization score is 62.1%, which is also not a good score. Too much keyword stuffing is to blame.
If you worry that AI writers can replace humans, then stop — AI content is fairly nonsense without human revision. To make these texts look natural, especially when it comes to text to speech generation, you need to train the tool with the help of some iconic pieces of content. Otherwise, you will not get the desired tone of voice and readability.
Still, you cannot ignore AI writing tools in your SEO content strategy. They can help you streamline your SEO routine and speed up many content creation processes.
Here are the most common cases for using AI tools in SEO and marketing:
Enriching existing content with SEO keywords is a time-consuming task, but AI writers can make it much faster and easier. Sure thing, you’ll need to review the result anyway and make edits where necessary, but AI assistants take on the routine.
What you need is to clearly form the prompt, i.e., mention what keywords to add and how many times, and to what parts of your text. In this case, you can get a perfectly-optimized copy within seconds.
While AI content writers are unable to write better than professional human writers, they are really very good at compiling outlines for blog posts.
It often happens that writers get stuck on ideas for sections and on how to form the structure of their future posts. This is where AI assistants come in handy. Like, you don’t even need to train them long to get a usable result — the only thing you need is a topic.
Many AI writers even have special features like Instagram post creation or Facebook description generator. The thing is that they really work and let you create catchy short captions without any trouble at all.
The same is true for titles, outlines, headings, meta descriptions, etc. As long as you provide enough details and give clear instructions, you can be sure that AI writers can offer you a great bunch of usable ideas.
If you are knowledgeable about the topic, you can use AI writers to create a generic copy. Then, you can fill out the missing details and edit the copy for readability. It ends up being a joint effort, where the AI is the writer and you are the editor.
You are not going to create outstanding content this way because you are skipping the research part and relying on the tool and what you already know. Still, it’s a good compromise for when you need to create the copy quickly. It could also be a useful solution for copywriters creating copy for others.
From what I’ve been able to produce with the tools above, I don’t see AI replacing human writers any time soon. All these tools are writers and writing assistants, but not creators. They use the database they’ve been trained on but do not produce anything new.
Google’s policies are quite right when they require human revision for AI texts — otherwise, the internet would drown in tons of keyword-stuffed and unreadable texts.
Users’ feedback with AI tools and my experiment tells the following:
So, use AI writers as you want, but don’t overdo and always check what they write.
By the way, how’s your AI routine going? I know that many of you turn to ChatGPT and BARD and do it pretty often. Are you happy with the results? How much do they help you? Share your experience in our Facebook community.