ChatGPT is an artificial intelligence chatbot that can give directions and perform tasks like writing essays. There are a number of issues that should be understood before deciding on content and methods to use for his SEO.
The quality of ChatGPT content is amazing, so the idea of using it for SEO purposes should be addressed.
Explore.
Why ChatGPT Can Do It
Simply put, ChatGPT is a type of machine learning called a large-scale learning model.
A large-scale learning model is an artificial intelligence trained on vast amounts of data that can predict the next word in a sentence.
The more data you train, the more types of tasks you can perform (such as writing articles).
Large language models may develop unexpected features.
Stanford University wrote about how increasing training data enabled GPT-3 to translate text from English to French.
Large language models like GPT-3 (and ChatGPT’s underlying GPT-3.5) were not trained to perform specific tasks.
They are trained with a wide range of knowledge that can be applied to other areas.
This is similar to how humans learn. For example, if a human learns the basics of carpentry, he can apply that knowledge to build a table.
GPT-3 functions similarly to the human brain in that it contains general knowledge that can be applied to multiple tasks.
A Stanford University article on GPT-3 explains:
“Unlike chess engines, which solve specific problems, humans are ‘generally’ intelligent, capable of learning everything from writing poetry to playing soccer to filing tax returns.” can.
In contrast to most current AI systems, GPT-3 is approaching such general intelligence…”
ChatGPT incorporates another large language model called InstructGPT. It is trained to take instructions from humans and long-form answers to complex questions.
This ability to follow instructions allows ChatGPT to receive instructions to write essays on virtually any topic and to do so in a specified manner.
You can write essays within constraints such as word count and inclusion of specific topic points.
6 things you should know about ChatGPT
ChatGPT has been trained on a variety of publicly available texts so you can write essays on virtually any topic.
However, ChatGPT has important limitations that you should be aware of before deciding to use it in your SEO project.
The biggest limitation is that ChatGPT is unreliable in producing accurate information. The reason this is inaccurate is that the model is only predicting words that come after the word before the sentence in a paragraph on a particular topic. Accuracy is irrelevant.
This is a top concern for anyone interested in creating quality content.
1. Programmed to avoid certain types of content
For example, ChatGPT is specifically programmed not to generate text on topics of graphic violence, sexually explicit acts, and harmful content such as instructions on how to make explosive devices.
2. Ignorance of current affairs
Another limitation is that it doesn’t recognize content created after 2021.
So if you need your content to be up-to-date and fresh, ChatGPT in its current form may not work for you.
3. Built-in bias
An important limitation to be aware of is that we are trained to be helpful, honest, and harmless.
These are not just ideals, they are intentional biases built into the machine.
Programming to be harmless seems like the output avoids negativity.
That’s good, but it also subtly alters the article from being ideally neutral.
In other words, you need to get behind the wheel and explicitly tell ChatGPT to drive in the desired direction.
Below is an example of how the output changes due to bias.
I asked ChatGPT to write a story in the style of Raymond Carver and another story in the style of mystery writer Raymond Chandler.
Both stories had bright endings uncharacteristic of either writer.
To get the output in line with my expectations, I had to guide ChatGPT with detailed instructions to avoid bright endings and narrative resolutions with Carver-style endings.
The point is that ChatGPT has biases and you need to be aware of how they affect your output.
4. ChatGPT requires very detailed instructions
ChatGPT needs detailed instructions to output higher quality content that is likely to be more original or have a specific point of view.
The more instructions, the more sophisticated the output.
This is both an advantage and a limitation to be aware of.
The fewer instructions a request for content contains, the more likely the output will share similar output with another request.
As a test, I copied the query and output that several people posted on Facebook.
When I asked ChatGPT the exact same question, the machine produced a completely original essay that followed a similar structure.
Although the articles were different, they shared the same structure and touched on similar subtopics, but the language was 100% different.
ChatGPT is designed to choose completely random words when predicting the next word in an article, so it makes sense that it wouldn’t be plagiarized per se.
But the fact that similar requests generate similar articles highlights the limitations of simply asking “please do this”. ”
5. Can you identify ChatGPT content?
Researchers at Google and other organizations have long worked on algorithms to correctly detect AI-generated content.
There are many research papers on this topic, here is one from March 2022 using output from GPT-2 and GPT-3.
The title of the research paper is Adversarial Robustness of Neural Statistical Features in Detecting Generative Transformers (PDF).
Researchers were testing to see what kind of analysis could detect AI-generated content that uses algorithms designed to evade detection.
They tested strategies such as replacing words with synonyms and adding misspellings using the BERT algorithm.
They found that some of the statistical features of AI-generated text, such as the Gunning-Fog Index and Flesch Index scores, were used by algorithms designed to evade detection. was also useful in predicting whether the text was computer-generated.
6. Invisible Watermark
Even more interesting, OpenAI researchers have developed a cryptographic watermark to help detect content created through OpenAI products such as ChatGPT.
A recent article called attention to a discussion by OpenAI researchers available in a video titled Scott Aaronson Talks AI Safety.
The researcher says that ethical AI practices such as watermarking could evolve into industry standards, just as Robots.txt became the standard for ethical crawling.
he said:
“…for the past 30 years, we have discouraged large Internet companies from agreeing to certain minimum standards for fear of being sued, a desire to be seen as responsible players, or for any other reason. I’ve seen
One simple example is robots.txt. If you don’t want your website to be indexed by search engines, you can specify that and the major search engines will honor it.
Similarly, you can imagine something like watermarking — if you can demonstrate and show that it works, is cheap, doesn’t compromise the quality of the output, doesn’t require a lot of computation, etc., it It will only become an industry standard and anyone who wants to be considered a responsible player will include it. ”
The watermark developed by researchers is based on cryptography. Anyone with the key can test the document to see if it has a watermark that indicates it was generated by AI.
The code can be in the form of punctuation usage, word choice, and so on.
He explained how watermarking works and why it’s important:
“My main project so far has been tools for statistically watermarking the output of text models like GPT.
Basically, whenever GPT produces a long piece of text, the word choice has an otherwise unnoticeable I hope there is a secret signal.
We want to make it harder to take GPT output and pass it along as if it came from a human.
While this can obviously help prevent academic plagiarism, it can also help prevent propaganda mass production, for example. That is, spamming every blog with seemingly on-topic comments supporting Russia’s aggression in Ukraine, and not even a vandal-filled building in Moscow.
Or impersonating someone’s writing style in order to denounce them.
These are all things you want to make difficult, right?
Researchers shared that watermarking beats algorithmic efforts to evade detection.
Should you use AI for SEO purposes?
AI content is detectable
Many say Google has no way of knowing if content was generated using AI.
AI detection is more or less a solved problem already, so I don’t understand why anyone would say such an opinion.
Even content that deploys anti-detection algorithms can be detected (as described in the research paper linked above).
Detecting machine-generated content has been the subject of many years of research, including research on how to detect translated content from another language.
Does auto-generated content violate Google’s guidelines?
In April 2022, Google’s John Mueller said that AI-generated content violates Google’s guidelines.
“For us, they basically fall into the category of auto-generated content that has been around since the very beginning of the Webmaster Guidelines.
And people have generated content automatically in many ways. And for us, if you’re using machine learning tools to generate content, it’s basically shuffling words, looking up synonyms, doing translation tricks that people used to do. Same thing.
My suspicion is that the content quality might be a bit better than really old school tools, but for us it’s still auto-generated content, and it still violates webmaster guidelines for us. means that it is spam. ”
Google recently updated the “auto-generated” content section of its developer page about spam.
Created in October 2022 and updated near the end of November 2022.
This change reflects a clarification about what causes auto-generated content to be spam.
It first said:
“Auto-generated” (or “auto-generated”) content is content that is programmatically generated without creating anything of its own or adding significant value. ”
Google updated that sentence to include the word “spam”.
Spammy auto-generated (or “auto-generated”) content is content that is programmatically generated and does not create anything unique or add significant value. ”
This change seems to clarify that automatically generated content alone does not constitute spam. What makes that content problematic is the lack of all added value and general “spam” characteristics.
ChatGPT may contain watermarks at some point
Finally, OpenAI researchers (a few weeks before ChatGPT’s release) said watermarks would “hopefully” come in the next version of GPT.
So ChatGPT may be upgraded with a watermark at some point, if not already.
Best use of AI for SEO
The best use of AI tools is to scale SEO in ways that make your employees more productive. It usually consists of letting AI do the tedious work of research and analysis.
Summarizing a webpage to create a meta description may be acceptable usage as Google clearly states that it does not violate any guidelines.
Using ChatGPT to generate outlines or content briefs can be an interesting application.
Relying on AI to create content and publishing it as-is may not be the most effective use of AI unless the AI is first reviewed for quality, accuracy, and usefulness.
Featured image by Shutterstock/Roman Samborskyi
