Before using ChatGPT for SEO and content, know these six things.
ChatGPT is a chatbot with artificial intelligence that can take instructions and do tasks like essay writing. Before choosing how to use it for SEO and content, there are many things to understand.
The notion of leveraging ChatGPT SEO and content should be explored because of the incredible quality of the content.
Let’s explore.
Why ChatGPT Can Do What It Does
In a word, ChatGPT is a large learning model, a subset of machine learning.
A large learning model is a type of artificial intelligence that has been trained on enormous quantities of data and can anticipate the word that will come after one in a phrase.
The number of tasks it can do depends on how much data it has been trained on (like writing articles).
Large language models may acquire surprising skills.
Stanford University describes how GPT-3 was able to translate text from one language to another despite not having been specifically trained to do so because of an increase in training data.
Large language models like GPT-3 (and GPT-3.5, which is the foundation of ChatGPT) are not trained to do particular tasks.
They get extensive training in a variety of subjects that they may subsequently apply to other fields.
This is comparable to how people learn. For instance, even if a person was never specifically taught how to make a table, they may use the knowledge they get from learning the foundations of carpentry to make one.
GPT-3 functions like the human brain in that it has a general knowledge base that can be used for a variety of tasks.
Humans are “usually” clever and can learn anything, from writing poetry to playing soccer to filling out tax returns. in contrast to chess engines, which solve a specific issue.
GPT-3 is getting closer to such general intelligence compared to most contemporary AI systems.
The Stanford University article on GPT-3
InstructGPT, a large language model that was trained to take instructions from people and provide in-depth responses to challenging inquiries, is also called ChatGPT.
Because of its way of following instructions, ChatGPT may be given instructions to write an essay on almost any topic and can carry them out virtually as they are described.
It can write an essay within the limitations of the word limit and the inclusion of specific topic elements.
Six Things to Know About ChatGPT
Because ChatGPT was trained on such a wide range of publicly accessible content, it is capable of writing essays on almost any subject.
But before choosing to use ChatGPT for an SEO project, it's crucial to be aware of its limitations.
The major drawback is that ChatGPT is unreliable at generating correct data. Because the algorithm only predicts what words should follow after the preceding word in a phrase in a paragraph on a certain topic, it is erroneous. It doesn't care about being accurate.
Anyone interested in producing high-quality material should pay special attention to that.
1. Programmed to Avoid Certain Kinds of Content
For example, ChatGPT is specifically programmed not to generate text on the topics of graphic violence, explicit sex, and harmful content like instructions for making explosive devices.
2. Unaware of Current Events
The fact that it is unaware of any content produced beyond 2021 is another restriction.
So ChatGPT in its present state may not be beneficial if your content needs to be current and new.
3. Has Built-in Biases
The fact that we are trained to be kind, trustworthy, and harmless is a crucial constraint to be aware of.
These aren't merely ideas; the machine is intentionally biased in that direction.
It seems that programming to be harmless prevents negativity from coming through in the output.
That's great, but it also subtly changes the article from what may otherwise be impartial.
In a sense, one needs to take the wheel and tell ChatGPT directly to go in the desired direction.
Here is an illustration of how the bias changes the output.
I requested that ChatGPT write two stories, one in the Raymond Carver style and the other in the Raymond Chandler mystery writer style.
The upbeat endings to both stories were out of character for both authors.
I had to give ChatGPT specific instructions to avoid clear upbeat endings and avoid a clear closure to the story to achieve an output that matched my expectations since that is how Raymond Carver's stories often ended.
The point is that biases exist in ChatGPT, and one must be mindful of how they may affect the results.
4. ChatGPT Requires Highly Detailed Instructions
ChatGPT requires specific instructions to output higher-quality content that is more likely to be highly original or make a particular point.
The output will be more complex with more instructions provided.
This has both a benefit and a drawback that should be considered.
The likelihood that the output will be similar to that of another request increases with the number of instructions in the content request.
I duplicated the query and the output that other Facebook users had posted as a test.
The machine created an entirely original article that had a similar structure when I asked ChatGPT the same query.
Although the articles varied, they had a similar structure and covered related subtopics, yet they were written entirely in distinct words.
It makes sense that ChatGPT doesn't plagiarise itself since it is meant to choose entirely random words when anticipating what the next word in an article should be.
The fact that similar requests result in similar articles, however, highlights the limitations of simply asking, “Give me this“.
5. Can ChatGPT Content Be Identified?
For many years, researchers at Google and other companies have been developing algorithms for accurately identifying content generated by AI.
I'll name one from March 2022 that used the output from GPT-2 and GPT-3 among the other research publications on the topic.
Adversarial Robustness of Neural-Statistical Features in the Detection of Generative Transformers is the title of the research study (PDF).
To avoid discovery, the researchers used several types of analysis to see whether they could spot content generated by AI that used evasive algorithms.
They experimented with several strategies, including adding misspellings and using BERT algorithms to replace words with synonyms.
They found that even when a text had been generated using an algorithm intended to avoid detection, statistical characteristics of the text produced by AI, such as Gunning-Fog Index and Flesch Index scores, could be used to determine if the text was artificially generated.
6. Invisible Watermarking
The product that OpenAI researchers have constructed with cryptographic watermarking to help with content detection produced by OpenAI products like ChatGPT is more intriguing.
A recent article drew attention to a conversation by an OpenAI researcher that is shown in the video Scott Aaronson Talks AI Safety.
According to the researcher, ethical AI methods like watermarking may develop into industry standards, much as Robots.txt established a standard for moral crawlers.
“We've seen over the previous 30 years that the major Internet corporations may concur on specific basic criteria, whether out of concern for legal repercussions, a wish to be regarded as a responsible player, or for any other reason.”
Robots.txt is one straightforward example: if you don't want search engines to index your website, you can say so, and the main search engines will respect it.
Similarly, you could imagine that something like watermarking would just become the industry standard and that anyone who wanted to be seen as a responsible player would include it if we could prove that it works, is affordable, doesn't degrade output quality, doesn't require much computing, etc.
Scott Aaronson Talks AI Safety
Based on cryptography, the researcher created a watermarking system. Any person with the key may test a document to check whether it contains the digital watermark that shows it as being generated by an AI.
The code may take the shape of word choice or punctuation, for instance.
He explained how watermarking works and why it’s important:
“Up to this point, my major endeavor has been a tool for statistically watermarking the results of a text model like GPT.”
Anytime GPT creates a large body of text, we want there to be a subtle, later-verifiable hidden signal in the word selections that would otherwise go undetected.
We want it to be considerably more difficult to pretend that a GPT output is human-generated.
This might be useful in reducing academic plagiarism as well as other types of mass propaganda production, such as flooding blogs with ostensibly relevant comments in favor of Russia's invasion of Ukraine without the need for a Moscow-based troll farm.
or using someone else's writing style to falsely accuse them.
All of these tasks might be made more difficult, right?
According to the researcher, watermarking thwarts algorithmic attempts to avoid detection.
Should You Use AI for SEO Purposes?
AI Content is Detectable
Many claims that Google has no way of knowing if the content was generated using AI.
Since detecting AI is a challenge that has mostly already been resolved, I find it difficult to understand why someone would retain that viewpoint.
It is possible to identify content even when it uses anti-detection algorithms (as noted in the research paper I linked above).
For many years, researchers have been studying how to identify content that has been produced by machines, particularly content that has been translated from another language.
Autogenerated Content Violates Google’s Guidelines?
In April 2022, Google's John Mueller said that AI-generated content is against Google's policies.
“For us, they would essentially still be considered automatically generated content, a category that has been included in the Webmaster Guidelines virtually from the beginning.”
And people have been generating content automatically in many different ways. For us, using machine learning tools to generate content is essentially equivalent to manually rearranging words, searching for synonyms, or using translation tricks like people once did. That kind of stuff
Though we still use automatically generated content, which means it still violates the Webmaster Guidelines, my hunch is that the quality of the content may be a little bit higher than with the truly outdated tools. Therefore, we would classify it as spam.
The “auto-generated” content section of Google's developer page concerning spam was recently updated.
It was made in October 2022 and revised close to the end of November 2022.
The changes reflect a clarification of what constitutes spam in auto-generated content.
It initially said this:
“Automatically generated (or “auto-generated”) content is content that’s been generated programmatically without producing anything original or adding sufficient value;”
Google updated that sentence to include the word “spammy”:
“Spammy automatically generated (or “auto-generated”) content is content that’s been generated programmatically without producing anything original or adding sufficient value;”
That adjustment seems to make it clearer that content that is generated automatically need not simply be spam. That content is troublesome since it lacks all the value-adds and often has “spammy” characteristics.
ChatGPT May at Some Point Contain a Watermark
In addition, the OpenAI researcher said that watermarking would “hopefully” be included in the GPT version that would follow ChatGPT's release a few weeks prior.
So if ChatGPT isn't currently watermarked, it may be upgraded with watermarking at some point.
The Best Use of AI for SEO
The ideal use of AI tools is to scale SEO in a way that makes workers more efficient. For the most part, this involves delegating tedious research and analysis to an AI.
Given that Google specifically says that doing so is not against its policies, summarising webpages to create a meta description may be a legal practice.
It could be interesting to use ChatGPT to generate a content brief or overview.
If content production is delegated to an AI without first being checked for quality, accuracy, and usefulness, distributing it as-is may not be the most efficient use of AI.