Is AI all it’s cracked up to be for bloggers?

Everywhere you turn, there’s a new app or a new service that promises AI can do everything for bloggers.

Is Artificial Intelligence or AI about to take over the world of content creation, elbowing out the way the bloggers who amuse, inform or inspire us? Are we really entering into a new world order where AI writing tools will replace the gifted word-smith?

Or is it just overblown hype following the launch of ChatGPT, the first language model to catch the public’s imagination, much to the “bemusement” of its developers at OpenAI

In this blog post, I examine in detail what bloggers and other content creators need to know when using AI including its limitations, and where it excels. I base this on my experience as a blogger who has been using ChatGPT almost every day since the end of 2022.

Let’s start off by understanding what the two different types of AI are:

Narrow AI, or Weak AI, is a form of artificial intelligence that’s designed for a specific task or a narrow range of tasks. It is limited in scope and can’t do things beyond its programmed capabilities. It can’t “think”.

Examples of Narrow AI include: virtual personal assistants like Siri or Alexa, ChatGPT, image recognition software, and recommendation algorithms used by online platforms. It’s also deployed very successfully by software such as Lightroom to help with photo editing.

General AI, sometimes called Artificial General Intelligence or Strong AI, is a hypothetical form of AI and is not (yet). a reality. Unlike Narrow AI, it’s able to understand, learn, and apply knowledge across a wide range of tasks, much like a human being.

In fact, General AI would have the capacity to think and reason in a way that overtakes human intelligence and is able to adapt. In other words, it would be able to truly think and adapt to various situations.

At the moment, General AI is still in the realms of science fiction, but like many things, may become a reality.

ChatGPT is a large language model-based chatbot launched by San Francisco-based OpenAI as a “research preview” in November 2022. It’s based on the GPT (Generative Pre-trained Transformer) architecture and is a natural language generator. It falls under the category of Narrow AI.

The free version of ChatGPT is based on ChatGPT-3.5. An upgraded version (ChatGPT-4) was released and made available to paying API customers customers on 6 July 2023 (Source: OpenAI).

GPT has been trained with publicly available data, including from the internet, to predict the next word in a document. As such, it’s an example of Narrow AI in a language model.

ChatGPT doesn’t “think” independently. It’s the product of sophisticated mathematical calculations of what words are most likely to come after other words. The text that ChatGPT generates is based on a probability distribution. It’s like a very advanced and sophisticated version of the predictive text on a mobile phone.

As ChatGPT doesn’t think independently, it won’t generate exceptionally good content independently. It only knows what it has been trained on, and that knowledge hasn’t been independently verified. It can get better if you provide detailed prompts and keep going back to refine the question. But it doesn’t “learn” and sometimes won’t remember what you’ve asked from one question to the next.

Do also be aware that over-engineered prompts can lead to even more inaccurate results as GPT gets less accurate the more variables it has to process. However, if the prompt is too short, your responses are likely to be bland and generic..

ChatGPT is just a language model, and not a source of the “truth”. It’s prone to what are referred to as “artificial hallucinations”.

As a paper on the prestigious NIH National Library of Medicine website concludes, although ChatGPT can write credible scientific essays, “the data it generates is a mix of true and completely fabricated”. So the answers it produces can sound really plausible and confident, whilst full of complete nonsense.

OpenAI itself freely admits this. In a March 2023 interview with the MIT Technology Review, John Schulman (Co-founder of OpenAI) said their biggest concern was “factuality, because the model likes to fabricate things.”

The website of OpenAI goes further, and admits it can be “confidently wrong”. They explain that is because it has been trained on a “corpus of data” from the web, which includes “correct and incorrect solutions to math problems, weak and strong reasoning, self-contradictory and consistent statements, and representing a great variety of ideologies and ideas.”

This means you need to verify what it says on a given subject.

ChatGPT doesn’t have traditional “knowledge” or even know anything. There’s a randomised element to the way it generates responses, and it may very well give different answers to the same prompts.

ChatGPT hallucinates as it’s a large language model that follows patterns in texts, and doesn’t inherently know when something is incorrect. See the quote below from Jess Peck which sums up ChatGPT neatly:

One big problem with both ChatGPT-3.5 and ChatGPT-4 at the moment is that their “knowledge” stops as of September 2021, which means they are unaware of anything that happened after that date.

Depending on the niche of your blog, this can be a big problem. The fact they’re both stuck back in 2021 means that when you ask them for hashtags or blog ideas, they’ll just regurgitate what was trending then.

Just think how different the world was in September 2021 – low inflation, low interest rates, many countries still had Covid restrictions in place. You can find out more about how different 2023 is in this blog post: 101 Blog Ideas for Autumn 2023.

How can your blog stand out if it’s just saying the same old thing?

I’ve been using the free version of ChatGPT since the end of 2022 I started experimenting with it for use in blogs, checking code for web development, and as a general thesaurus.

I find it easy to use, but very limited in use. The reasoning is simplistic and the writing style oddly formal, but not particularly articulate. It doesn’t have the wit and lively turn of phrase that I enjoy reading and try to emulate myself.

Ideas for blog post topics and suggested outlines for blog posts are bland and generic. I don’t like the writing style, even when I prompt it to write in British English. At best it can provide a checklist, but I wouldn’t use anything it produces without extensive editing.

I’ve used ChatGPT to try and fix glitches in coding. I’d say it’s successful about 25% of the time and is mostly a rabbit hole that has no exit, other than quitting ChatGPT. A complete waste of time.

When I’ve asked it specific questions about factual situations, it’s been very hit and miss. The free version doesn’t show the sources, so I have no idea if it’s accurate enough without then double-checking elsewhere. At least with Google, you can often find primary (and reliable) sources.

Another big gripe I have with ChatGPT is that its memory is stuck in September 2021. This means that if you ask for suggested hashtags, topics for blog posts, and anything topical, it will advise you based on September 2021, back when interest rates in the UK were still 0.25% and CPI inflation 3.1%! This means it’s useless for my main blog The Independent Landlord, as it doesn’t reflect anything accurate about the new Renters Reform Bill or investing in a high inflationary, high interest rate new world.

What I have found it very useful for is synonyms and asking it to rephrase things. That said, I’ve needed to be very selective in incorporating the suggested synonyms into my work. I use it as a tool, and not as a replacement “directing mind”.

AI can be of help for bloggers, if used selectively, with a judicious amount of editing. It very much depends on the niche. I’ve heard ChatGPT can be useful for recipes and to summarise books (provided they were published before September 2021).

There are lots of AI tools that will draft meta descriptions, check spelling, produce images, help with graphic design and paraphrase blocks of text. This can all save a lot of time.

However, it AI isn’t discerning, and currently needs a lot of close attention from content creators before it’s fit to publish.

Google claims to be agnostic about whether content is generated by humans or by AI. What counts is having high quality original content.

However, as ChatGPT will not ordinarily produce original content without significant human intervention, isn’t trustworthy and is just a language model, I can’t see how it can produce -A-A-T high quality content. It will produce content that is like other content on the internet, as that’s where it has “learned” its trade. It will be difficult for ChatGPT content to stand out, because it is simply regurgitating what’s out there already.

Also, despite the above, Google is already differentiating between human-generated content and AI-generated content with its policy towards reviews:

Bloggers should therefore be very discerning about how they use AI.

Bloggers and content creators should view AI as a valuable tool that will become more and more useful as it becomes more and more sophisticated.

Bear in mind that it’s not much more than a very advanced form of predictive text. Sometimes it will predict well, other times it will be worse than weather forecasts!

To give an idea of where ChatGPT is at the moment, this is the response to the prompt “Please give me 5 tips for bloggers so they get the best out of ChatGPT”:

Google wants to show helpful content to users. If users don’t stay on a page for very long, that will harm the page’s rating.

What’s your experience as a reader when you read the text in the blue box? Did you learn anything? Do you like the turns of phrase? Did you find it engaging? Did you feel satisfied with it? Is it genuinely helpful? Would you leave the page? I would!

The phrase “intriguing introductions” is proof in itself to me of how far ChatGPT has to go before it stops sounding cheesy and AI-generated. It sounds like a 13 year old has been “enriching” their writing with a thesaurus as a homework exercise!

I also take issue with the assertion that ChatGPT is good for fact-checking. Given it’s propensity for “artificial hallucinations” and being “confidently wrong”, I would not use ChatGPT for fact-checking. ChatGPT has been built scraping content from the whole web, and we all know how much of what’s out there is inaccurate. If ChatGPT”s inputs are inaccurate, how can its outputs be relied on?

When using AI to write blog posts, it’s important to keep quality at the forefront of your mind. It’s what Google refers to as creating helpful, reliable, people-first content.

Here is a 7 part checklist for you to self-assess your AI-generated content before publishing. This checklist will help you to create readable, useful content for your audience that will also rank in the search engines:

1. Does the content just rehash what the AI tool has scraped from the internet?
2. Does it provide added value to the AI-generated response and what others are saying on the topic?
3. Is it the sort of content you’d enjoy reading and want to save or share, and even sign up for the newsletter?
4. Does the content appear trustworthy? For instance, is it well-researched with links to primary sources when discussing factual information? is there evidence of the expertise involved? Or is it full of non-substantiated assertion?
5. Have you checked every fact stated? If the content draws on other sources, does it avoid simply copying or rephrasing what they say, and instead provide substantial additional value and originality on the source?
6. Will the reader feel like they’ve learned something after reading the content, or go back to Google?
7. Is it easy to read, without clumsy phrasing and too many basic transition words like “Overall” (which ChatGPT overuses).

Artificial intelligence has huge potential for bloggers and content creators. We’re still at the start of the journey and the quality of the output will only improve.

Think just how far mobile phones have come, from even the Nokia 3210, which itself was game changer as it made texting easy and it could fit in a back pocket. I’m old enough to remember how amazing we all thought the Nokia 3210 was in 1999, and look how far phones have developed since then.

It will be the same for AI. We’re just not there yet.

However, as Neil Patel warns, when AI content becomes ubiquitous, search engine algorithms will prefer human-written content because it’ll get rarer. Even if you do start the writing process with AI, it’s likely you’ll need a real person to edit it.

As for me, I like techie things so I’ll use keep using AI as a tool. However, part of the pleasure of blogging for me is the pleasure of writing. I believe it comes across in what I write. So I’ll keep generating content myself.

To answer the question at the beginning of this blog post, no, I don’t think AI is all it’s cracked up to be. But just as AI has become an aid to editing photos in Lightroom and Photoshop, it will be an aid to content creators and bloggers. We just need to be careful how we use it.

About Suzanne

Suzanne is lawyer and expert blogger, based in the UK, and runs the successful blog, The Independent Landlord.

She set up The Web Smith to help entrepreneurs build fantastic blogs to increase organic traffic to their websites and grow their businesses.

The key to building a blog with high organic traffic is to consistently create high quality content that the target audience wants to read. Then this content needs to be optimised for search engines, ie SEO.

Suzanne works with entrepreneurs to help them devise and implement a strategy for their website that’s right for their business. As part of this, she provides one-to-one consultation, SEO audits, website audits, content creation and editing.

You can sign up for The Web Smith free newsletter here.

Read More