No products in the cart!
Please make your choice.View all catalog
This article is intended for SEOs including freelancers, and in-house and agency professionals.
Learn how to leverage ChatGPT for SEO tasks and avoid bad use cases.
It will help you increase efficiency in SEO and will also help navigate the landscape of generative AI and LLMs so that you don’t get tripped up by using ChatGPT for tasks it is not well suited to.
The trouble with AI using LLMs is that they are designed to make stuff up – it’s literally what they were built for – they know very well what words are likely to come next.
But do not mistake this for properly understanding and having real insights.
The technology is very impressive, but as with many tools, it needs to be handled by someone with a good understanding of the tool and crucially the underlying technology.
This is akin to the “Every problem starts to look like a nail when the only tool you have is a hammer”.
Image produced by DALL-E
Ask yourself this – would you let your predictive text messaging write your keyword research strategy for you? That’s essentially what you are doing.
As with many things, the quality of the answer is related to the quality of the question.
How you formulate prompts  and what you get out matters a lot.
GIGO = Garbage In Garbage Out
You need to test your outputs and read the content before blindly trusting it.
LLMs and generative text generators do suffer from hallucinations  – they make up things that are not true.
They are designed to make things up – don’t be surprised when they do!
Just as Google isn’t a truth engine , it’s a search engine (it will find something relevant to what you are searching for) LLMs are not fact generators.
“AI Content is better when it is augmenting and adding to human-produced and cultivated content.”
Here are some of the things we think the technology is very well suited to.
One of the best uses I have found in SEO for generative text is to do the boring bulk meta descriptions.
But I do this with a twist.
I use as much data as possible from a page, gathered with importxml and a screaming frog crawl, I use a lot of page information to formulate the prompt that ultimately produces the unique and specific meta description.
This won’t do any SEO magic but it will efficiently deal with a large amount of pages missing their meta descriptions.
See where this should be prioritized in our deep dive on deciphering SEO audits .
Create a meta description of length 130-155 characters not to exceed 155 characters considering the following information:
=GPT(CONCATENATE($E$1, A3, B3, C3, D3))
Where $E$1 is the prompt
A3 is the URL
B3 is the Page title
C3 is the Heading
D3 is the body tet
You can use Xpath to pull in these features and or use a Screamingfrog crawl output.
You then provide a lot of real detail and context from the page itself.
A well-set-up prompt and GPT can be useful for assessing your content with Google’s helpful content guidelines  in mind.
Read more here from Matt Digity on how to Audit Your Content with Chat GPT According to Google’s Guidelines .
Creating title suggestions is a good use of the technology.
The use of titles generated by AI likely won’t be an issue in terms of SEO. You can even add a degree of specificity to get the format and length of titles you seek.
I like to use this for inspiration.
The title of this article could still be one of a few different variations from my original thought.
Presently my title is “The Good, the Bad, and the Ugly of Generative Text for SEO” I had a couple of other title ideas too, but for the sake of this article, I used Chat GPT  to give me some title suggestions. Sometimes they can inspire you.
In this example, it kind of missed the mark because it didn’t have enough context and didn’t convey what I was going for. However, if I refined my prompts and the responses and gave it additional feedback, or even just the whole text when finished it probably would do a much better job.
Or ….. I could just think of an appropriate title 🤔.
I have found document outlines and just outlines in general to be quite a good use case for the technology. They are broad enough to be useful for most generalized topics where there is a lot of already published content.
That may be a key point here though, you are likely to get the best results when there is already a plethora of information available on a topic, which ironically makes it less likely you will be able to provide something valuable.
Nonetheless, Chat GPT  for general outlines is fairly good and can be useful to get you started on something that gets refined later. I sometimes use it for that application on my stuff.
I asked Tom Winter what he thought the best use of the technology was. Here is what he said.
Generative AI is revolutionizing the way we create content.
When used effectively, it can produce outstanding outcomes. The most successful strategy I’ve found is the CyborgMethod™, which combines human insight with the speed and efficiency of AI, allowing for the production of content on a grand scale without compromising quality.
By integrating AI into each step of the writing process, you tap into a wellspring of creativity and vastly increase our output. For those interested in exploring the potential of this approach,
I encourage you to look into our AI Writing Challenge at https://seowind.io/ai-writing-challenge/ .Tom Winter – seowind.io 
Generating schema based on supplied details and information can be very useful.
What would take an age to do manually, generative AI can do en mess – and again, do this with very specific inputs and prompts.
I like to use Kristina Azarenko’s SEO Pro Chrome Extention  to quickly inspect schema.
Here is what Chat GPT  gave me. Note the data is not correct, you still need to check and make sure the data fields are correct, but for quickly generating the code in the appropriate format it can be a quick and handy little use case.
FAQs are a useful feature that enhances user experience and not enough websites have good FAQs. FAQs are used in the SERPs, however, their coverage of the SERPs  has been decimated in recent algorithm updates. Nonetheless, I still consider these to be useful features (and I suspect they will find their way back via SGE).
And of course, once you have your FAQs generated you can use the ChatGPT to put them into schema for you.
I have quite a strong suspicion that schema FAQ will be used heavily in SGE (if it ever actually rolls out).
Rephrasing is a good use of generative text technology.
Sometimes you just get writer’s block (I know I have several times while writing this) or something just doesn’t read quite right.
Using Gen AI to rephrase text can give you additional ideas.
I often use it just to get me started.
Below is an example of the rephrasing tool in Semrush’s Content Writing Assistant.
A good use of gen AI is to extract key points and action items.
I use it to transcribe client calls and have it give me the action items. This way you don’t miss adding and assigning an action item from a meeting. This is one area where AI shines.
Similarly to extracting key points and action items, gen AI can take a text and summarize it for you.
This can be great for making TLDR sections.
Using GPTs to produce RegEx can be a big efficiency hack.
Getting custom RegEx right is quite difficult and can be time-consuming and painful to debug and get right, but GPTs can easily knock this out for you.
I have been incredibly impressed with the speed and efficiency at which GPTs can produce reasonable workable code and also debug code.
Here is an example I ran through Chat GPT.
🤖 The Good, the Bad, and the Ugly of Generative AI for SEO 👍👎🤔. In this article, we explore 🔍 some different use cases for the technology and explain how to leverage 💡 the technology and some areas where you probably shouldn’t! ⚠️ We feel that AI is best used when it augments 👥 human-produced content, not as a replacement for it ✨.
I am not a fan of Gen AI for Organic Pages and content, however, for other content forms, it can be a useful efficiency tool.
I like to use Anyword  for getting started with PPC headings and descriptions. You can create personas relevant to your campaigns and it will give you the alignment of your ads and the demographics of those personas.
What this platform does is compare the performance of multiple different ads and the performance and gives a score for how well the ad should perform.
Image from Anyword.com 
One platform I have not yet tried is AdCreative.ai  which from what I hear is “like Canva  and AI had a baby”. This platform suggests that it produces multiple different creatives from minimal inputs. While it is likely that modifications and tweaks will be needed it is likely a useful creative efficiency booster. I am excited to try this platform.
Image from AdCreative.io 
Now let’s look at some examples of where the use of this technology is not so good.
One of the main reasons for being considered bad here is that this is done in the absence of any other checks and balances and without expert review.
Just letting AI blindly produce content is very risky. The jury is still very much out on the longevity of AI-produced content and its viability in organic search. However, I predict that most use cases will see a sharp decline after their initial ~6-month increase in traffic (most people don’t show you that side of the case study).
Yes AI content can rank.
Yes, Google changed its language on the use of human-generated content that seemed to give a free path to AI content (as long as it’s helpful). Read more on Google’s shifting approach to AI content .
There is also a growing body of evidence and case studies showing that the long-term fate of AI content is not good.
An audit is a poor use of Gen AI technology. The technology is generative and uses other audits as the basis of the text generated to pass off as an audit. It is not a real audit, it just looks like it.
It misses important analysis and audit specifics. It misses context and situational awareness.
Note that there are specific analysis add-ons and features that can be used with Chat GPT and other generative AI platforms.
They can perform some analysis on supplied data so you can do some real analysis on data.
However, generative text is not an appropriate technology for conducting analysis. You need real functions and interpretation of the data and function outcomes to provide the insights that give the analysis value. AI missed the mark significantly in this area
This one may seem like a surprising one. After all the base technology here is NLP (Natural Language Processing) and the ability to “understand” context and intent.
Therefore, search intent classification should be very good.
The trouble is though, that the context element is missing and the alignment to the searcher, the search query, and the SERP intent is not there.
I have seen many use cases where people are asking Gen AI to predict something.
The very notion of this approach is flawed.
They know everything they know based on history. And we know in most cases history does not predict the future (but we can learn from it and sometimes it does repeat itself).
The very reason most investment advice states “past performance does not guarantee future gains” is because you can’t predict the future. You don’t know what will change.
At best you can project and use models but this isn’t how Gen AI works.
I do accept you can give it a data set and ask it to use a model but it’s not the right technology for these types of tasks.
The worst thing about a strategy developed by chat GPT is that it looks like an actual strategy. At first glance, it does pass for a strategy. But it’s made up based on, and only based on other documents talking about strategy.
A strategy is only really useful when it’s developed for your business with your goals and resources in mind. Template and boilerplate strategies don’t work for a reason.
Using generative AI and LLMs to produce a strategy is up there with the worst uses for the technology.
Keyword research is another area that is poorly suited to generative AI technology.
Some points for it is that they do leverage NLP and can understand relevance quite well.
However keyword research, and more specifically modern keyword research is not well suited to generative AI.
In this article, we have reviewed some of the uses of generative text technology.
We have explored good, bad, and ugly use cases. We have provided some examples and scenarios with discussion.
Note that this is a rapidly evolving space with new developments coming all the time.
If you have an example that belongs on this list, contact us and let us know.
Disagree with something? Contact us, we would love to discuss it.