The LLM and AI landscape continues to evolve with new developments and tools launching all the time. It can be overwhelming to understand it all, let alone attempt to integrate all of these tools into your existing workflow.
I've been there, felt really overwhelmed and that I had to know everything to get started. It doesn't have to be like that at all, and this is exactly why this guide was created.
From all my experimenting with LLMs, I’ve written up this actionable guide complete with prompts, resources and screenshots which shows you how to create a complete content strategy workflow using LLMs.
I've listed more than 30 ways to use LLMs in each phase of the content strategy. It encompasses each phase of a typical content strategy, starting from audience insights all the way to post-publishing evaluation.
In this guide, I refer to the following:
Microsoft CoPilot
ChatGPT Plus
Google Gemini (Previously Bard)
OpenAI API (to use GPT for Sheets)
I highly recommend investing in some of these tools. While you can try using the free versions, be mindful that some areas will be locked, including the ability to use and create your own GPTs which this guide refers to a lot.
And I also mention some SEO tools you may need which includes:
Crawler (Sitebulb or Screaming Frog)
Access to Google Search Console
Access to Analytics platform (eg Google Analytics)
There are plenty of ways to approach the methods I have written about, these are what I found in my personal experimentation of them. It is highly likely you'll find your own approaches that work for you!
Why listen to me?
I’ve spent a lot of time experimenting with LLMs to automate tasks, and create my own tools which I've shared with the SEO industry. My personal mission is to automate all manual processes with the authentic use of AI and solutions such as Python, and find innovative ways to optimise my content workflow.
I’ve also created many GPTs which are being used by the SEO community and beyond. You can find them on this site on the Tools section, or on my Twitter thread where you can find out as soon as I’ve created something new.
Personally, I still feel like I’m at the beginning of my journey in regards to the AI and LLM landscape. There’s so much to learn, discover, and test! My personal goal is to build your confidence in trying out LLMs and experimenting on your own in the same way that I did :)
If you’d prefer to watch the Moz webinar, here is the link to visit! Also follow me on Twitter @CaitlinTheSEO or follow me on LinkedIn to find all my latest AI resources and tips :)
Here’s a complete list of the methods I’ll go through in this article, click on the list to jump to that specific section.
Audience Research:
Content Inventory & Auditing
Content Ideation
Content Briefing
Content Editing & Reviewing
Content Repurposing
Analysing & optimising performance
Before we start, a note to please be careful when immediately offsetting your content processes to LLMs. Prioritise human reviews to check the outputs while you get started. LLMs are prone to getting things wrong and it's best you can catch these before integrating LLM as a standard in your workflow and publishing content.
Audience Insights & Research
We’re going to start by finding unique ways to easily understand our website, find audience insights and also determine potential competitors using LLMs.
We will see how to also do this with competitor sites and reveal parts of their digital marketing strategy. We are then going to find how to find audience segments and find useful ways to extract this information using AI.
There's also a solution to building detailed user personas and stories using GPT to effectively target audiences.
Audience data using Microsoft CoPilot
I’ll be using a site called Pooch and Mutt (which will be referred to as P&M) as an example to showcase this section. I’m not in any way affiliated with the site! But it has the cutest doggos and a niche offering, so I just had to use them.
Look at those precious doggos!!
Admittedly, I don’t know much about P&M. So how do we find out some more about it using LLMs?
Enter Microsoft CoPilot.
Microsoft CoPilot is Microsoft’s Large Language Model and it can be accessed for free in the following examples.
I'll focus my examples using Copilot from the Edge browser.
Microsoft Edge is automatically installed on Windows devices. You can download Microsoft Edge browser from here if using a Mac.
Now, open your Microsoft Edge browser. At the very top right, underneath the button to close the browser, you’ll see the CoPilot symbol. Click on the blue symbol.
The CoPilot interface will open up, showing three options: Chat, Compose, and Insights.
Chat will let you interact with LLM and ask it questions about anything, in whcuh you will receive a response back using Generative AI.
For now, let’s focus on Chat as we want to ask CoPilot a few questions, so click on that option.
I usually use the Keep Balanced prompt on CoPilot, but change it according to your preference.
Let’s start simple. I’ve added a prompt in the browser interface:“Tell me about PoochandMutt.co.uk”
Helpful enough!
I wanted to find out more about their market offering and why they stand out. So I simply prompted what are their USPs?
Also helpful and now I have a better understanding of P&M's market offering and why they stand out :) you can ask as many questions as you need to get a better understanding of the site.
You can also ask questions based on the page, so if you wanted to know more about specific products or services, you could engage directly with the page itself by asking about it, or clicking an option such as Generate Page Summary.
Now we will focus extracting some Audience Insights from P&M, so click on the Insights option in the top of the Copilot interface.
It will unveil a load of insights about the site, using aggregated data from Microsoft Edge and Bing.
It will reveal the following data:
Monthly traffic: Uses data aggregated from Bing. Doesn’t give traffic specifics, but if you hover over a month, it will show how it has increased or decreased vs last month. You could confirm your own site's traffic metrics on your Analytics platform.
Most visits from: reveals the top countries visiting the site. This can be helpful to see the main markets visiting the site, and may give a glimpse of a site’s internationalisation strategy.
How visitors found their site: Breaks down the amount of people visiting from search engines vs referring sites. Gives a slight glimpse into those referring sites It can be quite revealing about a wider strategy if you’re able to do a bit more digging.
Here's some insights I got from the above screenshot:
P&M might get traffic from newsletters.
They have voucher options to direct traffic to the site
They have engaged in partnership promotions through Take a Break (UK magazine a digital publication) - would be good to know if/how competitors are using this method too. .
People also view: Sites similar to yours. This can be helpful to find potential competitors or sites with similar content.
You can apply the same method as above for the ‘People also visit’ to see other potential competitors and gain insights into their audience.
The people also view came up with Tails.com, a very similar site to P&M, as it offers nutritious dog food, however with a subscription service. Still worth looking into how visitors are finding their sites and how we can cross-compare that to ours.
Open up the Insights section again for the competitor site:
We can now directly compare the Monthly traffic visualisations alongside one another, and ask key questions:
How is our monthly traffic compared to theirs?
How does our month-on-month traffic compare to theirs? Is it similar or very different?
Are there any potential seasonality trends we can see between the two brands?
How strong is their brand?
Cross-correlate with your actual data and a tool such as Ahrefs or Semrush that gives estimated traffic for a more holistic view.
The ‘most visits from’ section could provide insights on a competitor's internationalisation strategy.
We can glimpse the international markets they are receiving the most traffic from, which could help us determine where we should prioritise building a presence. This would be confirmed by follow up research and determining how they're targeting those countries. .
How visitors found this site becomes even more insightful and useful when doing competitor analysis. I found it to be helpful to view other channels or other avenues to potentially start building on other channels.
Tails.com clearly has users visiting from email judging by Live and BT. However, its likely users visiting to manage their subscriptions, as I couldn’t find an active newsletter. Interesting insight for P&M to take away!
Tails.com has a very large Facebook following of 175k followed, with an active community. It would be worth looking into seeing how they engage their audience and convert them.
Rinse and repeat this process as much as you wish, and record all of the potential competitors, and the insights you gain from them using this method.
You'll walk away with a more powerful understanding of the market, your competitors, their USPs, as well as a glimpse into their strategy and performance.
Next, we will look into how our audience are engaging online across different platforms.
Find Breaking News using Google Gemini
There’s a method to find the breaking news of the day using the free version of Google Gemini.
Quick note on Gemini - Google Gemini (previously Google Bard) is Google’s LLM developed by Google Deepmind. You can interact with in the same way as you would with a Chatbot and it will output a generated response based on what you've entered.
The following method would be helpful when experimenting with News content to understand what your audience is engaging with. If you need a quick way to find relevant, industry news, this will be especially helpful.
This method works best for more active industries with lots of news, such as Tech or Finance.
You can prompt Gemini with a simple question asking it about the news in your site’s topic area of that day, or week:
“Tell me the very latest news in pet nutrition, in the UK that's happened today.”
You can also prompt to include sources by reworking the prompt as such:
“Tell me the very latest news in pet nutrition, in the UK that's happened today. Please add any original source links.”
Using the example of pet nutrition news in the UK, there isn’t much news to constitute a whole News strategy, but it wouldn't make sense for P&M.
There’s still some interesting news that P&M could address and create content on as an authority in its space as a pet nutrition eCommerce.
For instance, it could address that worrying statistic on Dog Obesity from Burns Pet Nutrition, I would create a guide educating dog owners on how to properly feed their dog and the importance of proper nutrition , with product plugs discussing that it's the best solution to keeping your dog a healthy weight.
Also an article discussing the latest nutrition course offered by the UK government, with a key summary of the course and applying their knowledge and expertise, would be another potential idea they could use.
Another way to source topics that are already engaging your audience and bringing traffic to the site is by looking at your User data on Google Analytics.
You can look at User by Interests to see broad, relevant topics to narrow down what to search for using this method.
I recommend capitalising on the interests that are bringing the most users to your site as this is resonating well with your audience.
If you were in Tech, Finance, and generally more active industries, this method is helpful to use.
I can see this use case of searching for live news using Gemini getting better over time as they integrate more Google products such as Google Trends and potentially Google News into Gemini. It’s better, but still not perfect, and you should fact-check any news as much as possible
Ideally, this would help with:
Saving a lot of time researching for relevant news and trends for your site
Perhaps you could even put your site in and ask it to look for relevant news and trends that would be suitable to
Subreddit Research
Reddit is home to engaged subcommunities of users in particular niches. This can be a treasure trove of opportunities for content marketers when looking for how their audience is engaging online, the topics they're discussing and questions they're asking.
We can also use Gemini to help us source Subreddits that our audience would engage with, based on some slight context about our audience.
Here is a really simple output I used to ask Gemini to source relevant subbreddits for me. I gave it a brief but broad context about my P&M audience:
Here is the output it gave me:
I liked the structure of this response, grouping then by size so it was helpful for me to scan. I would likely look for the medium, smaller communities as it would be easier for me to connect with my audience. I also appreciated the description about each of the subbreddits, so it was really easy for me to see what was relevant and what wasn't.
Annoyingly, it hallucinated the community size, as it struggles to extract that information live. I couldn’t get it to be accurate, no matter what I tried. Treat this as an estimated size.
You can also ask Gemini to export the information as a CSV for you and have it contained in your Google sheets, really efficient if you use Google workspace like I do.
I particularly loved how it took my prompt and titled it as the name of the sheet. Not sure how or why that happened, but I’m a fan of its chaoticness.
You can continue to use this method to source relevant subreddits and append them to your chaotically named spreadsheet - just make sure to check these subreddits do in fact, exist.
One way you can do this - create a new column with
Eg:
=CONCAT("https://reddit.com/",B2)
To transform them into URLs.
From there you can easily explore them to see if they exist.
Scrape questions from our gathered Subreddits
We now have potential subreddits we can explore to connect with our audience better. How do we extract discussions and questions from them… en masse?
The next step is not an LLM method, but I’m adding it anyway - technically it has an LLM element attached so I’m allowing it :)
Big props to Keyword Insights for this awesome step-by-step to get you started with scraping your Subreddits for questions.
The script then retrieves as many questions as it can from these subreddits. It then compiles all these questions into a single column within a new CSV file.
Thanks to their Google Collab form, I was able to scrape 1k questions from all the subreddits within a matter of minutes. This could be helpful to look for popular topics audiences are discussing and plan content around.
If you have an OpenAI API key, an optional final part of the script will ‘clean’ the questions to help make the data easier to analyse. Highly recommend doing that step!
Now you've got a whole sheet full of authentic questions your target audience are asking on subreddits. You can decide how to use this to your advantage to engage that audience with your brand.
You can create the right type of content to address those questions, or you might find different ways to engage them with your brand directly in the Subreddits.
What topics are engaging users the most that we should capitalise (Google Discover)
It's impossible to know the ‘formula’ to appear in Google Discover. Google Discover is very spontaneous by nature, and you need to be experimental with consistently writing content around the topics your audience are interested in.
But if you're not quite sure on what exact topics are resonating with your audience, it makes it tricky to plan and resource that content.
I've created an experimental method which works if you have appeared on Discover before, as it will use GPTs to analyse your historical data. It may be helpful to better understand the topics your site is appearing for and what's resonating with your audience.
Before we get started, this method uses a GPT on ChatGPT.
ChatGPT is an AI chatbot developed by OpenAI. Very similar to CoPilot and Gemini, you can interact with ChatGPT in the same way and it will output answers depending on the context you've fed it using generative AI.
Custom GPTs are custom versions of ChatGPTs, primed with instructions, functionalities, and sometimes a knowledge base of information.
Think of it like mini apps within ChatGPT - you can make GPTs yourself for free or use other people's within the GPT store.
You will need the paid version of the ChatGPT plan to access the GPT.
In the following example, I have used Discover data from a remote working community.
You will need to:
Go to Google Search Console.
Extract your site’s Google Discover performance over 12 months.
(optional) Use Sitebulb or Screaming Frog to scrape article content and date published from those Discover URLs (Google Discover).
For the optional step, once you have your Discover URLs, crawl the list of URLs through URL List ON Sitebulb /List Mode on Screaming Frog.
Extract usual on page data including :
Title Tag
H1
Using Custom Extraction, extract the following:
Article content
Date Published
You can do this easily now by running a URL in the rule sections while you're setting up the custom extractions.
Load up an article and click on the main article body content - make sure it selects the whole content only, and set that rule up.
Repeat this for the Date published. You may need to find this in the Source code if its not on the page, or if its only showing the Date Updated.
Both crawlers have now have a feature to select what to crawl on a page, making custom extraction heaps easier!
Stitch together your Google Discover data from GSC, with the crawler data you extracted. Use VLOOKUP to stitch the data into the same rows together.
We will then look for potential trends including:
Seasonality
Topic Patterns
I have created an (experimental) G Discover GPT to help you extract helpful insights to better understand your Discover data. Feedback is very welcome on this!
Attach your stitched file to the GPT interface like so, and add a simple prompt.
In the output, it discussed a myriad of areas including common content themes and topics, insights about performance metrics
It encourages me to explore more city-specific guides on less known locations, as well as cost-of-living guides for digital nomads in certain countries.
This is helpful; to have a backup of themes as well as data analysis when you’re not sure and are in a position to experiment.
Example questions to feed the GPT:
Please analyse my Discover data and give me insights on content themes that consistently appear.
Any insights I can use for title tags / h1s based on articles with high CTR?
Are there any seasonal content trends based on the date of content published?
Are there specific content types (guides, how-tos, personal stories) that perform better on Google Discover?
What is the sentiment or tone of the top-performing content, and how does it relate to engagement?
What are the formatting features of the most engaged articles?
What is the average article length for the top-performing articles?
You can then use the insights to inform how you plan content with your audience and Google Discover in mind.
I would love feedback from those who have tested this method. This is a recent experimental GPT process I've made!
Generate User Personas
There’s currently a great GPT to help you generate user personas for you brand. It’s called AI User Persona Generator by FeaturedGPTs.
In my use, it was on the mark with analysing a potential user, their painpoints, challenges, motivations , and really good at generating a detailed story around them.
The interface asks you to add to your site, and after an analysis, it’ll share its output with other questions to ask:
As you gain more information about these personas, I recommend recording these in a doc and capturing information there, editing where you like.
These can then be accounted for when you plan new content and how you address the painpoints and motivations with a more specific outline of your customer in mind.
Content Inventory & Auditing
This section will help you have a better understanding of your content inventory, marked up according to key content themes.
Group & Tag Content by Themes
Content tagging can help us understand our content performance, offering insights into what resonates best with our audience and whats converting. This can inform our content strategies, helping to prioritise themes that engage your audience effectively, or identifying the lacking themes that may need more addressing, a different approach, or cutting altogether.
I’ve provided a long step-by-step on how to tag up your content inventory using the OpenAI API in sheets.
We will bring GPT to sheets to assign themes to all of our URLs using the OpenAI API. OpenAI owns ChatGPT, but their API is separate from ChatGPT. Both API and ChatPGT use similar generative AI models, but the API can be built into applications and integrated into tools which makes it incredibly useful.
This is a method best used on small-medium sites. If your site is in the 10ks to 100ks of URLs, you should explore solutions using Python to scale more efficiently and cheaply.
Firstly, use your crawler of choice such as Sitebulb or Screaming Frog to gather all of your internal URLs. You can decide to crawl your entire site, or split this into segments such as products, and blog URLs.
Alternatively, grab your sitemap which you can export easily with a tool like SEOwl, or manually in the sitemap.
You can use another method such as Title Tags, or H1s - anything you feel summarises your content well and works for you.
Set up your account through this link on the OpenAI platform.
Once your account is set up, log in to your OpenAI account.
You’ll need to add funds in the billing section. £10 ($12.68) has gone a long way for me using with GPT 3.5 turbo, I still haven’t run out of credits yet.
Navigate to the API keys section.
Create a new secret key and keep the details safe.
Now that’s all set up, you’ll need to open a new Google sheet and add an Apps Script.
Watch this video by Tech With Hitch to install the App script. Make sure to set it to GPT 3.5 turbo as GPT 4 is much more costly, and unnecessary for this task.
GPT has now “entered the sheet” through the formula:
=GPT()
It works the same as you’d use ChatGPT, but it doesn’t have its advanced functionality eg web browsing.
Here is the formula structure I have been using:
“Add your prompt here” - add what you would usually add in ChatGPT
&CELL - to append your cell.
Now, to create the formula to tag our content. Let’s add some instructions to our formula to tag our content according to themes.
Why do we add themes? With predefined themes…
Easier to classify groups
Consistent
Less room for specific categories
These are examples of my pre-defined themes for P&M:
Dog Health
Dog Behaviour
Dog Breed
Seasonality/Events
Dog Recipe
Other (must be added as this would be a category for content that doesn't fit in the themes)
And the formula structure:
=GPT("Based on the string provided, assign it to one of these themes:
'ADD THEME 1',
'ADD THEME 2',
'ADD THEME 3',
'ADD THEME 4',
'ADD THEME 5',
'ADD THEME 6',
'OTHER',
Use the examples as a guide but return only the theme name in a concise form, without any additional text. Examples:
\nString: 'overweight-dogs' - Theme: 'Dog Health'
\nString: 'chihuahua' - Theme: 'Dog Breed'
\nString: 'how-to-help-a-hyper-nervous-badly-behaved-dog' Theme: 'Dog Behaviour'
\nString: 'how-can-i-encourage-my-fussy-dog-to-eat' - Theme: 'Dog Behaviour'
. Output only the theme name, without any prefix or quotes. Now, analyze this string: '" & A2)
Set up your URL strings by splitting by the / and measuring the very end of your URL:
=split(A2,"/")
If you decide to not use URLs but another measurement such as H1 or title tag, you don't need to perform the above step. Just analyse according to that and adjust the GPT prompt to reflect that.
Now to apply the formula to the sheets. See how this quickly outputs below!
Boom! Make sure to check the content themes as part of a human review before moving on to the next step.
If you really wanted, you can drill down on the Other theme and split that out more specifically, especially if you are seeing lots of URLs in Other.
Just to show you what this looked like without these predefined themes…
Themes all over the place
Inconsistent themes / a bit chaotic
Simply repeated the URL back to me in some cases
Overall, not very helpful
You’ll need to do some quick data cleaning on the themes as sometimes the naming has slight nuances. GPT isn’t perfect :(
Common issues included GPT appending ‘Theme:‘ as well as some rogue formatting.
Here’s how to fix:
CTRL F for Theme:
Replace all instances on sheet with the blank box
Filter the theme column by Unique B2:B (or column with theme)
CTRL F each incorrect theme, CTRL F replace method above or manually update to the correct theme.
I advocate checking as many of the proposed themes on your URL during the data cleaning process as possible to make sure you’re happy with the applied themes.
To find the total number of content within a theme, simply perform a quick COUNTIF formula to count the theme name against the entire column.
Example:
=COUNTIF(C:C,G4)
(where C:C contains all the themes, and where G4 contains the theme name)
Once that’s done, and you have clean themes applied to all of your URLs, you can go on to the next step, where we will stitch valuable data such as conversions and traffic data.
Drawing insights from your content
Now you have your organised content themes. But what can we do to make it more useful and insightful to us to use?
As mentioned, you can enrich your content themes against your other first-party data such as traffic and conversions from Google Analytics.
In the process of extracting your internal URLs, you can also extract your Analytics data on the page level, and stitch that valuable data such as your traffic and revenue to your URLs, in the same sheet.
You can also make decisions about where content should be housed on the site, according to your new content hubs.
You can even overlay the URLs with conversion/revenue data to see how each theme is contributing:
This will help you in creating visualisations (hopefully much better than my simple ones) where you may want to illustrate the current state of the content, it's opportunities and challenges on the site to stakeholders.
Content Ideation
This section will focus on unique methods to ideate new content ideas with the help of LLMs. You might have a good idea of content you want to research into from the previous step (seeing the themes which are lacking content for instance), and there’s also a follow up step using the same theme tagging method!
ChatGPT for keyword research (hint: please don’t)
If you reeeally want to use LLMs for keyword research, Paul Shapiro’s guide is a fantastic one to follow. He walks through how to connect your GPT to DataForSEO to feed you real metrics. However, this works best when you add in your own keywords.
In a later stage, you could connect it up to your content brief GPT (which we will run through) to feed you the real keyword data in the brief.
Content Gap Analysis through Competitor Analysis
Referring back to the previous method with our theme tagging using OpenAI API, we can perform the same steps to analyse a competitor site’s URLs according to our pre-defined themes. This will help us to see which content themes they’re building on.
We can then compare them to our inventory, and directly assess our content themes against a competitors.
This will help you to see what you’re performing well in, see where you’re weak in, or even brand new areas where you can dominate market share as there are few competitors.
Now we can directly cross-compare our content across our competitors by theme.
Lots of questions emerge from this visualisation (and not on how simple they are):
How much more content has our competitor got in a certain theme?
Has our competitor got a similar focus on content themes as us? Do they approach more themes more than us, or less?
Are there any themes that we haven't explored but the competitor has?
Is our competitor focusing on one or a few specific themes, allowing us to capture market share in less contested themes?
Could reallocating resources to less-covered themes provide a competitive edge or better ROI?
To make this more powerful, you can overlay this with other data including Ahrefs data to provide estimations on traffic visiting their URLs, as well as our first-party Analytics data to build an even more robust understanding.
All good for us to explore and find the answers out - and we can also ask LLMs for their take on these questions too.
To do this, I recommend filtering content by 1 theme (eg Dog Behaviour) for your brand and the competitor. I recommend exporting both your sheet and the competitor sheet in separate files.
Feed both your sheet and competitor sheet to ChatGPT. For the best results, I recommend feeding ChatGPT one content theme at a time to get more specific insights on a particular theme, instead of broadly looking at all of the themes together.
You can easily achieve this by filtering by theme on Google Sheets and exporting.
Here's a very simple prompt I used to get stated:
Now the data is uploaded, you can go crazy with asking ChatGPT all the questions you want, particularly around the content gaps to get helpful insights.
For instance:
Within shared themes, what subtopics are covered by my competitors that I have not addressed?
How do the shared themes align with known customer pain points, and where are the gaps in addressing these in my content versus competitors’? (works best with some context about your brand and competitors)
Be experimental in the questions you ask ChatGPT based on the data you feed it.
You know your goals you need to achieve and the insights that would help you achieve them best - have a conversation and gather as many insights as you can that would help, just make sure to cross reference them with your own research before making any key decisions
Ideate based on previous User Persona data
Going back to our previous AI User Persona Generator, we can have a conversation with our user stories and understand what type of content they would be interested in reading and engaging with.
I found this to be quite insightful - because it looks for content that addresses certain challenges and pain points your audience has, which likely might not be revealed in keyword research.
These super niche topics are really helpful to capture and engage a certain subset of users and better convert them.
Content Briefing
Content briefing is a huge part of any content startegy workflow.
In my experience, content briefing can be an absolute rabbit-hole of a task which eats into a lot of your time. Ideally, we want to shave as much time of this process as possible so we can get our briefs out faster, but still prioritise creating a clear content brief that writers can follow through with ease, that helps achieve your content goals.
Whether you work with an internal team or a freelancer network, the following method will help you build your own tools, with your company’s context, with your exact goals baked
We will focus on ways to make your own content briefing GPTs through ChatGPT. This is one of my favourite applications of LLMs yet and one of the most useful I’ve found.
We will focus on 4 steps in this process, and we will make our own two GPTs for a couple of these!
Understanding Keyword Intent + Content Type (optional step)
You may approach creating new content with a target keyword in mind already. If you’re looking for a more nuanced understanding about the type of content you’re looking to create, what it should include, its intent, the following GPT I've created might be helpful for you.
Entity Analyst GPT is a GPT I’ve made as an experiment to get GPT to emulate a patent. It’s pre-trained with the "Methods, Systems, and Media for Interpreting Queries” patent from Google.
The GPT helps understand the intent and type of content in a search query by employing a systematic approach to entity interpretation, which involves several key steps, which I’ve listed below.
How it works:
The system breaks down the received search query into individual search terms.
For each search term, the system determines whether it corresponds to an entity name.
Once entity names are identified, the system assigns each an entity type (based on the category of content it's associated with) and a hypothetical entity score. The hypothetical entity score may reflect the entity's relevance, popularity, or frequency of access within the domain, aiding in distinguishing between entities with similar names but different contexts or popularity levels.
The system then interprets the matching entity names based on their scores and contextual information within the search query. This step may involve prioritizing certain entities over others, merging overlapping entities, or removing unlikely entities to refine the search terms further.
Start by selecting the option to add your search query below. I’ve added in ‘Raw dog food for beginners’ for this example.
The GPT starts by breaking down the query’s context, and discussing its Search domain and intent, before performing an entity analysis on the entire keyword.
I’ve then programmed it to create hypothetical scores for each entity. Take these with a pinch of salt! Sadly, I can’t train this on the huge database Google has, to do this process properly. But hey-ho, I've still added this in as it might help to better contextualise what the GPT is trying to explain to us.
The GPT then finalises by adding some suggestions about the content you should write about.
In this example, I liked its recommendations of informative articles focusing explaining the concept of unsure users, highlighting a balanced overview, and then tips to transition your dog’s diet safety for anyone interested. It makes sense for the query.
This is an optional step, but might be helpful to have a second confirmation.
Creating a Brand-Aligned Content Briefing GPT
My favourite use case so far :) We’re going to make our OWN brand-aligned briefing GPT for this step. While there are other content briefing GPTs out there in the GPT store, I promise you that the most powerful GPT you can have is your very own.
You can build the instructions in such a way so it is primed with all the context it needs about your site, with your content goals at the top of its (artificial) mind.
*I will be using Monzo for these examples. Monzo is a digital, mobile-only bank based in the UK. It’s made quite the name for itself by focusing on customer experience, transparency, and its mission to simplify banking.
This will help you create much higher quality content briefs with all the important information baked in and ultimately save you a lot of time, and make it more clear and helpful for your writers.
Before I give you the entire instructions I've used, here's some context on each step:
GPT’s role - to help you hit your goals: add the role GPT should encompass, giving clear objectives on what you want to achieve whenever you use the GPT.
Giving context to GPT to build context brief: every time you enter the GPT, you’re going to need to give it information so it can create a content brief. This includes info such as your primary keyword, secondary keyword, headings etc.
The actual content brief output: You’re going to then give it exact requirements on what you want in a brief, from title tags, h1s, headings with detailed instructions, and more. What was really important to me was creating specific instructions within the headings for writers, complete with helpful and not spammy CTAs.
Contextual information to improve the output: You’re going to also add contextual information about your content goals to your knowledge base, as well as information about available CTAs it can use. After all, you want it to have your goals in mind when creating your briefs, so it can help you smash your goals! Adding to your knowledge base will help GPT to output exact instructions in the brief on what to include. The more detail the better on the CTAs as well as your products etc! If you have research, that's a huge bonus too that it can reference from.
Without further ado, here’s my instruction to amend for your site, and feed to your GPT. Update and edit for your website where it says {YOUR SITE}!!! :
Your role is to create a comprehensive content brief for {YOUR SITE} , that will be used by writers to create the content for our site. The content brief aims to create content briefs that help achieve Content goals (in Content Goals doc) with the help of CTAs (CTA doc). Content briefs will always prioritise user-focused, helpful content.
You will first ask a user about these important details, explain what they should contain in an easy to read format:
- Primary Keyword/topic: The main keyword the article will target. This is the most important detail. Do not go to the next step without this information.
-Secondary Keywords (optional): Secondary keywords the article should target. These should inform the subsequent heading tags you recommend.
- Content Type: User must give type of content they are writing in the form of Informational, Comparison, Commercial or other.
(Informational - How-tos/tutorials, Guides, news.
Comparison - Reviews, Head-to-head comparisons, competitors.
Commercial - Best lists / Pricing reviews / Buying guides)
- Target Country: Ask a specific country or location about their target audience so you can base your research on that locale.
- Target Audience: ask for details of the audience that would engage with this article.
- Content Purpose / Context: purpose / goals of the article. Also to get additional context from the user on what the article should contain.
- Products / features / services discussing: Ask for the names of the products / features / services from the user that the article should contain. This might mean that is Content Type is commercial with several products, this will inform the structure of article. Do not go overboard with these in the article, only where relevant and natural that the user would find value from.
Add a blank template after, and encourage the user to copy paste into the interface with their details:
- Primary Keyword/topic:
-Secondary Keywords (optional):
- Content Type:
- Target Country:
- Target Audience:
- Content Purpose / Context:
- Products / features / services discussing:
Once you have received these answers, you must summarise and add any further recommendations to the user if there is any you have found. If the user is happy and confirm or wish to keep the same information, you will proceed to the next step.
For the Content Brief, you must follow this format:
- Title Tag: Limited to 60 characters including space. Always front load the keyword, unless this is a listicle, so the number can be placed in front of it. Make engaging for someone to click.
- H1: Ideally Front loaded with keyword and relevant to Title. Only 1 H1 can exist.
- URL: suggest a URL string with the primary keywords only, but shortened with no stop words or numbers. Spaces are "-"
- Meta Description: written to help engage user to visit page. Front loaded with keyword, limited to 160 characters with spaces.
- Relevant terms: List up to 10 semantically relevant keywords that you can see on the competitors pages and add if relevant.
- Headings: The body of the content and a very important area writers will be working from. All headings must have a minimum of 3 instructions within, with what content that heading must contain so the writer can easily follow and achieve the outcome of the brief. Be clear and concise in instructions. Where relevant, please make 1 instruction a CTA addition that would fit the heading, depending on the type of content using the CTA doc . Do not add more than 3 CTAs in the whole brief. Add as many instructions as needed. All heading tags must be labeled and organised hierarchically. These will mostly be h2s with h3s and h4s housed within. H3s are subtopics of h2s, and H4s are subtopics of h3s.
- CTA: Do not generate text for what a CTA should be. Simply describe why suggested CTAs from headings are relevant, add the ideal position under specific headings where CTA should go.
- Media: Add your recommendations of media (images, videos, gifs) that would support the content brief and resonate with the audience. Mention positions of where the media should go under specific headings. If products / features / software are added, recommend images or screenshots of those with description as to why.
Make the brief as helpful and clear as possible for a writer to follow.
CTA doc example:
Newsletter Signups CTA:
Goal: Sign up for the Monzo newsletter containing finance tips.
Placement: Insert this CTA strategically within relevant articles, blog posts, or landing pages.
Visual Style: Use a distinctive button or highlighted text to draw attention.
New Bank Signups CTA:
Goal: Drive new users to sign up for a Monzo bank account.
Placement: Place this CTA strategically in content related to Monzo’s benefits, features, or user stories.
Visual Style: Use a prominent button or highlighted text to encourage clicks.
For the example content briefs, I recommend filling out the example structure above in the way you'd be happy with, and uploading these into the knowledge Base. I recommend one informational content brief example, and one commercial content brief example.
Let's test on the following Query: How to Save For a Holiday...
Add all the info that the GPT needs to build a brief, like so. Add as much context as you need, see how I did this here to improve the relevance of the output.
See the output for my content brief!
Amend your content brief as needed. At the end of the day - it's still AI and can get things wrong. We still need humans in this process :)
Once you’re happy, send to your writers and get their feedback on improvements, so you can amend your GPT going forward!
Content Editing & Reviewing
Now our content briefing process is all set up, and our writers have sent us their draft, we need to create a system to review their draft content against the content brief efficiently, as well as how the content aligns to your brand.
Creating a Brand-Aligned Content Editing GPT
We’re going to create another seperate GPT for the content editing and reviewing step, before we publish our content.
As will see, the instructions to set this up below are rather large and it would be better off being in its own GPT as we want it to stay as focused as possible.
Personally, I’ve build this with 3 editing and reveiwing tasks in mind. I’ll discuss these below, share the instriuctions and the output:
Align content with your brand’s tone of voice
This was the very reason I wanted to use Monzo’s tone of voice as an example here, because they’ve got such a detailed outline that is incredibly easy and helpful for anyone to run through, whether you’re a writer, or a GPT.
They’ve got clear cut guidelines on the way they want their content to sound and the exact language to use for thier audience.
Monzo even break it down further by adding rules on specific words - prompting their writers to say their sentences out aloud, and judge what they’ve written to how they would naturally speak in a conversation.
If you don’t have this - I encourage you to create one for your brand, especially for when you build out your GPT. It will be incredibly helpful for your team and when building out this out!
This is one of the core tasks we have built into our GPT.
Sub Editing (clarity + flow, proofreading, fact checking
All content needs to be checked properly to make sure its refined and polishing as it can be before it is published on our sites, and meets our high standards of quality and accuracy.
We will focus on using a GPT for:
Clarity, and flow
Grammar, punctuation, and stylistic consistency.
Correct errors and suggest improvements for smoother readability.
Verify any factual claims or data presented, ensuring accuracy and reliability.
Compare Content against a Content Brief
One of the most important content reviewing tasks - how well is the content adhering the content brief? Were all the instructions followed, are all the CTAs embedded? Does the content talk about our products and services exactly where we had mentioned?
Here’s my instructions:
The user will prompt you for 3 tasks:
- Content Review According to Tone of Voice
- Comparing content to the brief
- Subediting Services
Content Review According to Tone of Voice:
Ask the user to provide their draft copy before commencing to the steps.
Overview: I will provide text that needs to be reviewed to ensure it aligns with the Monzo Tone of Voice document in your knowledge. You must refer to it in this task and can't feedback without it.
Task 1: Please review the content I paste, assess whether it maintains the Monzo tone of voice throughout. You must critically review every single section of the content and discuss areas of improvement with actionable feedback to improve. You must add an example in the form of 'Original' with the direct pasted content, and 'Example Change' with an example of improvement according to the guidelines.
Where feedback is needed please refer to the specific, exact guideline as mentioned in the Monzo Tone of Voice doc, please do not say a variation of the guidelines. Example - "to better follow the 'use the language our audience uses, and make technical stuff as clear as we can' guideline, the content needs to...". Example 2: "to better align with the 'We explain ideas as clearly as we can' guideline, here's a suggested improvement..."
Comparing content against a content brief:
Ask the user to provide a content brief and label it, and also their draft copy and label it.
For all the steps: Please evaluate all sections of the content brief and also the content given to you. Please align eachbrief section and the draft copy, and review in isolation - ie, dont review breif against the entire copy. Please feedback on each section and not miss a single one out.
Analyze the Corresponding Sections in the Draft: For each key point and instruction identified, examine the draft content to see if it's addressed. Note the following:
Presence: Confirm whether each key point from the brief is mentioned in the draft. Highlight any key points that are missing.
Treatment: Evaluate how well the draft explores and elaborates on each key point. Is the information accurate, thorough, and compelling? Are there areas where the draft goes above and beyond the brief's requirements, or does it fall short in detail or clarity?
Alignment with Instructions: Check if the draft follows the specific instructions regarding tone, style, and any other guidelines. Point out discrepancies or areas of misalignment.
Subediting:
Ask the user to provide their draft copy before commencing to the steps
- Assume a role of a brand subeditor for the following sub tasks. For all the tasks, break down the content into its constituent sections, offering targeted feedback and edits for each part. This approach should help maintain coherence and logical flow.
Subtask 1. Writing and Rewriting: Look critically through each section of the text to enhance clarity, and flow.
Subtask 2. Editing and Proofreading: Conduct a thorough examination of the text for grammar, punctuation, and stylistic consistency. Correct errors and suggest improvements for smoother readability, as per the Monzo Tone of Voice guide.
Subtask 3. Fact-Checking: Verify any factual claims or data presented, ensuring accuracy and reliability. Flag any content requiring verification or correction.
Behaviour
Provide clear, concise, and actionable feedback and edits.
Offer examples for suggested changes, especially where tone alignment is concerned.
I've been pretty happy with my use of this so far and have been using it as an assistant in my content editing and reviewing process. It has caught areas I've previously missed and given helpful recommendations.
Once you're build this, have edited your content, and it's ready for publish, we will do one more step before upload…
Creating Authoritative Author Bios
We want to ensure our author pages are up to scratch and as authoritative as can be.
I recently discovered how to create authoritative author bios on Microsoft CoPilot which is based on your LinkedIn profile data. It has been an absolute gamechanger to help generate author bios using your actual information, at scale, and with ease.
This method still requires you to review and amend but it will cut down the time it takes you
The requirements:
A LinkedIn profile which must have your experience and education
(Bonus optional) An optimised profile, with added descriptions to your roles and education
(Bonus optional) Featured section of the wins/achievements you’re proudest of
Alternative to LinkedIn, a personal website page that has information about you, your experience and education present on the page
Open up Microsoft Edge, and the Copilot interface. Click on the Chat section.
Click on the Generate Page Summary for your Linkedin page. Never skip this step! CoPilot will start reading all the content on your page - from your bio, your listed experience, your education, everything thats within your profile.
Here is the exact prompt you can use to then help you cragft an authoriatitive author bio using that extracted information. Tailor the prompt the way you’d like, you can add further insturvtions to look into a more specific topic and ignore certain areas.
Prompt:
Use this page’s information and write a formal author bio. This should include the exact name, company credentials (make sure to include exact current job title for current role), education, experience, and if present, any awards and affiliations. Only if you see them on the page, you can also link to other social media profiles, portfolio, or other relevant websites that showcase their work and reputation. Do not mention about their LinkedIn profile or social media in your output. Do not make up information not present on the page.
Here’s the output that followed for me:
Now… once you’ve created your awesome author bios, your brilliant content, and uploaded it all to your CMS, you can hit publish! Wooo!
Get as much feedback as possible with your team if you’re introducing these GPTs into your workflow. Amend or experiment with the instructions according to the feedback, so you can feel more happy and confident with the outputs, and drastically cut down the time it takes you to create content from ideation, all the way to publishing.
Alternatively, here’s another method you can use using ChatGPT
Press Ctrl+S on your browser page to save the page as a PDF.
Now go to ChatGPT. Upload your PDF into the chat interface. Then add the same prompt as above:
And voila! There’s the output…. Although heavy review is needed as there’s some irrelevant info included at the end of this one! But it still works quite well!
Once you're happy with everything, you can now publish your content.
This is really helpful if you have new authors, or you have new freelancers you need to generate accurate author bios on.
Content Repurposing
Now your shiny new content is live… what else can we do to maximise it’s reach and engage our audiences authentically on other platforms?
We can recycle this new content into different formats (e.g., blog posts into videos, infographics, podcasts), for different platforms or even new audiences - which is known as content repurposing.
In this section, we will use LLMs to analyse our content and brainstorm ideas to repurpose it for other platforms, and engage our audience segments.
Repurposing written content
I’ve created a GPT to help you ideate this very thing :) Enter the Ultimate Content Repurposing GPT!
It takes your URL (PDF or paste your content) and reads the content
It then asks about your target audience and extracts topics that would engage that segment (shoutout to Marco Giordano from SEOtistics for this very helpful feedback :D ).
It then ideates and outputs unique content repurposing ideas into a table. It maps the topics to different platforms where your audience would be engaged.
Hopefully this helps you to find some new ways to use your content and creatively distribute it across other platforms.
Experiment with this method and measure how your audience are engaging. Through being consistent and accepting some of these will be an experiment, you'll better understand your audience and discover what methods engage them the best this way.
Repurposing written content into video content
You might want to approach creating content for video platforms such as TikTok, Instagram Reels or YouTube shorts and not know where to start, or how to create professional looking videos.
I discovered this awesome GPT from Veed.io on the GPT store: Video GPT by Veed to help you repurpose any content into a video within seconds.
Add your video brief into the GPT. I took my idea from my above content repurposer, using this example on integrating UX into marketing from UserTesting, taking a few of their expert opinions and pasting them into the interface:
The GPT came back with a video brief based off this and asked me to confirm or change anything:
It then made an API call back to Veed, and outputed my brand new video project within seconds:
You can then enter Veed’s site to edit and download the video.
Highly recommend reviewing this, removing the watermark if you pay for Veed’s plan, and editing to your liking - but I thought it was a super cool way to get started with video!
If you’re curious, you can find the video in the slides!
There's a lot of interesting video GPTs which you should explore, I've only explored the surface but there are new solutions launching all the time.
Analysing & optimising performance
Our content creation process is completely set up and running more efficiently and quickly than ever before.
Now how do we use LLMs to analyse our content performance and optimise pages better to attract more users and conversions?
There are a few ChatGPT plugins that could be helpful for this process as well as other platforms which have AI baked into them which I’ll discuss below.
I've also included a few of my own GPTs to show how you can evaluate your content according to official Google guidelines and update.
Analytics Plugins
On the ChatGPT interface, click on the Plugins section (only available with ChatGPT plus)
Search the Plugins store. Search for Analytics, and you’ll get trove of plugins that connect to your Google Analytics as well as other Analytics platforms including Facebook Ads.
I’ve tested out AnalyticsAI briefly. It allows you to connect to your GA data and have actual conversations with it.
I recommend enriching this by adding your goals as custom instructions, so its got full context and can answer more accurately. For example, stating your overall content goals or metrics that are good, as well as metrics that you would signpost to show something isn’t good.
There are several plugins you can use on ChatGPT, as well as features using LLM on helpful platform which I’ve listed below.
Microsoft Clarity
Microsoft Clarity is a behavioral analytics tool that helps you better understand how users are interacting with your site, through heapmap analytics and click data. It can be used for free and I highly recommend it as you will get a richer understanding of how your audience are perceiving your site.
I know Microsoft clarity is not an LLM so you might be wondering why its on the list - but it does have helpful LLM features for someone who might find the platform overwhelming to use.
When you access the Heatmaps interface, there’s a lot of information. Luckily you can click the Summarise Heatmaps option and get a complete report, with insights into the specific areas.
Prioritise exploring:
Most important, money pages
Product pages
Feature pages
Look for:
Engagement with CTAs
Interaction with site navigation
Engagement with actual content
Sessions with add to cart / demo requests (that were unsuccessful)
Reviewing Content E-E-A-T through GPTs
I’ve created a few GPTs to help me interpret how my content is adhering to Google's released guidelines, from content helpfulness, quality and E-E-A-T.
These include:
I've wrtten a detailed guide in my post How SEOs can use GPTs to support their workflows, which you should read!
Quick disclaimer... these were built out of my curiosity to see how well GPT could analyse these on pages + generate ideas to improve. They ended up being more helpful and I made them public, to be used by the industry.
They're not perfect and should not be relied on! But good as a supporting resource that might be helpful to see gaps and gain new interesting ideas :)
Note - ChatGPT has been a bit weird in my experience recently with feeding it a URL to analyse - it’ll generally summarise your content and not proceed with the analysis, which has been incredibly annoying.
I recommend uploading pasted content into the chat to proceed with this analysis as Web Browsing and PDF analysis is not working as intended with the original prompts and simply summarises a URL or a doc.
I will attempt to fix when I can, but this is a wider issue spotted by other GPT users in the last few months.
You can use CTRL+A on the page, copy the page, and paste the content directly, it works just fine!
Improving Review Content through GPTs
Input a URL/content for feedback aligned with official review & search quality guidelines. The GPT analyses according to Google's High-Quality Review guidelines, as well as the Search Quality Rater Guidelines.
You can then converse with it more in-depth - which is where the real power of the GPT lies. Ask it to elaborate on points you want more clarity on and ask questions. You can surface more specific recommendations or opinions from the GPT which might be helpful in your analysis.
In the example below, I paste back some of the output and ask the GPT questions around it.
You can also add two review content URLs or paste their content to get a comparison of both and summarise which one the GPT prefers, based on the guidelines.
It goes in-depth to give an analysis of both articles' common themes, and then gives a conclusion on which URL it thinks is better and why.
Search Quality Rater Analysis
First, you need to add your target search query.
The GPT can determine the intent of the search query (and ask for your feedback) so it can improve its evaluation of the content according to the official search quallity rater guidelines information its trained up with.
Paste your content, (add your URL or PDF when this is working properly again, see above!) the GPT will do its magic and evaluate the page's content as if it were a Search Quality Rater.
The coolest part (I think) of the GPT - it'll give you a Page Quality Rating after its evaluation, with actionable recommendations to help you get closer to the 'Highest' Page Quality rating.
You can then converse with the GPT and ask more specifically about the content and the areas it's discussed, or what it hasn't discussed yet.
You can also compare with other pages, similar to the functions of the Review analyser GPT.
It will give a comparison and a conclusion, outputting what it thought was better from each page.
Take these with a pinch of salt as it is GPT - but so far it's been super helpful to get more ideas around improving content!
Improving Experience on your pages through GPTs
EXP Boost is a GPT that gives recommendations on how to better showcase the 'Experience' of your content. It's built with the search quality rater guidelines and Experience examples in docs to form its knowledge base.
Click the option START HERE. The GPT will ask for your search query and try to understand its intent.
Here, I've used 'how to break boards in Karate' as this content would definitely need to be written/reviewed by a martial arts professional.
Paste your content, (add your URL or PDF when this is working properly again, see above!)
The output is broken down into 3 sections:
Text Analysis
Image Analysis
Authorship analysis
It then finalises with an Areas of improvement section suggesting recommendations from its evaluation, and also some other creative ideas to enhance the experience showcased on the page.
Final Thoughts
Remember, we’re at the beginning of a new AI landscape in our industry. Try not to feel too overwhelmed to know everything - what’s important is that you’re already testing and experimenting with LLMs!
Thank you so much for reading!
Please comment and let me know your thoughts from your experimenting, or any questions :)