Earlier this year, a team of us SEOs at Bring packed our bags and headed down to BrightonSEO, ready to soak up all the latest industry insights and maybe even a bit of seaside sun. One thing quickly became apparent: AI was the talk of the town. 

Since its launch in November 2022, ChatGPT has had a monumental effect on marketers’ approaches to various aspects of search engine optimisation and, in particular, content production. So much so that Google has even changed its guidelines around how it rewards—and penalises—AI content. 

(Spoiler: as long as it’s helpful for users, Google doesn’t seem too fussed.)

We already know that AI can be an invaluable tool for creating content at scale. But as time goes on, more marketers are finding new ways to use AI to make their lives easier without compromising on results.

That’s why we decided to roll up our sleeves and put four AI prompts that we learnt from BrightonSEO to the test. The mission? To find out if they’re worth adding to our content marketing and SEO toolkit — or if they’re better left on the conference floor.

Why do we need to test prompts for AI content creation?

AI might be the latest shiny tool in the organic search toolbox, but like any tool, it’s only as good as how you use it. With so many new AI prompts and strategies being tossed around, it’s easy to get swept up in the hype. 

But before we dive headfirst into integrating AI into our content marketing or SEO workflows, it’s crucial to take a step back and ask: Is it actually effective?

Consider this: according to Salesforce’s State of Marketing Report, over half of marketers say they use predictive AI and even more cite using generative AI. Yet, despite this optimism, a significant number of them have concerns about embracing AI, with a lack of strategy, inaccurate outputs, and general distrust in generative AI ranking highly amongst their list of concerns.

Testing AI prompts isn’t just about seeing if they work — it’s about understanding how well they work, where they fall short, and whether they can truly deliver value. 

After all, there’s no point in using AI to create content if it causes more headaches than it solves. By putting these prompts through their paces, we’re not just looking for quick wins— we’re making sure that any AI we bring into our workflow genuinely enhances what we do, saves us time, and ultimately drives better results.

Our chosen AI prompts 

In the spirit of curiosity (and a little bit of healthy scepticism), we decided to put four AI prompts from BrightonSEO to the test. 

Here’s our shortlist of contenders:

  1. Using AI to create content clusters and identify content gaps
  2. Getting AI to help us group keywords by user intent
  3. Asking AI what it needs from us for it to create stronger content
  4. Requesting AI to read Google guidelines

Here’s what we found when we put them to the test…

1. Creating content clusters and identifying content gaps

Idea: 

AI keyword research might not come with the same data-driven backing as traditional SEO tools, but it can understand topics and how they relate to each other. With that in mind, we saw potential in using AI to create content clusters and identify any gaps in our existing content strategy.

The prompt we picked up at BrightonSEO was straightforward: feed the AI a sitemap and ask it to group the URLs into content clusters. Then, have it identify any gaps and suggest ideas to fill them. 

Here’s the prompt we started with:

Read this sitemap: https://www.bringdigital.co.uk/post-sitemap.xml

Group the URLs into content clusters. Identify content gaps and give some ideas to fill them

The idea was that AI could give us a visual representation of our topic clusters, showing us what we’re covering well and what we’re missing out on.

We tried it out:

We plugged in Bring Digital’s sitemap and waited for the magic to happen, but, as you can see from the above, our initial attempt didn’t quite go as planned. ChatGPT wasn’t able to directly access the sitemap we provided. Instead, it offered tips on how to review the sitemap manually — a thoughtful gesture but not quite the time-saver we were hoping for.

Refusing to give up, we took a different approach. We scraped the URLs from our sitemap ourselves and fed them into ChatGPT as a list, tweaking the prompt slightly to provide more context: 

 

Here are all the URLs from Bring Digital’s blog. Bring Digital is a digital marketing agency specialising in SEO, PPC, paid social, affiliates, and translation services. 

[list of blog URLs]

Group the URLs into content clusters. Identify content gaps and give some ideas to fill them

 

Chat-GPT’s response:

This proved more fruitful: we ended up with a list of grouped blog topics that were surprisingly more accurate than expected. Sure, there were a few questionable categorisations — some URLs ended up in odd places — but overall, the results were solid. As with anything AI-created, it wasn’t anything a QA and human touch couldn’t fix.


As for the content gaps, the AI did give us some ideas, though they were fairly high-level and vague. The suggestions needed refinement before they could be considered actionable, but they did provide a good starting point for brainstorming new content topics.

Results:

Overall, this prompt had a few niggles but nothing we couldn’t iron out to get some decent nuggets of information for our theoretical blog content GAP analysis. ChatGPT’s ability to quickly group URLs into mostly accurate content clusters could save us a lot of time in a comprehensive blog audit, especially when dealing with a vast amount of content.

The GAP results gave us some top-level ideas to work with, but much like the clustering task, would need a dose of human touch to refine into A+ ideas. 

Overall, this prompt could be a good place to start for a comprehensive blog audit task or a blog for a really niche sector where ideating could be tricky. We’ve knocked off points for accuracy, but we were impressed with how much time the AI could save us if we wanted to cluster a long list of pages.

Score: 4/5 

 

2. AI to understand user intent 

Idea: 

With the first test under our belts, we moved on to round two: using AI to help us better understand user intent. 

This time, we weren’t starting from scratch but rather working with real keyword data and a clear blog topic in mind. 

The idea behind this prompt was to feed ChatGPT a list of keywords that competitor articles were ranking for and let it do the heavy lifting — identifying the intent behind these keywords and weeding out unique opportunities for us to target in our content.

We tried it out:

The instructions began with some manual labour: we grabbed the top three ranking results in the SERPs for our chosen topic (image optimisation for SEO), identified the keywords they’re ranking for, and uploaded these Chat-GPT as an Excel document to identify unique keyword opportunities.

 

After following the prompt’s instructions, we got exactly what we asked for: a list of 145 keywords our competitors are ranking for without duplicates. We could have gotten this ourselves in Excel or Sheets with a simple formula, so initial impressions weren’t great.

Determined to push the AI further, we got a bit more specific. We asked ChatGPT to refine the list, whittling it down to a handful of primary keywords that we could use to optimise our blog post. 

And this is where things started to get interesting — or perhaps frustrating. ChatGPT provided us with five target keywords, but one of them, “how to optimise images for SEO,” wasn’t actually in the original list of competitor keywords we fed to the AI. Therefore, it had no keyword volumes to back it up.

It seemed that ChatGPT had taken some creative liberties, possibly because we mentioned wanting to write about this topic. It even gave us a semi-convincing explanation, claiming the keyword was chosen for its “tutorial aspect.” While that might sound plausible, it left us questioning the AI’s reliability in this task.

Either way, we had five target keywords for our theoretical blog post.

Results:

The amount of work required to extract something useful from the AI felt disproportionate to the results. We couldn’t help but think that conducting the keyword research ourselves might have been more straightforward and less time-consuming.

In a last-ditch effort to salvage the situation, we asked ChatGPT why it selected those particular keywords. 

 

The response, while coherent, didn’t quite convince us. 

We’ll stick to our tried-and-tested methods of conducting competitor keyword research ourselves.

Score: 1/5 

 

3. Ask AI if it has any questions or requirements

Idea: 

We’ve grown used to ‘perfecting’ a prompt or relying on a template to get an end result from Chat-GPT, but what if we stopped to ask the AI if there’s something crucial that we’re missing that’s preventing it from delivering even better results?

This was the idea behind our third prompt from BrightonSEO: after providing the initial prompt, we added a simple follow-up question: is there anything else you need to know to help you do your job? 

The theory was that ChatGPT would request any extra details or information that could help it better understand the task and deliver more tailored, accurate content. Plus, it might help us avoid missing any crucial components in our prompts.

We tried it out:

We decided to test this idea with a straightforward task: writing category page copy for an eCommerce website selling high-end fashion clothing. The specific page we were working on was for ladies’ coats.

Here’s our usual prompt, including target keywords and notes on the brand’s tone of voice,

I’m writing a piece of category page copy for an e-commerce website that sells high-end fashion clothing. The category page is for ladies’ coats, including a mixture of formal and casual pieces in a range of fabrics like suede and leather. 

You will be writing out the category page copy for me, optimising the copy for keywords that I would like to target. The keywords to target are:

  • Ladies’ coats
  • Ladies’ leather coats
  • Ladies’ suede coats
  • Trench coats womens
  • Black ladies’ coats
  • Women’s waterproof coats

The idea is that I would like the page to rank highly in the SERPs for these keywords so that I can boost the visibility of the e-commerce page and drive traffic to the category page. 

The copy should be written in the ecommerce brand’s tone of voice and should follow the same structure as other category pages on the website.

Here are some notes on the ecommerce brand’s tone of voice:

[Brand] does not sell fast fashion; its produce is sourced from artisans and designer brands. The emphasis is on quality over quantity, which should be reflected in the creative — and sometimes poetic — deployment of a diverse vocabulary. That being said, description should never compromise the clarity of what is being communicated.

Do:

  • Use an engaging mixture of simple and complex sentences
  • Employ a variety of adjectives to help bring to life what’s being described in unique ways
  • Include the small details about products that help emphasise their high caliber
  • Deliver facts rather than opinions wherever possible
  • Make good use of short, imperative statements

Don’t:

  • Bury key information in dense paragraphs
  • Use headlines that misrepresent the information contained within the blog post
  • Create content that is longer or shorter than necessary to convey all the information the reader wants to know
  • Use multiple words where one will do
  • Rely heavily on jargon

Here’s an example of an existing piece of category page content:

From delicate printed silk to sumptuous velvet and cool cotton, our collection of ladies’ dresses uses only the finest materials to produce stunning contemporary designs that are full of life and colour. With light and casual creations that celebrate the coming of summer, elegant evening dresses and everything in between, our selection is capable of rising to any occasion. Discover colourful tea dresses embolden with pastel hues and elegant patterns, or more luxurious velvet wrap dresses for women that add a touch of glamour to any outfit. We also offer a range of women’s dresses crafted from 100% natural fibres, such as cashmere, cotton, and silk, for a timeless and reliable style. Shop online today.

Please write the category page copy for the ladies’ coats page.

 

Chat-GPT’s response:

The result was mixed — while the AI did a decent job of matching the tone and style we were after, but the length and structure of the content were wildly different from the example we’d provided.

When we asked Chat-GPT if there was anything else it needed to create the content, here’s what we got:

The AI responded with a request for a few additional details to help refine its output. Once we provided this extra context, Chat-GPT confirmed it had everything needed to write a first draft. Here it is:

 

With this new information, the AI’s second attempt was a marked improvement. The copy was much closer to what we were looking for—about 90% there—with only minor tweaks needed to fine-tune the tone of voice.

Results:

The difference between the two prompts is substantial. By simply asking the AI if it needed more information, we significantly improved the quality of the second output. 

While the process of providing all this extra detail was a bit time-consuming, it was clear that this prompt had potential. We’d recommend scaling this process for a larger batch of content to make it worthwhile. 

If you fed ChatGPT an Excel sheet with all your target keywords and category page USPs, we could see this being an accurate and effective way of creating optimised category page copy.

All things considered, this prompt worked extremely well. Yes, it required some prep work, but the end result was well worth the effort. We decide how to use the prompt to make it worthwhile, so when it comes to scoring this prompt, we give it full marks.

Score: 5/5 

 

4. Optimise content for SEO using Google guidelines

Idea: 

With three prompts down, it was time to dive into something more technical: optimising SEO content using Google’s guidelines after you’ve crafted a piece of blog content yourself (or AI has given you a helping hand).

The idea behind this prompt was simple enough: ask ChatGPT to review our content and grade it based on Google’s quality rater guidelines, spam policies, and helpful content criteria. Ideally, it would then provide us with actionable steps to improve our content according to these guidelines.

We tried it out:

We kicked things off by asking ChatGPT to write a generic guide on how to warm up before going for a run. Once we had our content, we asked it to take into consideration Google’s guidelines for quality raters, spam content, and helpful content. 

This is where we hit our first snag. ChatGPT, as helpful as it usually is, couldn’t actually “read” the Google guidelines directly. Instead, it fell back on its general knowledge of these guidelines to offer us some broad-stroke suggestions for improvement. 

The tips it provided were pretty fundamental — nothing groundbreaking that we wouldn’t have come up with ourselves during a standard content review.

Not entirely satisfied with this result, we decided to tweak our approach. Instead of asking ChatGPT to read the guidelines, we provided it with a breakdown of the key points from Google’s quality rater guidelines, spam policies, and helpful content framework.

We got slightly more detailed recommendations and a wider variety of them, even touching on technical SEO and user experience. This approach yielded a top-level checklist that could be quite useful for a final content optimisation pass, especially if you’re looking to ensure you haven’t overlooked any of Google’s core requirements.

Results:

We found that this prompt worked better when we provided the AI with a primer on Google’s guidelines rather than asking it to rely on its “general knowledge.”

Once we passed that hurdle, the AI offered a list of actionable improvements for our content. While the recommendations weren’t revolutionary, they could certainly serve as a solid checklist to ensure we’re meeting all of Google’s requirements and haven’t forgotten anything throughout the creation or optimisation process.

The prompt isn’t a magic bullet, but it does provide a helpful way to double-check that our content aligns with Google’s expectations. To make this method even more efficient, it could be useful to have a boilerplate prompt about Google’s guidelines that we can give to the AI for each piece of content we want it to review, allowing us to skip the priming step for each new piece of content we review.

Score: 3/5 

 

Our final thoughts

If you’ve ever attended BrightonSEO, we’re sure you’ll know that the sheer volume of information packed into those two days is nothing short of overwhelming. New ideas and strategies often sound revolutionary when you first hear about them — promising to change how you work, save time, and boost results. But as we all know, the real test comes when you try to put these ideas into practice back at the office.

After testing out the four AI prompts we learned at BrightonSEO, we’ve come away with a mixed bag of results. Some of the prompts showed real promise and could potentially save us significant time and effort in our content creation and SEO processes. Others, however, fell a bit flat, proving that not every shiny new tool is worth adding to the toolkit.

The standout winner for us was prompt #3 which asked AI if it had any questions or additional requirements before crafting content. 

This simple tweak turned out to be a game-changer, dramatically improving the quality of the output with just a little extra input from our side. It’s a prompt we can see ourselves using regularly, especially for larger batches of content that need to hit the mark with minimal revisions. Think: optimising lots of commercial pages in one swoop.

Overall, these experiments highlighted the potential of AI in our workflow but also stressed the importance of a critical eye and human touch. Not every AI-generated result is gold, but AI can be a powerful ally in enhancing our content marketing and SEO efforts when used strategically.