You and I keep reading about how AI is going to change the world — or not, depending on NVIDIA’s earnings that week — but that companies are struggling to implement AI in their workflows.
I wrote a couple of weeks ago that the biggest winners of the genAI revolution thus far have not been OpenAI, but consultants Accenture who have booked slightly more genAI revenue than the ubiquitous ChatGPT maker.
We have used AI in our processes for years. I first started with GPT-NEO whilst on the GPT-3 waitlist. Our philosophy has always been to use AI to help us do better work. It’s not about productivity but quality.
Today I want to share 5 new AI internal tools we’ve built in the last 5 weeks. Where finding AI use cases is difficult, we have no end of use cases in production. From identifying missing topics and assessing search intent fit, to rapidly briefing the team on new subjects and suggesting optimised graphics, these tools showcase the breadth and depth of what’s possible today with cutting-edge models.
Identifying content gaps at scale with RORY AI
RORY AI is our new research product that can identify not just what competitors are covering, but the importance of those topics to the searcher and against the search intent.
RORY is designed to analyse massive amounts of SERP data to uncover not just what your competitors are writing about but, more importantly, what they’re not writing about. It looks for gaps in the content landscape that represent untapped opportunities.
Then, RORY goes beyond just identifying topics. Its superpower is its ability to assess the importance and relevance of each potential topic for your target audience.
You need to know the importance of a topic – not just the frequency – because then you can cover topics because they’re useful, rather than because everyone else is covering them.
We’ve been using RORY for a while now, but we recently added the ability to input an existing URL and get an instant report on the most important topics that page is missing. With just a few clicks, you can see exactly how to take a piece of content from good to great. We implemented the recommendations for a client and saw 100%+ improvement in clicks within 7 days. The results are awesome.
I was in a meeting looking at some older content last week, and it was remarkable how much of a difference RORY has made versus our old process (which used mainstream and very expensive SaaS software to identify topics but not importance).
RORY-powered content gets the point a lot faster, cuts fluff, and is a lot more useful to the searcher.
Ensuring content relevance with the “product/search intent fit” analysis
One of the most common challenges in SEO is determining not just whether you can rank for a keyword, but whether you should. Just because you can optimise a page to rank well doesn’t mean that content will actually resonate with searchers and drive meaningful conversions.
It might be possible to achieve a ranking for a target keyword, but if this gets you traffic and no conversions, then this might not align with your priorities.
We’ve solved for this by introducing the “product/search intent fit”. You’ve heard of product/market fit: this is adapting the same idea for organic search.
We understand what searchers are looking for and how well the client product fits with this. We can identify the benefits and the limitations and then make a call on the strength of the fit.
This then gets rolled up with the FALCON prediction, SERP information, and other analysis into a programmatic strategic call: is this the best fit for you?
Here’s an example:
The search intent for “how to know which RAM is compatible with my laptop” reveals a highly specific and technical query that may not align well with the client’s broader product offerings. While the topic relates to computer components, which is within the client’s expertise, the primary user need is for detailed, step-by-step guidance on identifying RAM compatibility. This presents a challenge in seamlessly integrating the client’s products without appearing overly promotional or deviating from the user’s core information-seeking goal.
This analysis continues, and the overall recommendation was to not proceed. This goes to a human expert who can decide what they want to do with this analysis.
No more wasted time chasing vanity keywords or creating content that drives empty traffic! With the new score, we can ensure every piece of content we create is laser-targeted to both the needs of searchers and the priorities of the business.
Rapid topic briefing
One of the challenges of a best-in-class content agency is you have lots of different topic areas. We ask 5–7 people to become experts on any given topic throughout the process, and they might need to be an expert on WooCommerce at 9am but an expert on signage for manufacturing businesses at 11am.
This context switching is hard and it takes some time to get up to speed. We solve for this with the Rapid Topic Brief, a 1-pager that gets you up to speed quickly on any given topic.
The Rapid Topic Brief uses a lot of the same technology as RORY: it goes out and finds all the information it can on a given topic. We’re taking advantage of the huge context windows in Claude 3.5 and Gemini 1.5 Pro that let you feed in hundreds of pages of context.
The Rapid Topic Brief then condenses the information down into a nice 1-pager. Text is cool, but what’s even better is then writing a podcast transcript and using text-to-speech to generate a 5-minute audio briefing on the topic. It’s surprisingly good!
Graphic suggestions
Jamie is our wonderful Graphic Designer, and Jamie powers through 80–100 graphic requests for our content each month.
I’ve long struggled with how we can use AI here: AI image generation is getting better, but it’s still obviously flawed, and whilst text generation gives you something a human expert can build on, image generation is just trying to do the work for you.
Jamie and I went through his process and identified an opportunity: thinking of what your graphic should be is part of the creativity of the work, but at 100 graphics a month, it’s also a lot. We have thus built a solution: AI-powered ideas for graphics.
The solution is very cool: we use the newest computer vision AI to understand the client style in the last 4 graphics Jamie produced for them. We then collate this into a client style and generate 5 possible graphic ideas for each post.
Because we’ve analyzed the most recent graphics for the client, the suggestions are in their current style.
The ideas range from very sensible to very out there. The idea is just to give Jamie some ideation to work with. This fits our philosophy perfectly: use technology to help us do better work than ever. A really nice quality of life improvement.
Cannibalisation analysis
Cannibalisation is the remarkably alarmist term for where 1 article “eats” the performance of another. I have no idea why such aggressive terminology is used, but here we are.
Cannibalisation can be a huge issue. You generally don’t want 2 articles competing with each other for the same clicks.
To date, we’ve used data from the Google Search Console API to flag if any content from the same domain is ranking for the target keyword. This runs in one of James’s giant sheets. It works OK!
I’ve moved the identification out of James’s giant spreadsheet. The GSC API is bizarrely inaccurate on keyword reporting, so we’ve swapped out GSC as the source of truth for a better keyword API.
Analysing cannibalisation programmatically is wonderful. Claude 3.5 is superb at taking in a huge amount of context and creating a report for you. We feed in the SERP, the full HTML of the articles in the SERP, the full HTML of the 2 articles with possible cannibalisation, and prompt against a series of questions to get insights. The new caching from Anthropic makes this wildly more affordable than it was even 6 weeks ago: about $1 to check everything rather than $10. Given the value that creates, it’s hands-down worth it.
Having Claude read the HTML is a great example of where a computer can do something a human is going to find difficult. You need the HTML rather than the text, as we need to understand linking. A human is going to have a terrible time reading the HTML of 20+ pages, but to AI, it’s no problem.
Search intent scoring
The final piece is a really nice quality-of-life improvement that makes it easier and faster to understand the opportunities in a SERP.
We’ve added scoring 1–10 of how well each URL meets the search intent.
This lets you see how well content in the SERP meets the intent on average, but also URL-by-URL. You can quickly understand: content in the SERP is really competitive! Or, content in this SERP is not competitive!
When you filter the URLs by type, you get even more insight: blog content on this SERP is really good at meeting the search intent, but we also have these short category pages, and that’s an opportunity.
This is output at the keyword stage and refreshed when we start working on a piece of content. It’s extremely useful and a great addition to the mix.
Businesses need smart and useful implementations of AI – now
The 5 AI tools we’ve developed in 5 weeks at Ellipsis are a testament to the transformative power of AI in the SEO industry. From identifying missing topics and assessing product/search intent fit to rapidly briefing writers on new topics and suggesting graphic ideas, these tools showcase the breadth of applications for AI in content optimisation.
The quantity and quality of data you need to deal with to do SEO well in 2024 is high and getting higher. AI is not just a productivity booster or a novelty – it’s a strategic asset that can help us deliver better quality, more targeted, and more effective content.
I am committed to our core philosophy: using technology to help us do better work than ever before. The AI revolution is here, and we’re excited to be at the forefront, harnessing its power to deliver exceptional results for our clients.
