Validate your long-form content with the help of AI assistants

Mark says: “Use AI assistants to validate your long-form content at scale and see if it’s answering your users’ intent. The kind of content can include any pages with lots of content on them – even if they're e-commerce category pages.

Generally, companies that have invested in producing content for many years have focused on content quality and understanding what users are searching for. Essentially, this process is for anywhere where you're trying to match complex user intent, which now includes pretty much any business that has a website.”

Are you using the validation process to match user intent?

“Yes. If you take on a new client, you’ll often do a content audit to try and work out if the content they already have is good.

You can find loads of information about expertise, authority, trust, experience, and knowing what we need to look for in content. You will also have reams of keyword research from all kinds of tools that give us hints as to what users are searching for – and you've got a wealth of experience within the company of what they know about their customers and what they regularly ask.

It's great having all that information but trying to work out if the content that you’ve got for a particular subject answers all of those questions is a very long, manual task – or it has been. One of the best uses of LLMs now (which leans into their strengths, especially now that they can access live URLs), is using them to feed in that research.

If you put the article title into AlsoAsked, you'll get a few hundred People Also Asked questions out, and you can load those into something like ChatGPT. You can then say, ‘Does this URL answer those questions?’, and you can very quickly get an idea of where you need to improve that content. It would have taken a very long time to manually go through and pick apart.

You can do this with all different types of keyword research, whether it’s People Also Asked data, suggest data, or even content briefs that you're building when you're looking to improve the content. With ChatGPT, there are also opportunities to use things like their API to scale it. If you've got 1,000 or 5,000 pages you need to check, it’s possible to make very quick strides into that.”

Is ChatGPT your preferred AI tool?

“At the moment, for this particular task, yes. To be honest, I've been very cautious about the use of AI. ChatGPT is built on GPT-4 which is a large language model. The generative part of these algorithms can lead to what's commonly called hallucinations. How they work is they take tokens and make statistical guesses about what comes next.

I've seen people try and use LLMs for things like analysing SEO data and forming strategies, which I don't think they're really capable of doing yet. You'll certainly get an answer that sounds convincing a lot of the time because that's what LLMs do. They give you what sounds like the most probable answer, so they're great at fooling humans.

This particular language analysis task is something I've found, through testing, that LLMs can do with a high level of accuracy. It’s useful, and it saves a lot of time and money.”

Could you go into more detail about how the validation process works?

“Let’s take People Also Ask research as an example. We'll have a public URL which has our keyword research data in. Then, if it's an optimized title, you can just drop the title of the page into Google or AlsoAsked to get a list of nearest intent questions around that topic. You can also amplify that with other things that you know manually or even other suggestions from something like ChatGPT.

Once you have that list of questions in an accessible document online, you can then build a simple prompt by posting your target URL and saying, ‘Does this URL answer all of the questions posed on this other URL?’ You can also tweak that prompt to get the answers in a table that specifies where it thinks the answer is, highlights any gaps, and maybe even suggests fixes. I've seen that executed really well.

To push that slightly further, you can use the Search Quality Rating guidelines PDF and ask the LLM to evaluate the content against that. It does it well at a basic level but, when it comes down to things like page experience, it gets a little bit confused. It might say that there isn't anything about page experience on the page because it's looking for the content rather than the user experience. You have to be careful where you draw the line.

However, it’s really useful for that initial task of going through 2000 pages of existing content for a new client, using some quantitative data about that content from analytics (how many visitors it's getting, how well it ranks, what the engagement metrics are, etc.). At this stage, you've got no idea what the coverage is, and this process can really quickly give you an action plan to start moving things in the right direction.”

Would you create a Google Sheet with all those URLs and ask ChatGPT to go through them and see what the content is and what’s missing?

“Absolutely. Normally, there will be a pattern on the website for the type of URLs you want to use. You can use a tool like Sitebulb or Screaming Frog to get a list of those URL patterns.

We tend to prioritise them. You can do as many as you like but, when it comes to updating content, there's no point giving someone a work list with three years’ worth of content for them to update. You should focus on particular areas of the site because it's good to have a control when you’re testing and qualifying the results of what you're doing.

If you've got a website that covers several different categories, you might focus on one particular category and work on improving that content. Then, you can compare that against things you haven't worked on to see the impact you’ve had (number of keywords ranked for, impressions, organic clicks, etc.), and separate that from something like a Google update.”

Would you focus on a category of content first or identify pages that are ranking on page two and aim for quick ranking changes?

“You can't avoid the quick wins; that's always a really good place to start with content. If you have stuff that's not quite in the top few rankings and it’s possible to get there, you absolutely want to start there. That's where you'll see an impact.

However, if you want to scale it, you’ve got to keep track of what you're working on. If you want to work on 50 disparate URLs that fit those criteria, that's fine – as long as you still record them. There are advantages to having a topical approach because, when you look at rankings, Google will generally favour websites for particular types of categories that are inside one topical area. If you did a Venn diagram, there's normally significant overlap between those URLs that are performing well and are topically related.”

Are there any current trends in terms of the type or format of content that might improve an existing page?

“Right now, we’re about one week out from the end of the helpful content update, which has seen some massive shifts in how websites have been ranking. Some sites have lost 70/80% of their traffic from what they claim are good pages, but it’s too early for me to draw conclusions on that latest sprint.

Personally, I wouldn’t focus too much on the pattern of the algorithm or what the algorithm is looking for and instead focus on a higher strategic goal. Do you have a genuine expert answering these questions and are you posing the right questions?

I have seen some patterns in the sites that have been impacted by the latest helpful content update in that they tend to lay out their content too much like an SEO: question-answer, question-answer, question-answer.

The advantage of using an LLM for precisely this kind of task is that, when you ask it whether a specific question is answered within that page or document, the question doesn't need to be explicitly written on the page. If the answer is embedded within a paragraph, the LLM seems to be more than capable of finding that and identifying whether it's been answered.

That's a more natural way of writing but I don't know whether that's a pattern Google's picked up on. It's certainly something that's on my mind when we're giving content briefs to the people who write the content. At Candour, we produce almost no content. We always try and get the correct person to write about it – whether it's an in-house client expert or a freelance writer who has a lot of experience in that topic.”

Would you ever allow ChatGPT to also write the content for you?

“You can use it as a starting point for incredibly generic things that you might need to phrase in a certain way. However, I don't think it's a good long-term bet.

Firstly, being an LLM, if ChatGPT can generate a suitable answer for that question, then that question is already objectively solved somewhere else. You are just re-wording it, so you're not adding any additional value.

If you consider the SGE that Google is working on, their strategy has historically been to try and keep users on the SERP – because that's how they can best monetise them. If you, as an SEO or a company, have access to an LLM that can generate an answer to a question, why would Google choose to rank your content when they could generate that answer themselves?

You have to be very careful about where you use AI-generated content. You may not be worried about whether some content ranks and it might still be helpful to users. That’s where you can use it. However, for things that you’re trying to rank for, I don't think that's a good long-term strategy. You are essentially getting in the way of Google making additional revenue which, in my experience, isn't a good place to be.”

If an SEO is struggling for time what should they stop doing right now so they can spend more time doing what you suggest in 2024?

“Stop analysing, auditing, and disavowing toxic or spammy links. In conversations with the SEOs I speak to regularly, that's just taken as read. However, when I share that point of view in other places, I have quite a number of people telling me that I'm wrong and it's really important to look for spammy links.

You will see AI-collaborated articles on LinkedIn that say things like, ‘toxic links can harm your website and your brand’ and, ‘you need to be looking at spam scores and toxic link scores’. Google has very flatly said (and I don't think there's a reason for them to carefully word this) that they discount rather than penalise those kinds of links.

That makes perfect sense to me because, if they put a negative value on what they class as bad links, there would be a whole economy for doing that to your competitors. We've also just seen Bing depreciate their link disavow tool because they're so confident that they can identify them.

In my book, there are a lot of other things you could do that would provide value. I'd rather have another good link than get rid of 10 of what someone says are bad links.”

Mark Williams-Cook is Director at Candour and Founder of AlsoAsked, and you can find him at AlsoAsked.com.