Information gain and the new frontier for SEO: content in a ChatGPT world

Ellipsis is a fully remote team, and we’re fortunate to receive a large quantity of job applications when we’re hiring.

We’re a fully remote and expertise-driven company, and ask candidates screening questions. I like to see how candidates write and solve problems.

Recently, when we’ve been hiring we’ve noticed that a third or half of the responses are saying the same thing. The most egregious ones are (hopefully) entirely automated and start “As an AI language model, I can’t…”, but we had dozens and dozens of applications all making identical points.

Have no doubt that ChatGPT is going to change SEO Content forever. But have no doubt that ChatGPT will drive further homogenisation of SEO Content.

I say further homogenisation, as SEO Content has been heading towards extreme same-ness for the last couple of years. Popular software like MarketMuse, Frase, or Clearscope encourages content teams to build their topic coverage on competing content. The idea is that we can look at the topics competing content covers, cover more topics than competing content, and therefore satisfy the search intent best.

This drive for topic coverage completion met a need for ballooning word counts and finding something to say in ever-bigger content pieces. Even a couple of years ago, Kinsta published a post about speeding up WordPress that was so long – 25k words – that they literally made a book out of it.

If you’re trying to target the same keyword, the topic coverage philosophy demands you write a 30k word piece. In this world, ChatGPT is wonderful: we’re in a word count race and it’s a great way of adding word count.

Information gain and the urgent need to say new things

Google filed a patent in 2020 that assesses something it calls information gain. The patent describes a system for judging the amount of new information in a search result, against other results the searcher has already seen.

On a post on the topic, Animalz cover this really nicely. When we have a 25k word piece about speeding up WordPress, does anyone actually want a 30k one?

When every topic, every keyword, every conceivable avenue of growth is already dominated… their content is long and comprehensive. It’s immaculately search optimized, and if we’re honest it’s pretty darn similar to the content we had in mind.

If we start tying these threads together, you can see we have a problem:

  • SEO Content was already going in the direction of homogenisation
  • Topic coverage trends encourage longer and longer content
  • ChatGPT gives everyone the same answers to the same questions
  • The inevitable outcome is that content gets longer, more homogenous, and less useful

If you and the other sites on page 1 are being judged for information gain as a ranking factor and you’ve generated broadly the same ChatGPT content as the other page 1 content, you can see how you might have reasons to be worried your rank might not stay. This is from the patent:

Doing SEO Content in 2023 is going to require doing different things, differently. One of those needs to be introducing new topics and new concepts into content.

You should be really concerned about your ChatGPT content being the same as everyone else’s

Once the thrill of being able to “write” an OK quality 2000 word blog post in 10 minutes wears off, it’s time to be really concerned about your ChatGPT content being the same as everyone else’s.

You see this in the job application responses. I flagged the most egregious “As an AI language model…” job application earlier, but it’s the bit-of-effort-AI ones that are the most harmful. These are 2 responses from different people, and they make remarkably similar points. I could have picked any of 50 applications with a different version of the same thing.

SEO Content is winner-takes-all. ChatGPT has made it easier to create low and medium quality content. But I see this the same way Google does: your content needs to be extremely good to “win”, and that’s not going to happen by using AI by itself.

To win with search, you have to be better than the competition while writing genuinely helpful, relevant content. You won’t be better than the competition if you’re using the same chatbot to generate the same text.

There’s an additional risk which isn’t being highlighted often enough, too: using ChatGPT as your sole basis / outline builder without the right prompts and nuances also arguably restricts your own thinking, blocking you from exhibiting your own (true) expertise. I’m sure many of the applicants we saw could have done a decent job answering our questions with no AI help, but their using ChatGPT as their sole source of information means they didn’t feel the need to think any further about what they were writing. It looked great, so they didn’t need to think about whether there was something else they could/should be saying. This can ironically be a huge restraint!

Quality content in the age of infinite content

In a world where chatbots are apparently the end of writer jobs, we’ve hired 3 full time in-house writers this year. We’re doing this in order to double-down on content quality and subject matter expertise. We now have the slam-dunk opportunity to beat all our competitors by using more human expertise to provide ever-higher quality content.

Our goal with content remains the same: provide great answers to searcher questions. This is fundamentally what Google wants to do too.

There are certainly going to be efficiencies in the content process. I’m really interested in using the AI as a solve for information gain and making new connections. We’ve added information gain scoring to both the SERP and individual results to FALCON AI so that our Strategy team can see how much unique content is already present in already-ranking articles. We’re scoring this on a 1-5 scale and the average has been a solid 3 so far: most keywords are samey.

For keywords with a high information gain score, we can go looking for new angles. We’ve included prompts in our content briefs on new angles for years, but a GPT-4 powered solution is much more alluring. Using LangChain, we’re able to search for related concepts from authoritative sources including academic papers, and then return a note for the writer about how the concept could be applied to the topic.

In a post about eCommerce pricing strategies, for example, we might reference some serious empirical research like Khan, Romana & Jain, Dipak. (2005). An Empirical Analysis of Price Discrimination Mechanisms and Retailer Profitability. Journal of Marketing Research. Why would we reference it? Here’s what FALCON came up with:

The content brief mentions potential drawbacks and risks associated with offering bulk discounts. Khan’s paper can be an excellent reference to identify these risks and suggest ways to mitigate them.

I love this kind of AI leverage: making the work of human experts better by surfacing easy-to-use insights that we wouldn’t otherwise find.

We added this content brief kick-off process at the start of March. We still spent 209 hours in April on content outlines, up 11% month on month. That’s because what we’re using it for is to help add information we might otherwise miss, rather than replace our research and creation process completely. It’s adding more value, rather than simply creating efficiencies.

You have to get EVERYTHING right to win with SEO

SEO is made up of many moving parts, and there’s usually no single, straightforward answer to most questions which doesn’t start with “it depends…” AI raises the stakes and adds more variables: in a world where it has just gotten incredibly easy to create low-quality content which mimics everybody else’s, the competition for the top positions will get tighter. That means getting more areas absolutely right is crucial to ranking.

We’ve already made our bet: take the efficiencies of AI and invest in higher quality content, going beyond the basics which tools like ChatGPT have made it incredibly easy to access. Your competitors can now create low quality content easily, and so can you. We plan on staying ahead.

SEO Content is a winner-takes-all marketing channel. Everyone can publish a piece of OK-quality ChatGPT content which ticks the boxes and turns the lights green. We’re instead focusing on publishing work we’ve been through a 1,000 iterations on, with the right human expertise in the loop every step of the way, raising the bar for quality content which ranks.

SEO Content is winner-takes-all, and ChatGPT makes it easier to be middle-of-the-pack. We’re going to continue being the winners.

Picture of Alex Denning

Alex Denning

Alex Denning is the Founder and Managing Director of Ellipsis®, a world-class SEO Content agency. Alex is the inventor of FALCON AI®, the next-generation AI SEO platform that helps you predict how your content will perform – before you create it. Alex is an international speaker and publishes regularly on the future of SEO. @AlexDenning on Twitter