In our third monthly roundup of the latest AI news and developments related  to SEO, we take a look back at what happened in March 2024. While last month was fairly eventful – particularly for Google with their release of Gemini 1.5 – March has been a little quieter. 

Might March be the calm before the April storm..?

What is happening? 

A month of updates

March 2024 core update

On the 5th of March, Google began rolling out the March 2024 core update which is more complex than its usual core updates, involving changes to multiple systems. Amongst other improvements, the update prioritises user-focused content, surfacing less content in the search results that is intended to attract clicks, ultimately reducing clickbait.

March 2024 spam update 

On the same day, Google also announced the March 2024 spam update that rolled out across a two-week period. This incorporated three new policies to tackle increasing levels of bad practice, namely expired domain abuse, site reputation abuse, and scaled content abuse. The latter relates to the practice of creating large amounts of unoriginal content with the sole purpose of manipulating search rankings. It’s interesting to note that one of the specific examples Google gave was “using generative AI tools or other similar tools to generate many pages without adding value for users”. 

As it happens, Google uses an AI-based spam-prevention system called SpamBrain that was first launched in 2018 and has been continuously improved upon – it has made further improvements to that system as part of the March 2024 spam update. 

Search Quality Rater Guidelines

Again on the 5th of March, Google updated its Search Quality Rater Guidelines to the latest version that contains a single mention of AI in relation to an example website being evaluated that is considered deceptive and untrustworthy: This page has a deceptive purpose: the page claims to be an informational resource for parents, but the website terms of service state that it is published for AI enthusiasts and is not intended for the general public. The website terms of use states that “some articles” are generated by artificial intelligence and may have errors or be out of date; there is no indication to which pages this statement applies. The information in this article is not trustworthy and is Lowest E-E-A-T.

SGE expands to non-test users in the US, and trials in the UK

Google’s Search Generative Experience (SGE) has been expanded to non-test users who haven’t opted in for a small number of users in the US, but only for a subset of queries where searchers will supposedly benefit from the enhancements. These types of queries tend to be more complex in nature, or involve questions where it may be helpful to get information from a wide range of sources. An example given is [how do I get marks off painted walls]. Part of the reason Google is moving in this direction is so it can gain feedback from users who have not opted into SGE specifically to learn how the broader population may find this technology useful.

Google has also started trialling SGE in the UK to a small proportion of logged-in users. 

The relaunch of Bing Deep Search

Initially off to a shaky start and subsequently stopped from being fully tested, Bing Deep Search is now live again. Despite a lacklustre reception from users, Microsoft maintains that it turned out to be more popular than the company anticipated, although it is intended to be more of a research tool. 

Deep Search uses Bing’s index and ranking system, but enhances these with OpenAI’s GPT-4, so that the initial query is expanded into a more comprehensive description of what an ideal set of results should include. It is designed to return relevant results that often don’t show up in typical search results by using ‘querying techniques’ to locate pages that might match an expanded query, rewriting the query, and searching for those variations too. 

While it comes with a disclaimer that ‘GPT-powered results takes ~30 secs’, some users are reporting that its speed has improved significantly. 

Copilot uses GPT-4 Turbo, Copilot GPT builder rollout, and Pro free trial

Microsoft Copilot now uses GPT-4 Turbo instead of GPT-4 in the free tier across all three modes of chat (Creative, Balanced, and Precise). Pro users however have the option to select either version via a toggle in the interface. Microsoft has explained that the three chat modes use GPT-4 Turbo in differing amounts, with Creative and Precise using it almost fully, but Balanced only using it in certain cases.

Additionally, Microsoft launched Copilot GPT builder for Copilot Pro subscribers; this allows users to create their own AI models without any specific coding knowledge. When Copilot Pro was announced in January, the GPT builder was highlighted as one of the subscription plan features but is only now rolling out, albeit in a phased manner.

In order to entice more users to the Pro plan, Microsoft is launching a free one month trial (available for users of the Copilot app). The company has also expanded its availability to more markets, with it now available in 222 countries in total. So as well as giving users access to the latest OpenAI models, the ability to build Copilot GPTs, and the use of Copilot inside Microsoft Office apps (for existing Microsoft 365 Personal or Home subscribers), Copilot Pro is now also available in the Office web apps, so users don’t need a separate Microsoft 365 subscription to use it in Word, Outlook, and other free web apps.

What we think

Just as Copilot offers three modes of chat, Google search representatives tend to respond in one of three ways:

• The consistent reiterations: Examples like “hreflang being one of the hardest parts of SEO” are frequently repeated.
• The ‘it depends’ responses: Questions about ranking often receive these answers (valid, but potentially overused).
• The changes in stance: Core Web Vitals and AI are examples where Google’s position has evolved.

But it does seem that, since the hype around ChatGPT 3.5, Google is pushing AI much more heavily and with a more concrete stance. Recent updates and revisions to the Search Quality Rater Guidelines related to AI now clearly state that using generative AI to manipulate search rankings will only backfire; if AI is to be used, the content that it outputs would need to provide real value to users and ideally be original in nature. While future developments might make this more possible, in its current early stage, the very nature of pre-training seems unlikely to produce these kinds of results. So while AI tools can be very useful for many elements of SEO, using them to create the core content itself is counterintuitive.

Ironically, to tackle spam often partly produced with AI, Google uses AI itself to prevent it from surfacing. This approach of “fighting fire with fire” is likely to continue, as Google has used machine learning in various forms for years now.

Google’s approach with SGE is one of cautious exploration. With SGE, they are clearly still treading carefully – both in terms of its rollout and the features it includes. It still feels like they are very much in the experimental phase, and while Google is growing increasingly confident about certain types of queries that are genuinely enhanced with this format, the narrow segmentation of availability indicates that it is looking for much more feedback based on testing before it pushes it more aggressively – which in fairness is probably a wise decision.

These concerns about reduced traffic are not unfounded. Many SEOs are concerned that SGE will result in reduced traffic to the sites they work on – in somewhat of a similar manner to when Featured Snippets were introduced. Google’s reassurance that it is the company’s priority to send traffic to content creators may hold some truth, though there might also be some validity to the claim that users are interacting with a broader range of sources.

Microsoft is clearly investing heavily in innovation with Bing and Copilot, despite their still-limited user base and performance shortcomings compared to the competition. Of course, the integration with ChatGPT keeps it very much afloat, but it’s interesting to speculate on where it would be without this. The expansion into multiple regions and the pushing of free trials suggests that it is willing to play a more aggressive game than Google – perhaps in part because it has less to lose. Either way, it’s clear that it will be heavily integrating Copilot into a range of Microsoft products even if Bing remains a relatively small piece of the search puzzle.

Planning ahead

While AI-powered content creation is on the rise, at RocketMill we advise a cautious approach, especially since SGE’s UK rollout hasn’t happened yet. However, staying informed about AI developments to identify opportunities as early as possible and using AI tools to streamline workflows remain crucial.

It’s more important than ever to pay close attention to Google updates (not just the elements related to AI) and focus on creating content and websites that genuinely provide value to users, rather than just trying to rank for high-volume keywords in isolation. Black-hat SEO tactics (including those targeted by the recent spam update) are increasingly ineffective and alternative approaches are required. For example, we avoid link-building campaigns in favour of alternative techniques that deliver better results for our clients and ultimately their customers.

As Google is growing more confident in SGE’s ability to provide value to users with complex queries that collate information from a wide range of sources, it might prove challenging to directly optimise for them. However, if the content you create is original, genuine, and valuable then it will more likely increase visibility. At this very early stage, focusing on elements of E-E-A-T and user service is more effective – for now let the AI dust settle where it may.