In our ongoing monthly roundup we look at AI news and developments related to SEO across June 2024. While May was extremely eventful, June was much quieter, with Google continuing to dial down the appearance of the new feature.

What is happening? 

Reduced AI Overviews presence

As May finished with a statement from Google addressing AI Overview’s teething problems, it was not surprising that a study by Brightedge conducted in early June concluded that the new SERP feature was now being shown for far fewer search queries than when it was part of the Search Generative Experience (from 84% to less than 15%). Speculation on why this might have occurred was not solely confined to quality concerns, but also the costs involved, as well as the effectiveness of Google Ads placed within this format. 

Besides highlighting the reduction in the presence of AI Overviews, the study also contained some interesting statistics, including:

• AI Overviews are 195% more likely to appear when Featured Snippets are present.
• Question searches are more likely to trigger them.
• Sitelinks are less likely to trigger them.
• Local queries are the least likely to show AI Overviews.

Evidently how likely AI Overviews are to appear also varies significantly by industry type. 63% of queries in healthcare show an AI Overview (while these tend to appear for more minor health issues, the quality and sources seemed to have improved), 32% for B2B tech, 23% for ecommerce, 18% for insurance, 14% for both education and entertainment, and just 5% for financial queries. They are unlikely to appear for restaurant and travel-related queries at all, though Google has started to use AI Overviews to write descriptions for Google Business Profiles.

While Google has reduced the overall presence of AI Overviews, many users are still unhappy with them existing at all in their current state. To address this and other related points, Google published a post in Google Search Help containing FAQs about AI Overviews sourced from its community forums. The question ‘Why can’t I disable AI Overviews?’ included the following answer: 

‘Before rolling out AI Overviews in Search, we tested the experience extensively through Search Labs to help us understand where generative AI can be most helpful. Like all the Search features we build, our goal is to help people find the information they’re looking for quickly and reliably. You might also find the Web filter useful if you’d like to focus on text-based links without features like AI Overviews. Learn more about AI Overviews: g.co/ai/overviewsinsearch.’

Many users felt this response did not directly answer the question and, continuing to criticise the feature’s quality and mandatory nature. While Google’s stock response is to suggest using the “Web” filter, this reverts back to a dated search experience and arguably misses the point of the new feature.

Gemini’s expanded context window

Towards the end of the month Google announced that Gemini 1.5 Pro now has the longest ever context window of two million tokens, which is an impressive upgrade from its previous limit of one million back in February of this year. As an example, processing six minutes of video requires over 100,000 tokens – and large code bases can exceed one million tokens.

However, as this new capability was announced, two studies raised questions about how well Gemini (and similar models) analyse complex information. The first tested their ability to understand long stories and answer true or false statements about the plots of fiction books, finding that Gemini 1.5 Pro was correct 46.7% of the time and Gemini Flash only 20% of the time. The second study was aimed at evaluating visual capabilities and showed limitations in understanding video content beyond basic details. The series of tests included visual question answering and  character recognition and found that the models have trouble once the visual context length increases. 

Apple Intelligence and chatbot integrations

June also saw the announcement of Apple Intelligence for iPhone, iPad, and Mac. This is a ‘personal intelligence system’ that combines generative AI with user data and, according to the company, can ‘…understand and create language and images, take action across apps, and draw from personal context to simplify and accelerate everyday tasks.’ 

As part of the announcement, it was also revealed that Apple would be integrating ChatGPT access into various experiences (including Siri) within iOS 18, iPadOS 18, and macOS Sequoia. While the company’s CEO Tim Cook claimed that Apple Intelligence will access information in a completely secure and private way, it didn’t take long for Elon Musk, citing privacy concerns with OpenAI, to threaten to ban the use of Apple devices in his many companies. Unphased, Apple simultaneously confirmed a likely upcoming integration with Gemini too. 

While these announcements are not directly related to SEO, speculation continues about a potential Apple search engine. Following on from these announcements, Applebot documentation was updated including Applebot-Extended, similar to Googlebot-Extended. This lets online publishers control how their content is used by Apple to train its generative AI features across its products including Apple Intelligence, Services, and Developer Tools via a website’s robots.txt file (User-agent: Applebot-Extended Disallow: /private/)

Applebot-Extended does not crawl website pages so these can still be included in the SERPs; it is only used to control the data crawled by the user agent.

What we think

Google’s reduced presence of AI Overviews is a positive step towards better quality SERPs. While testing might explain their initial prevalence, a longer testing phase before launch would have been preferable. With clear user benefits (though some would prefer the feature not exist at all), Google’s decision to dial down the presence of AI Overviews is likely also influenced by factors like cost and profit.

However, the “Web” filter is a poor alternative to an AI Overview opt-out (as Bing offers with Copilot). While useful in some cases, it generally hinders search experience and an interim solution should really be offered whilst their current change to the SERPs remains unpolished.

Beyond AI Overviews, Gemini’s rapid capability advancement is impressive, but correlation between somewhat overinflated abilities and real-world value remains uncertain. While a longer context window itself has not improved the tools immediately, it does enable escalation of training for the models indicating longer term quality gains. Ultimately, SEOs  are already benefiting from current efficiencies, and that is only going to continue as the models improve over time,  freeing our capacity to focus on creativity over routine tasks.

Planning ahead

AI Overviews are very unlikely to go away and we’re likely to see them in the UK at some point over the next six months. At RocketMill we already have weeks of testing behind us since the US rollout in mid-May, thanks to our work with brands across the globe.

Keeping a very close eye on which verticals and SERP features are most impacted by AI overall is paramount, as there are areas where a large amount of related effort is warranted, and others where time and energy spent may be more of a false economy. 

Research and testing must continue and develop, but this needs to run in parallel and be balanced carefully against the tried and tested foundations of SEO, whether that be more technical or content-focused in its nature. 

Google’s advice to focus on user experience can sometimes be overlooked in the midst of SEO optimisations. Nevertheless, it’s a core principle that benefits all marketers, essential for both short-and-long-term success.