Category

SEO

Date posted

27 Mar 2024

Read Time

Related Articles

SEO

How to supercharge your website content audits

SEO

AI and SEO: Key updates from April 2024

SEO

AI and SEO: Key updates from February 2024

February 2024 proved to be a busy month for AI news and SEO developments – particularly from Google. In this, our monthly roundup, we recap the latest news.

What is happening? 

From Bard to Gemini 

Google renamed their ‘Bard’ chatbot to ‘Gemini’; an all-encompassing name that refers to both the chatbot and the multimodal model behind it. 

Gemini Advanced

On top of the already available Pro 1.0 model of Gemini chatbot, Google has released Gemini Advanced. Gemini Advanced gives users access to their largest and most capable model called Ultra 1.0 and integrates with tools like Gmail, Google Docs, Sheets, Slides and Meet. It is more capable of executing complex tasks like coding and logical reasoning, and better understands the context from previous prompts.  

Alongside this, Google has also released a new Gemini app for Android (initially for US users) and is integrating it into the iOS Google app. Examples of usage while ‘on the go’ include taking a photo of a flat tyre and asking for instructions, or generating a custom invitation for a dinner party.

Gemini image generation

At the beginning of February, Google also announced that Bard (Gemini) can generate images powered by their updated Imagen 2 model that uses Google DeepMind’s latest text-to-image advancements via a diffusion-based model to improve in areas where text-to-image systems often struggle, such as rendering realistic hands and human faces, and keeping images free of distracting visual artefacts. 

However, despite their best efforts to generate images with responsibility in mind, three weeks later, the media jumped on the fact that Gemini was producing inaccurate images due to the way the model had been tuned by Google. It’s clear that even with the best intentions, trying to tune Large Language Models to avoid bias can result in unexpected hallucinations, overcompensation, and oversensitivity. For now, Google has paused Gemini’s image generation of people.

Gemini 1.5

Half way through the month it was perhaps surprising to see the next generation of Gemini announced already – version 1.5 – with significantly improved performance and the ability to achieve results with much less computing power than the previous release. 

One of the key breakthroughs is its long context window enabling it to process much more information (up to one million ‘tokens’ which are small building blocks such as part of a word, image, or video). Comparatively, ChatGPT-4 works with 8,000 ‘tokens’ and up to 128,000 with GPT-4 Turbo via their API.

It uses a new architecture called Mixture of Experts (so rather than using one large neural network, the models are divided into smaller ‘expert’ networks – and selectively activate only the most relevant pathways in its neural network). This ultimately means it can learn complex tasks more quickly and maintain quality, and also be trained more efficiently.

As an example, it can now process an hour of video in one go – and can accurately analyse various plot points and events, and even reason about small details in the movie – in just a few seconds. 

Beyond tasks of a standard AI chatbot, its unique strength lies in its ability to process sizable documents or multimedia files.

Non-Gemini Google news

Outside of the milestones reached by Gemini, Google has also made progress in bringing generative AI to Maps. The new functionality (intitally for US Local Guides) allows users to more accurately determine where they might want to go with a very specific set of personal requirements. 

Once a user has stated their preferences in a search query, Google’s Large Language Models can analyse Maps’ detailed information about more than 250 million places and trusted insights from the Maps community of over 300 million contributors to quickly make suggestions for where to go.

While we still wait for Google’s Search Generative Experience (SGE) in the UK (with Romania, Poland, and Turkey unofficially added to the growing list of included countries) it appears that SGE results are increasingly appearing alongside Google search results in the countries where it is accessible. A recent Brightedge study reported that in three industries, namely education, B2B tech and ecommerce, SGE appeared on 98% of queries over the previous 30 days.

That said, Google is treading carefully and has increased the SGE warnings for sensitive ‘Your Money Or Your Life’ topics, such as medical queries.

Managing ChatGPT’s memory

While Google and Gemini have overshadowed AI and SEO news in February, OpenAI continue to make strides in their chatbot’s development with the company experimenting with ChatGPT’s ability to remember conversations across all chats, reducing the need to repeat information. 

Users are also able to control what is remembered and what is not – and the more the feature is used, the better its memory becomes. For example, if you have expressed a preference for meeting notes to have headlines, bullets, and action items summarised at the bottom, ChatGPT will remember this and apply it to future tasks.

OpenAI are acutely aware that with enhanced memory comes increased privacy and safety concerns and are taking precautions to assess and mitigate biases, and steer ChatGPT away from proactively remembering sensitive information such as health information – unless you explicitly ask it to. The option to turn off the memory feature altogether remains – as does the ability to tell ChatGPT to forget specific information. 

What we think

When we were initially comparing the efficacy of the three chatbots Bard, ChatGPT, and Bing Chat (as they were then known) in 2023, we found Bard to be surprisingly effective – despite the prevailing opinion of the SEO industry that it was fairly lacklustre when compared to ChatGPT. This may have been partly due to Google’s underwhelming launch.

With the vast resources available to Google, it’s no shock that Gemini is starting to edge ahead of Chat GPT, particularly in relation to the amount of information it can process.

Gemini’s challenges with image generation underline the point of how difficult it will be for Google, OpenAI, and other similar companies to balance the avoidance of various biases while still producing accurate results for users in all cases.

The large range of SGE results across key industries points to Google being increasingly satisfied with the value this information provides for users, whilst they cautiously includes disclaimers for potentially harmful topics, as with all information they serve.

Similarly, OpenAI’s memory control for ChatGPT offers user benefits, but managing sensitive info is a responsibility they will need to cover.

Planning ahead

The recent developments in SEO related to AI carry great potential for companies, but we recommend being cautious at this stage before investing heavily in AI in many areas. For example, the recent updates to Google’s algorithms partly impacted websites that had heavily used AI tools to create content designed to rank well – and when it comes to Google, we are still waiting for the proper integration of SGE before we know how it might be possible to optimise for it. 

That said, there are evidently great efficiencies that can be made through the use of certain AI tools and we encourage their usage if not specifically used to manipulate rankings. 

Gemini 1.5’s ability to rapidly process huge amounts of multimodal information, for example, offers massive potential across a wide range of data-related tasks. The new multimodal capabilities means it continues to be vital to understand that the content we now create should also be spread across multiple modalities.