Jaywing and Google’s partnership event titled ‘The AI Advantage’, was a sell-out event.
As a follow-up to the popular session attended by leading marketing professionals from the likes of iD Mobile, Gusbourne, Studio Retail Group and Euro Car Parts, we invited Jaywing’s Head of Innovation, Joe Crawforth, to share his knowledge and insights into AI in marketing.
Joe covers AI’s history in marketing, including the introduction, capabilities and limitations of Generative AI. Joe also addresses the need for human creative input and Generative AI’s impact on marketing and search.
Artificial Intelligence (AI) has been around for decades. As early as 1952, Arthur Samuel developed a computer program to play checkers. The term Artificial Intelligence wasn’t officially coined until 1955 when John McCarthy held a workshop at Dartmouth College (USA). AI in marketing can be traced back that far when researchers started using techniques such as linear programming, game theory and decision trees to optimise marketing mix and pricing strategies. This was significantly accelerated and became widely used in marketing in the 90’s and 00’s with the creation of the internet.
Online advertising enabled AI, in the form of machine learning, to accelerate faster. With user data on demographics, interests and browsing behaviour now being available, it enabled marketers to utilise AI to target users with specific ads, leading to a more efficient and effective use of marketing budgets.
In 2015, Google introduced RankBrain and Smart Bidding. RankBrain is a machine learning (a field of AI) algorithm that enables an understanding of the relationships between words and concepts. This enables a better understanding of a user’s intent of a search query and has had a major impact on their ranking algorithm. Smart Bidding was introduced as a machine learning approach to automated bidding strategy. This was within what was known as Google AdWords, now Google Ads, to help marketers to set optimal bids.
In 2018, BERT (Bidirectional Encoder Representations from Transformers), a Natural Language Processing (NLP) deep learning (a field within machine learning) model was introduced by Google as part of its ranking algorithm. BERT works, not just by keywords, but is trained to understand the context of words, based on the sentence it is part of. This allows the unlocking of context around the words used in a search query. Responsive Search Ads within Google Ads, used machine learning to test different combinations of headlines and descriptions to find the best performing for each search query.
Google hasn’t stopped there. In 2021, Google introduced MUM (Multitask Unified Model) which utilises a new NPL model and T5 (Text-To-Text Transfer Transformer) which Google stated was 1,000 times more powerful than BERT. MUM wasn’t just limited to text, it was multimodal, meaning that it works across text, images, video and audio. For most Google users, MUM is what is still currently driving their search experience.
Generative AI has been anticipated for a long time and is using machine learning and deep learning models to generate new content, based on an instruction. Given an instruction, such as “write a short social media post on the importance of financial planning” or “show me an image of a cartoon monkey riding a bike on a beach”, the models take the input text and generate content based on the context it has learned from its training data.
Generative AI first came to the public limelight with OpenAI’s release of ChatGPT, based on the Large Language Model (LLM), GPT3, in 2022. OpenAI isn’t the only tech company at it; Google has its own LLM chat-based generative model, known as Bard. Based on their LLM PaLM (subsequently released Gemini although little is known of its true power to date), Meta created an opensource large language model, LLaMa. There has been an explosion of smaller companies, organisations, academia and individuals all participating in this space who are really pushing the competition and rapidly increasing the number of innovations in this area.
A large language model is a deep learning model that has been trained on large corporas of text from a multitude of sources, and subsequently fine-tuned to the tasks of generating content based on an instruction or a prompt. The outcome of these LLM’s is that, given an instruction or prompt, they will generate a response word by word. It does this by supplying the highest probable word, given past words, enabling it to form sentences, paragraphs and statements.
Generative AI isn’t limited to text and LLM’s. Images, video and audio can all be generated utilising tools such as OpenAI’s DALL-E, Google’s Imagen, Stable Diffusion, MidJourney and many others. These models are trained on a large somewhat exhaustive dataset of labelled images, video and sounds, teaching the model the textual context of these combined mediums providing impressive results.
As with all new technology, there are drawbacks in Generative AI and inherent limitations. All AI, as it stands, is trained on data. This means that, if there is a flaw with the data, this flaw will be learned by the model. In addition to this, the data that the AI model is trained on, is its only view of the world, therefore it has no independent understanding of the world we live in or the concepts that may come naturally to us. Developing models with a specific use case can mitigate this but it does demand care when general models are used for a new use case.
Bias and misinformation: these are by far, two of the biggest risks within AI in general. If a dataset contains a bias, this can lead to bias outcomes that are ethically challenging and can lead to discriminatory, unfair or false results.
Time based limitations: datasets generally have a cut-off therefore the model will only be trained on data that is prior to that cut-off. This means that it has no understanding or context of new events or information that has since become available.
Creativity restraints: models are only able to generate output based on its training data so it will not be able to come up with new ideas or solutions. This means that plagiarism is a risk that needs to be considered when utilising generated content.
Hallucinations: one of the biggest challenges surrounding LLM’s is Hallucinations. This happens when the content generated contains mistruths or is outright incorrect. Due to the nature of how LLM’s operate, by predicting the next best word to output, it doesn’t understand if the statement it has generated is accurate, but only if probabilistically the word makes sense to follow. This means that there have been well documented cases where AI has stated something as fact, which simply just isn’t true.
As a result, while Generative AI is an excellent tool to support humans in content creation, humans are still required for ideation and fact checking.
The real opportunity of AI is the chance to utilise it as a tool to create efficiencies in our day-to-day lives. In its present state, it isn’t in a place to completely replace humans with the creative hard skills that are found in the common workplace. It does however provide an edge to template, concept and generate initial content in which you can then adapt and utilise in a final piece of work.
Additionally, it is a new way to provide information that was perhaps previously only available via learned knowledge, encyclopaedias and the internet. It enables us to have a seemingly natural conversation with a machine to enable that knowledge retrieval. As it stands, Generative AI covers:
It still requires a human's creative input to generate the content.
Like steam was in the industrial revolution, AI is the new steam of the time. We are still early on in the journey to realise its potential, but the opportunities appear to be endless.
Within marketing, generative AI can support marketers through lowering the skills gap on harder skills, enabling a bridged gap between content creatives and strategists.
In a creative content marketing team, for example, content ideas may be briefed into the production team prior to Generative AI. This is likely to entail a long-winded briefing process of back and forth until the idea is fully formed. Strategists can however utilise Generative AI to ideate and create a much more formed brief, so the production staff have a much clearer picture, including generated examples to pull inspiration from.
Marketers can utilise Gen AI to complete tasks such as:
Jaywing has a unique approach of utilsing both humans and AI to get the best out of your search marketing. Not only are we experts in digital marketing, creative design and CRO, we have over 25 years of success in AI research, development and utilisation.
Search is one space where the two big players, Microsoft Bing and Google, are really pushing the use of Generative AI. This started with Microsoft announcing the new Bing experience, which incorporates GPT4 (OpenAI’s latest LLM) to offer a more conversational experience. Google have since announced Search Generative Experience (SGE) which is currently only available as an experiment in the US, providing an insight into what’s to come.
Generative search experiences, as shown by both Google and Microsoft, are changing the way users will interact with search engines. A more conversational style of interacting with a search engine will become natural, with the main aim for search engines being to answer queries at a high level without needing to progress further.
Organic results can already appear pretty low down on the search engine results page given all the additional features Google offers, such as local results, news and knowledge base. Some of these might be replaced, but based on information from the US, this suggests that the generative feature will go straight to the top, following shopping ads. The outcome of this is that traditional organic results will get pushed further down overall results, which will impact visibility on clickthrough rates.
To create generative results, search engines and the AI behind them will utilise content from sites. The higher the ranking, the more likely your content will be featured. Content is optimised in an increasingly conversational way; genuine questions, answered in an original, reliable, concise, informative, and engaging way.
The use of structured data and rich snippets will enable the AI to best interpret your content and what it is about. In the age of Generative AI, optimising your content for both the human and the AI is critical. As marketers you need to ensure that your content, not only carries your brand’s tone of voice and is conversion optimised, but also how that content will be interpreted by AI. The sooner content is optimised against both paradigms, the greater the running start your brand will have when the generative experiences come to us.
Generative search experiences will, not only impact organic search, but paid search will also be impacted in the way that the engines will offer your ads:
Google’s Performance Max has been around for a number of years and is a move away from the more visible and controllable environment of campaigns, ad groups, keywords and bid modifiers. Initially it appeared a stronger way for Google to control the paid search landscape, but on reflection and with the progression of AI and with SGE in mind, Performance Max makes much more sense.
Through utilising assets, Google can utilise their AI and optimise the ads they show across any of their platforms, SGE’s generated content included. Having a strong understanding of which search content assets perform well will be important. Providing Google with a large supply of assets may seem like a good way to understand that, but as with any optimisation, a test and learn approach will be taken.
The more assets you have for Google to test, the longer the process will take to hit its peak. Utilising AI to understand this, prior to adding the assets to Google, followed by allowing Google to fine tune your understanding of what works and what doesn’t, will improve the time it takes to achieve the best results available, based on your budgets and KPI’s.
AI is here, it’s here to stay but AI replacing humans is still something for science fiction writers to play with, at least for the foreseeable future. Nearly all the short to medium term opportunities come from adding AI to our toolbox, enabling it to enhance our capabilities. It can spot and highlight patterns, do the leg work, create drafts and suggestions. It can produce high volumes of output but select and review any output. Humans are still required for the creativity and novelty of setting tasks, which is the most interesting part of any process.
The adoption of AI and truly understanding how to utilise these new tools, will improve our effectiveness and output making our lives easier and enabling us to do what we are really good at: thinking, being creative and innovative, as at least for now this is beyond the realms of AI.
As marketers, we not only need to understand how to use these tools to our advantage, but also how the likes of Google are using these tools. In understanding this, it will give us the edge to better optimise our content and strategies to ensure that we are still staying relevant in the new world.
SGE utilises generative AI, providing you with: