Large Language Models (LLMs), like the technology underlying ChatGPT, are advanced AI systems designed to understand, generate, and work with human language. Imagine a highly skilled librarian who has read and memorized vast amounts of text from the internet. This librarian can help you write an essay, answer your questions, or even create stories. LLMs are like digital versions of this librarian, but with some differences.
At their core, LLMs are built using a technology called “Transformers,” which allows them to pay attention to different parts of a sentence or text to understand its meaning better. It’s akin to how you might focus on specific words or phrases while reading to grasp the overall context.
One of the key strengths of LLMs is their ability to learn from a large amount of text data available on the internet. This learning process is known as “transfer learning.” Here’s why it’s important:
1. Common Language Knowledge: Many tasks in language processing, like understanding grammar or context, are common across different tasks. By learning these once, the model can apply this knowledge to a wide range of language tasks.
2. Making the Most of Limited Data: Good quality, annotated data (where humans have marked the correct responses) is hard to come by. Transfer learning lets LLMs learn from whatever high-quality data is available.
3. Leveraging Abundant Unlabeled Data: The internet is a treasure trove of text data. LLMs can learn from this vast, unlabeled data pool, extracting patterns and knowledge.
4. State-of-the-Art Performance: In practice, transfer learning has proven to be highly effective, leading to groundbreaking performance in many language tasks like text classification, question answering, and information extraction.
In simpler terms, LLMs are like digital brains that have read almost everything on the internet. They use this knowledge to understand what we ask them and respond in a way that’s helpful, whether it’s writing a piece of text, answering a question, or even generating creative content. But remember, while they are incredibly knowledgeable, they don’t ‘think’ or ‘understand’ like humans do. They are more like very sophisticated pattern recognizers, using the vast information they have been trained on to make educated guesses about what to say in response to our queries.
Scalenut is an AI-powered co-pilot designed to manage the entire SEO lifecycle. Aimed at making the SEO process more streamlined and effective, Scalenut offers a range of functionalities that assist in various aspects of SEO strategy and implementation.
With Scalenut, users can effortlessly conduct keyword research and competitor analysis, giving them an edge in understanding market dynamics. This empowers you to make data-driven decisions, effectively targeting keywords that are not just high in search volume but also relevant to your specific audience. Scalenut also provides valuable insights into content gaps, suggesting areas where you can create impactful content to capture more organic traffic.
One of the standout features of Scalenut is its content creation and optimization engine. By leveraging advanced AI algorithms, it aids in the development of content that not only ranks well but also resonates with your target audience.
It takes into account critical factors such as content length, keyword density, and readability, ensuring that you produce content that is both SEO-friendly and user-centric.
For those looking for an all-in-one solution to tackle the complexities of SEO, from planning to execution, Scalenut offers a robust and versatile platform that adapts to your specific needs.
In the Search Engine Optimization (SEO) world, achieving relevance is a crucial goal driving strategic initiatives and tactical implementation.
A few weeks ago, Paul Thomas and a group of researchers from Microsoft captured Dawn Anderson and subsequently my attention by publishing a revolutionary paper titled “Large language models can accurately predict searcher preferences” on how to use large language models (LLMs) to generate high-quality relevance labels to improve the alignment of search queries and content.
Both Google and Bing have heavily invested in relevance labeling to shape the perceived quality of search results. In doing so, over the years, they faced a dilemma – ensuring scalability in acquiring labels while guaranteeing these labels’ accuracy. Relevance labeling is a complex challenge for anyone developing a modern search engine, and the idea that part of this work can be fully automated using synthetic data (information artificially created) is simply transformative.
Before diving into the specifics of the research, let me introduce a new free tool to evaluate the match between a query and the content of a web page that takes advantage of Bing’s team insights.
I reverse-engineered the setup presented in the paper, as indicated by Victor Pan in this Twitter Thread.
I'm still digesting the Microsoft paper about using LLM's to classify intent being met by content to replace trained humans and professional humans on categorization.
Enter the search intent, this is the narrative behind the information needed by the user.
We provide a simple traffic light system to show how well your content matches the search intent.
(M) Measure how well the content matches the intent of the query.
(T) Indicates how trustworthy the web page is.
(O) Considering the aspects above and the relative importance of each provides the score as follows:
2 = highly relevant, very helpful for this query
1 = relevant, may be partly helpful but might contain other irrelevant content
0 = not relevant, should never be shown for this query
Let’s Run A Quick Validation Test
While we are still working on conducting a more extensive validation test, here is how the experiment is setup:
We’re looking for top-ranking and least-ranking queries (along with their search intent) behind blog posts on our website;
We’re evaluating how the tool considers these two classes of queries;
We manually labeled the match between content and query (ground truth) and we are analyzing the gap between the human labels and the synthetic data.
The exact page (a blog post on how to get a knowledge panel), while trustworthy, is obviously a good match for the query “how to get a knowledge panel” and it doesn’t match at all the query “making carbonara” (ok, this one was easy).
Here is one more example. In the blog post for AI plagiarism, the tool finds relevancy for the query “ai plagiarism checker” but finds only partially the content relevant for the query “turing test”.
Current Limitations
While this tool is free, its continued availability is not guaranteed. It operates using the WordLift Inspector API, which currently does not support JavaScript rendering. Therefore, the tool will not function if you’re analyzing a webpage rendered client-side using JavaScript. I meticulously replicated the same configuration described in the paper (GPT-4 on Azure OpenAI) but the system is currently running on a single instance and you have to be patient while waiting for the final result.
What We Learned From Microsoft’s Research
Relevance labels, crucial for assessing search systems, are traditionally sourced from third-party labelers. However, this can result in subpar quality if labelers need to grasp user needs. The paper suggests employing large language models (LLMs) enriched with direct user feedback can generate superior relevance labels. Trials on TREC-Robust data revealed that LLM-derived labels rival or surpass human accuracy.
When implemented at Bing, LLM labels outperformed trained human labelers, offering cost savings and expedited iterations. Moreover, integrating LLM labels into Bing’s ranking system boosted its relevance significantly. While LLM labeling presents challenges like bias, overfitting, and environmental concerns, it underscores the potential of LLMs in delivering high-quality relevance labeling.
This is incredibly valuable for SEOs when evaluating how the content on a web page matches a target search intent.
Google’s Quality Raters
Google utilizes a global team of approximately 16,000 Quality Raters to assess and enhance the quality of its search results, ensuring they align with user queries and provide value. This Quality Raters program, operational since at least 2005, employs individuals via short-term contracts to evaluate Google’s Search Engine Results Pages (SERPs) based on specific guidelines, focusing mainly on the quality and relevance of displayed results.
Google Quality Raters follow a meticulous process defined by Google’s guidelines to evaluate webpage quality and the alignment of page content with user queries. They evaluate the page’s ability to achieve its purpose using E-E-A-T parameters (Experience, Expertise, Authoritativeness, and Trustworthiness). They also ensure that the content effectively satisfies user needs and search intent.
Although Quality Raters do not directly influence Google’s rankings, their evaluations indirectly impact Google’s search algorithms. Their assessments, particularly regarding whether webpages meet specified quality and relevance criteria, guide algorithm adjustments to enhance user experience and satisfaction. This human analysis is crucial for identifying and mitigating issues, such as disinformation, that might slip through algorithmic filters, ensuring that SERPs uphold high standards of quality and relevance.
Moreover, the Quality Raters’ feedback, especially on the usefulness or non-usefulness of search results, also aids in training Google’s machine learning algorithms, enhancing the search engine’s ability to deliver increasingly relevant and high-quality results over time. This is pivotal for YMYL (Your Money or Your Life) topics, which require elevated scrutiny due to their potential impact on users’ health, finances, or safety. The feedback and evaluations from the Quality Raters, therefore, serve as a valuable resource for Google in its continual quest to refine and optimize its search algorithms and maintain the efficacy of its search results.
To learn more about Google’s quality raters Cyrus Shepherd has written recently about his experience as quality raters for Google. Cyrus’s article is super interesting and informative as always!
Conclusions And Future Work
We aim to continue enhancing our content creation tool by merging knowledge graphs with large language models. Research like the one presented in this article can significantly improve the process of output validation. In the coming weeks we plan to extend the validation tests and compare rankings from Google Search Console with results from the Search Intent Optimization Tool to assess its value in the realm of SEO across multiple verticals.
If you’re interested in producing engaging and informative content on a large scale or review your SEO strategy, drop us an email!
Generative AI, in the context of search engine optimization (SEO) for publishers and news businesses, refers to the use of artificial intelligence techniques to create original and high-quality contentthat aligns with the preferences and demands of search engines and users.
In the modern AI world, generative AI has tremendous implications for publishers and news businesses. It enables them to streamline and enhance their content creation process, leading to improved search engine rankings, increased organic traffic, and better engagement with their target audience.
By leveraging generative AI, editorial teams have the potential to automate whole parts or certain segments of their content generation process, such as articles, blog posts, and product descriptions. This technology utilizes advanced algorithms and natural language processing models to understand the underlying structure and context of the desired content. It then generates human-like text that is coherent, informative, and tailored to specific topics or keywords.
Generative AI empowers publishers and news businesses to produce a larger volume of content in less time, allowing them to keep up with the ever-increasing demand for information. It also helps in addressing content gaps and optimizing for specific search queries. You can achieve this by generating relevant content that caters to the interests and intent of their target audience.
Moreover, generative AI can aid in personalization efforts by creating customized content based on user preferences, search history, and other available data. This personalized approach enhances user experience, increases engagement, and encourages repeat visits, resulting in higher user satisfaction and loyalty.
However, it’s important to note that while generative AI can be a valuable tool, it should be used responsibly and ethically. Publishers and news businesses must ensure that the generated content is accurate, reliable, fact-checked, and compliant with journalistic standards. Human oversight and editorial judgment are crucial to maintain credibility and trust with the audience.
Do I Really Need A Generative AI Strategy For My Publishing Business?
We have been involved in the publishing industry since the inception of the content marketing era, and that is exactly what we practice daily with our WordLift team. It would be inaccurate to suggest that we have yet to internally discuss and question the utility of these new tools and models readily available with just a few clicks.
We fully comprehend your standpoint: every publisher’s objective is to enhance their business and differentiate their content in uniqueness and originality from other online ventures. You genuinely care about your users and aspire to establish yourself as relevant and authoritative as possible.
That’s completely understandable and, more importantly, achievable. We have shared your position both in the present and in the past, constantly innovating on behalf of our customers. This is why we believe we are the ideal partner to assist you during these uncertain times. AI, knowledge graphs, and linked data with generative AI have been a thrilling journey to create scalable content that genuinely benefits users.
We strongly advocate for transparency and wholeheartedly support returning control to publishers themselves. In addition to investing significantly in innovative tools and cutting-edge end-to-end SEO solutions for content marketing and publishing businesses, we advocate for implementing “no AI” tags and a more detailed definition of what can and cannot be done with content throughschema markup strategies.
We are actively staying informed about the latest content and AI regulatory initiatives spearheaded by other influential content and AI industry figures. Both Coalition for Content Provenance and Authenticity (C2PA)and IPTC,the global standards body of the news media,are currently working on various options in this field. We anticipate the introduction of additional subsets of properties soon, specifically in terms of schema markup.
How Can You Integrate Generative AI Into Your Content Marketing Strategy And Toolkit?
Establishing the proper working framework and internal processes and fostering effective team partnerships is crucial in business and life. Implementing a framework solution, such as a content knowledge graph, empowers you to develop more relevant generative search experiences that are future-proof and enable resilience.
Based on our past client experience, this is easier said than done for medium-sized and large companies. They often need to catch up in anticipating these emerging search developments and addressing their users’ needs. If your company falls into this category, investing in organizational change, revamping workflows, and even redefining concepts will be essential for your content business.
Particularly from a search perspective, it’s important to note that SEO has evolved beyond search engine optimization. We are now in the era of organizing and optimizing search experiences. What we used to know about how SEO worked by 2022 already belongs in the past. You need a holistic strategy that covers your content business as a whole, not just some random ChatGPT experimentation done on an individual level.
Define Objectives: Begin by clearly defining your objectives and goals for incorporating generative AI into your content marketing strategy. Determine the specific outcomes you want to achieve and how generative AI can help you in reaching those goals.
Assess Data Availability: Evaluate the availability and quality of data that will be used to train the generative AI models. Identify the relevant datasets that align with your content marketing needs and ensure they are comprehensive and representative.
Model Selection: Choose the appropriate generative AI model that aligns with your objectives and the nature of your content. Consider factors such as the model’s capabilities, performance, and compatibility with your existing toolkit.
Data Preprocessing: Prepare and preprocess your data to ensure it is in a suitable format for training the generative AI model. This may involve cleaning and organizing the data, removing any inconsistencies or biases, and transforming it into a format compatible with the model.
Training the Model: Train the generative AI model using the preprocessed data. This step involves feeding the data into the model, adjusting parameters, and iteratively refining the model’s performance through multiple training cycles.
Evaluation and Fine-tuning: Evaluate the performance of the trained model using appropriate metrics and validation techniques. Identify areas for improvement and fine-tune the model accordingly to enhance its output quality and relevance.
Integration and Workflow Development: Integrate the generative AI model into your existing content marketing workflow. Develop a streamlined process for generating AI-driven content, incorporating the model’s outputs into your content creation and distribution pipeline.
Monitoring and Iteration: Continuously monitor the performance and impact of the generative AI integration. Gather feedback from users and stakeholders, and iterate on the model and workflows as needed to optimize results and adapt to evolving needs.
Throughout this process, it is crucial to maintain ethical considerations, ensure transparency in the use of generative AI, and comply with any relevant regulations and guidelines governing AI technologies.
How Wordlift Is Moving In This Direction To Help Publishers Use Generative AI?
We’ve been privileged to work with small and medium-sized companies but also huge brands that werepioneers in innovating intelligent customer experiences in the new generative AI search era.
We’ve also been very fortunate to have an internal team of highly-skilled flexible professionals that know our customers’ needs by heart and pride themselves in growing an online publishing business like yours.
Here’s how we develop our product for the generative AI era.
Creating KG-powered Agents to give the reader the opportunity to talk with an article and its author
We are exploring AI-driven experiences to assist news and media publishers and e-commerce shop owners. These experiences leverage data from a knowledge graph and employ Large Language Model (LLM) with transfer learning in context.
An example is in this article written by Andrea Volpini. In it, you can try the ‘AskMe’ widget, a function powered by the knowledge graph data embedded in the blog. You can ask questions such as “What is this article about?” or “What are Andrea’s thoughts on structured data?”. This is a first step towards empowering authors, putting them at the center of the creative process, and keeping them in complete control.
Introducing the Content Generation Tool by WordLift
This new feature is designed to allow our clients to create high-quality content at scale. This new feature leverages data from the Knowledge Graph, allowing you to generate compelling and customized content for your brand. With it, you can use a query to extract data from your KG and create a customized prompt template to generate engaging content. We have incorporated a robust validation process where you can define your rules to ensure the highest quality and alignment with your brand identity. These rules allow you to fine-tune the generated content, ensuring that the result perfectly encapsulates the desired tone, style, and messaging.
We are working on adding a new function to our SEO Add-on that will allow you to create content parts containing entities you select because they are considered most relevant to your business. This will make optimizing your website’s content easier and faster to rank higher in Google right from the start.
Creating a fine-tuned model to generate descriptions for products and snippets of text that can cover long-tail intents
It is all about your data. In this case, you can customize the template in its final part and train it using the best content already produced for your brand. For example, we used GPT-3 to generate product descriptions for an e-commerce shop automatically.
And so much more! These were just the latest developments👏
Other Frequently Asked Questions
How can generative AI benefit publishers?
Generative AI can aid publishers in developing content workflows and content strategies at scale. Combining the right team, organizational culture, tooling, expertise, SEO and AI best practices will be crucial for success.
Can generative AI replace human writers in publishing?
If your intention is to generate content on a large scale without taking user intents into proper consideration or devising a dedicated content strategy, then the answer is yes. However, if you prioritize high-quality content publishing, a human-in-the-loop approach is essential. This involves leveraging the expertise, best practices, and internal collaboration among teams to facilitate scalable, human-centered content creation.
Are there any ethical concerns with using generative AI in publishing?
Yes, standardization teams are actively working to determine the most suitable approach for labeling generative AI content, particularly with regard to author rights and addressing misinformation. This is why the combination of human expertise alongside technology proves to be the most effective approach at this current juncture.
How can publishers measure the effectiveness of their generative AI strategy?
This depends on different business setups and objectives but there are several ideas on how to achieve this. Publishers can measure the effectiveness of their generative AI strategy by analyzing content performance metrics, conversion rates, user feedback and satisfaction, SEO performance, and cost efficiency, conducting a comparative analysis, and monitoring the long-term impact on brand reputation and customer retention. A combination of quantitative metrics and qualitative feedback should be used to obtain a comprehensive evaluation of the strategy’s performance.
SEO is changing, and it is meant to become a lot different from what it used to be, primarily because of the changing nature of the user interface. Generative AI transforms how we create, as its name might imply, and how we access, find and consume information.
This doesn’t mean that SEO is dead; quite the opposite. It is finally becoming something different. And, there will not be a unifying horizon as we had with Google for the last 25 years (yes, happy B-day, Google!). New gatekeepers are emerging while Google is fighting back to retain its dominant position at various levels:
User experience: with the introduction of the Search Generative Experience (SGE);
Knowledge acquisition: with the advancements in robotics and autonomous driving, AI models start to experience the world (also known as knowledge embodiment) as we do to gain additional experience;
Knowledge representation: as seen in Gemini, Google’s latest large foundation model, new abilities emerge when a model is trained seamlessly using all media types simultaneously (cross-modality). At the architectural level, Google, starting with PaLM, invests in sparse models (as opposed to dense architecture) that more efficiently integrate with external APIs, knowledge bases and tools.
In this article, I will introduce some examples of how AI Autonomous Agents can generate content or conduct simple SEO tasks and the need for a multi-sided approach when developing such systems.
It is no accident that human brains contain so many different and specialized brain centers.
Marvin Minsky, 1991
The Anatomy of an AI Autonomous Agent
What are they
An AI agent, as the paper “The Rise and Potential of Large Language Model Based Agents: A Survey” explains, is essentially built of:
a brain primarily composed, in our vision, by the language model paired with a knowledge graph, acting as its long-term memory;
a perception layer that interacts with the environment (the environment might include the content editor or the SEO orchestrating the task) and;
a set of actions that the agent can accomplish. The tooling it has. From the APIs, it can use the sub-graphs of its memory (all the products in the catalog or all the articles written by a given author).
The Knowledge Graph is ‘the book’ the language model ‘reads‘ before making its prediction. The Knowledge Graph acts as the persistent memory layer of the AI Agent. The short memory (the conversation turns in the chat session) can also be stored back in the KG, but this is less relevant; their use remains limited to the user session.
The content generated or the Agent’s keyword analysis is stored in triples using the reference ontology or schema vocabulary. This way, the systems evolve and learn from interacting with content editors, marketers, and SEOs within the same organization.
The Evolution of AI Agents
From Symbolic to Deep Learning and Back
Let’s begin our journey in the early ’80s when artificial intelligence was a developing field, primarily rooted in symbolic agents. These agents represented a first attempt to model human cognition in binary code. They relied on symbolic logic, explicit rules, and semantic networks. Extremely elegant and yet computationally limited and not truly scalable.
Moving forward, we witnessed the development of reactive agents. These agents took a different route by reacting directly to environmental stimuli. There were no internal models, ontologies, or complex reasoning; they operated much like rule 30, the cellular automata discovered by Stephen Wolfram: a set of simple rules creating complex behavior.
Now, let’s traverse to reinforcement learning-based agents, with AlphaGo being the most beautiful representation. These agents found a harmonic blend of experience and optimisation. They interact with their environments and learn optimal behaviours guided by a reward system. These agents can learn intricate policies from high-dimensional inputs without human intervention.
Fast-forwarding to today, we arrived at the era of LLM-based agents.
Since the mass introduction of ChatGPT in late December of 2022, a new functional layer has been added to the traditional web application stack. A layer that is radically transforming the Internet, giving applications and services the ability to talk and, somehow, reason in a similar way as humans do.
In a few months, we have transitioned from leveraging basic language models to instruction-based language models to AI agents (Auto-GPT, GPT Baby and now Meta GPT). Each step lays the foundation of a sequence-to-sequence revolutionary neural network architecture known as transformers and its attention mechanism (a mathematical wonder that helps models keep the focus on tokens that matter).
This transition happened as we realized that being able to recognise a pattern in an isolated state is not the same as interpreting the same pattern when it is a component of a more complex system. In other words, a Large Language Model doesn’t know what it knows. While it can help close knowledge gaps, it is designed to hallucinate and will always remain unreliable.
By providing instructions during training, things have drastically improved; models rather than predicting similar sentences (What is the capital of Italy? >> What’s Italy’s capital? What is the centre of power in Italy?) have learned how to answer the questions (What is the capital of Italy? >> The capital of Italy is Rome). Yet, there is yet to be a solution to the fundamental unreliability as knowledge representation remains a problem. There is no “single best way” to represent a given knowledge domain. Each area requires its level of connection density and its rule-based ontology.
Peter and Juan are apostles The apostles are twelve Are Peter and Juan twelve?
Giuseppe Peano
The secret of what something means lies in how it connects to other things we know. That’s why it’s almost always wrong to seek the real meaning of anything. A thing with just one meaning has scarcely any meaning at all.
Marvin Minsky, “The Society of the Mind”, 1987.
An Agent implements what Minsky suggested in the early days of AI: the ability to combine “the expressiveness and procedural versatility of symbolic systems with the fuzziness and adaptiveness of connectionist representations”. Moreover, an Agent becomes the connecting tissue that blends existing computational functions and web APIs. It only needs to understand the tools (agencies) it can use and the mission of its task.
GraphQL as a Data Agency
Unless a distributed system has enough ability to crystallize its knowledge into lucid representations of its new subconcepts and substructures, its ability to learn will eventually slow, and it will be unable to solve problems beyond a certain degree of complexity.
Marvin Minsky, 1991
GraphQL is a querying language coupled with an execution engine designed with service APIs to help us extract the data we need from a knowledge base or to add new data.
Building a Data Agent that “talks” with our GraphQL end-point
Let’s do a first example by connecting a simple agent to a knowledge graph.
In this ultra-simple implementation, our data agent can:
translate natural language into a GraphQL query,
execute the query,
analyze the results and provide an output as indicated by its prompt.
If they find a parrot who could answer to everything, I would claim it to be an intelligent being without hesitation.
Denis Diderot, 1875
Autonomous AI Agents for Content Creation
The new typewriter
Given the fragmented nature of the audience and the need for trustworthy relationships between authors and readers, content generated by advanced language models shall take several factors into account:
Syntax: This refers to the grammatical structure of the text, essentially the backbone of any language. Syntax is governed by rules dictating how words are assembled into sentences. Advanced language models learn these syntactic rules by breaking down terms into smaller units called ‘tokens’ and identifying patterns in large text datasets.
Semantics: This is where syntax and meaning intersect. In language, a word is more than just a string of characters; it represents a point in a multidimensional space of concepts. In the field of Natural Language Processing (NLP), this is modelled through techniques like embeddings and graph representations.
Praxis: This involves the real-world application and use of language in social settings. Praxis adapts over time and can be influenced by altering the prompts given to language models, thereby changing the tone and context of the generated text. We use structured data and the knowledge graph to create the context for the autonomous agent.
We can produce grammatically sound, meaningful, and contextually appropriate content by understanding these elements. These layers characterize the human language in a framework initially introduced by C.W. Morris in 1938.
Introducing Graph Retrieval-Augmented Generator (G-RAG). An Agent that writes content like you do
Let’s advance now on a G-RAG Agent. Retrieval Augmented Generation(RAG) is a technique that combines a retriever and a generator to improve the accuracy of the prediction.
The retriever is responsible for searching through an external source (like a knowledge graph or a website’s content) and finding relevant information. At the same time, the generator takes this information to produce a coherent and contextually accurate response.
RAGs enhance the performance of Large Language Models (LLMs) by making them more context-aware and capable of generating more accurate and relevant responses.
The continuous nature of deep neural networks (like the one used by GPT models) makes an AI Agent capable of inductively analyzing large amounts of data, extracting patterns and predicting the upcoming sequence. The question we have to ask is though when designing these tools:
What data shall we feed into these systems?
What information creates the dialogic interaction (or the content) we want?
How can we steer a transformer-based language model to provide the correct answer (what does the prompt look like)?
What tools does the Agent need to have in its toolbox?
How do I describe these tools so the Agent can effectively use them?
The discrete nature of Knowledge Graphs, designed to organize a wealth of data by turning facts into triples (subject-predicate-object) and information into relationships, makes them a strategic asset for building such applications. Still, we cannot only rely on a symbol-oriented solution and language models.
When developing AI Agents for SEO, we realize there is no “right way” and that the time has come to build systems that combine diverse components based on the specific task.
An emerging trend involving orchestrator frameworks like LangChain and Llama Index and graph providers like NebulaGraph, TigerGraph and WordLift, among others, are pioneering a new paradigm: using knowledge graphs in the retrieval process. Semantically rich data, modeled with an ontology, is used to train the language model, build the retrieval system, and guard-rail the final generation.
Structured Data: a tool for the Agent to write SEO-friendly content using the content of your website
While it may be tempting to use advanced language models like ChatGPT or GPT-4 (with sophisticated prompts) for content creation, it is akin to asking a random person on the street how to design an SEO campaign for an international brand. While common sense may yield valuable insights, the overall strategy must be revised. Similarly, just as Google utilizes structured data to interpret the content of webpages, a Graph RAG (a Recursive Answer Generator based on a Knowledge Graph) can employ structured metadata to construct an index that serves as a retrieval mechanism. The agent will be instructed to find the relevant documents to be fed into the prompt.
Generate Content by using Structured Data
Much like the process outlined in fine-tuning GPT-3.5, structured data empowers us to parse through a website’s content, identifying key attributes that are crucial for building an efficient Retrieval-Augmented Generation (RAG) for content creation. Below is a practical example.
We employ WordLift Reader to generate one or more vector-based indices for our articles, specifically those marked as schema:Article. We then configure the fields that will be indexed (known as text_fields) and designate the fields to be used as metadata (known as metadata_fields). The reader automatically crawls the web page when its URL is added to the list of text_fields. Each index thus created can serve as a tool for our AI Agent.
Let’s now generate a brief paragraph on the fusion between knowledge graphs and LLMs using the content from this blog. The retriever has identified a blog post from our blog, and the LLM (a fine-tuned version of Chat-GPT 3.5 Turbo) is generating the expected paragraph. We use the detected sources (previously written blog posts) to steer the completion. Sources also represent a valuable opportunity:
To explain why the LLM is writing what it is writing (explainability builds trust with both the reader and eventually Google),
To add relevant internal links that will guide the reader further in exploring more content from the website.
An Autonomous AI Agent for Entity Analysis and Content Revamp
Your new Semantic SEO Agent
We are developing an agent designed to analyze search rankings and conduct entity analysis on existing web pages.
The following prompt describes the core function of this agent: it extracts entities from various web pages, identifying potential gaps in content.
When provided with a specific search query, the agent will determine which entities are essential for our web page to achieve a higher ranking.
As illustrated below, the agent uses the WordLift Content Analysis API to extract entities from a webpage on merkur.de (an established news outlet for Bavaria). Behind the scenes, the agent communicates with the API using the WordLift key. From the retrieved list of entities, it then hones in on those pertinent to the target query “Oktoberfest 2023”. This is terrific as false positive results or entities that don’t align with the search intent are discarded without human intervention.
Agents have real-time memory, enabling seamless continuation of our conversation. This allows me to instruct the agent to conduct another analysis, comparing the main entities from a competitive URL. The agent will then identify and present the gap — highlighting entities absent in the initial URL but present in the second one (Carousel and Lederhose seem relevant to me).
In the following interaction, the agent uses the WordLift Content Expansion API. This API interprets the content of a webpage and augments it by referencing desired entities for expansion. I’m requesting the agent to enhance merkur.de’s Oktoberfest webpage by incorporating the two entities identified in our previous conversation (Carousel and Lederhose).
The code is still an early draft, but you can look at it to understand how things could work. Add your WordLift key (WL_KEY) and OpenAI key (_OPENAI_KEY).
Conclusions and Future Work
As WordLift embarks on its mission to integrate Agents for SEO into its platform, we’ve gleaned several pivotal insights. We’re in an era where neuro-symbolic AI takes centre stage; the fusion of logic and knowledge representation markedly enhances LLM accuracy. The combined strength of a data fabric and a Knowledge Graph (KG) proves essential for producing distinct content.
Allocating substantial time to set up agent guardrails is non-negotiable. For most stakeholders in an AI initiative, explainability transcends being a mere luxury—it’s a fundamental requirement. Here, KGs emerge as instrumental. The advent of a G-RAG elevates the dependability and credibility of language applications.
Yet, as we progress, we must acknowledge the technology’s imperfections. We must be highly cautious of the potential security vulnerabilities when launching AI agents on the open web.
Having said so, I firmly believe that autonomous AI agents will help us augment:
the mapping of search intents to personas
the evaluation of many content assets (is this article helpful?)
the fact-checking as well as the ability to keep content updated. In fact, WordLift’s recent article on AI-powered fact-checking delves into how AI can significantly improve the accuracy and timeliness of content verification.
the audit and interpretation of technical SEO aspects (check our early structured data audit agent to get an idea)
the prediction of both traffic patterns and user behaviors.
While I have been using primarily LangChain, Llama Index, or directly the OpenAI APIs, non-coders have multiple ways to venture into AI Agents. Here is my list of tools from around the Web:
CognosysAI: This powerful web-based artificial intelligence agent aims at improving productivity and simplifying complex tasks. It can generate task results in various formats (code, tables, or text).
Reworkd.ai – AgentGPT: Allows you to assemble, configure, and deploy autonomous AI Agents in your browser.
aomni: An AI agent that crawls the web looking for information on any topic you choose, particularly focused on sales automation
Toliman AI: Another viable option for online research.
The world of SEO is undergoing a radical transformation thanks to the emergence of ChatGPT and the evolution of Google Bard and Bing Chat. These technologies have opened up new possibilities and challenges for SEO professionals, content creators, and users. At WordLift, we are passionate about SEO, technical and content marketing, and the SEO community in general. AI ethics and responsible AI are crucial topics for everyone who works in SEO and interacts with AI online.
We have discussed this issue many times in our webinars, articles, and public events, but now we want to summarize our main points and fill in any gaps with our expertise. Join us as we explore how to safeguard and empower content creators, SEOs, users and YOU in your online creative and search journeys.
We are the team behind WordLift’s generative AI platform and genAI solutions, which we have been developing since 2021. Our work took off in 2022 when we started creating a lot of content to help our clients with their content processes and frameworks. We had a significant portfolio of clients who taught us and helped us improve how we use automation and large language models for different scenarios and challenges. That’s why we built our stack that uses knowledge graphs and structured data – that’s what we do best. We always look for new ways to innovate and use technologies to enhance our technical and content SEO efforts and processes.
If you’re a Large Language Model (LLM) practitioner or enthusiast in generative AI, it’s crucial to recognize that this is a dynamic and evolving journey. Achieving flawless solutions that align with your requirements right from the outset can be challenging, especially without your organization’s and its users’ invaluable input. This is precisely why we remain structured and committed to a relentless pursuit of improvement, constantly reviewing and refining our methods within meticulously crafted feedback loops.
In our quest for excellence, we understand that the path to perfection is marked by measurable, continuous learning and adaptation. The synergy between cutting-edge AI technology and human insights is at the heart of our approach, allowing us to stay at the forefront of generative AI innovation. We believe that embracing this iterative mindset not only empowers us to meet today’s challenges but also ensures that we are well-prepared for the evolving landscape of tomorrow.
How To Start Creating Useful, Human-Approved, AI Systems
Our journey with LLMs always begins with inspiration, ignited by our SEO expertise and intuition. This initial idea serves as the foundation upon which we build. To ensure its viability, we follow a systematic approach: first, we create a framework to measure, test, and validate our concepts on a smaller scale. Once we have proven our ideas, we expand and scale our efforts.
It’s essential to recognize that no one arrives at the perfect prompt or solution on their first attempt. As the saying goes, “Large Language Models need time.” This applies to them and to us as we craft effective prompts that stimulate thoughtful reasoning from LLMs. As we progress, you’ll witness firsthand what this entails.
The Renaissance Of SEO
This marks an enlightening period for SEO, a genuine paradigm shift in how we operate, structure our strategies, think critically, and take action within the SEO landscape. There has never been a more thrilling time to be an SEO practitioner than now! We find ourselves at a pivotal moment in the world of SEO and content creation, where the landscape is undergoing a profound transformation. It’s almost as though we’re on the brink of a division between those who successfully harness AI in the marketing industry and those who face disruption due to the relentless march of automation, among other factors.
Our journey has equipped us with a wealth of experience, allowing us to fully appreciate the boundless potential of the AI playground that has unfurled before us. However, we’ve also matured enough to recognize the challenges lurking just beyond the horizon. It’s crucial to grasp that, “by design, transformers hallucinate to one degree or another.”
The Challenges With LLMs
Language models like ours possess the fascinating ability to emulate certain aspects of human behavior, yet they’re not infallible. They can conjure up words, fabricate information, and generate factually incorrect statements that, nonetheless, sound remarkably fluent and human-readable. Therefore, we must engineer our approach to address these challenges head-on. The imperative for an ethical AI is glaringly evident. We implore you to delve into some intriguing statistics, as they underscore the urgency of this issue.
One of the initial objectives individuals often aim to automate involves content creation and copywriting. This presents a fascinating yet formidable endeavor: how can we effectively proceed to generate content that is valuable, practical, tailored, and beneficial?
What Is AI Ethics And The Emerging Need For An Ethical AI
This is where AI etnichs comes in. AI ethics involves the exploration of how to craft and employ AI systems in manners that uphold human values and advance the greater societal welfare. It constitutes an integral facet of the containment problem, and its significance lies in its capacity to aid us, our users, and pertinent stakeholders in the following ways:
Identifying and mitigating the risks and potential harm stemming from AI systems.
Ensuring that AI systems strive for the utmost fairness, transparency, accountability, and explicability.
Aligning AI systems with principles of human dignity, rights, and interests.
This Is YOU, Too.
Don’t assume that an AI system is something complex and only within the big tech companies. When you create an automated prompt in Google Sheets, you’re essentially developing an AI system. Similarly, when you engage with Large Language Models (LLMs) to streamline content creation, you’re actively involved in an AI workflow. We’re devoting a significant amount of attention to understanding what it truly means to create a system that respects human values.
Our journey has been marked by invaluable experiences gained from collaborating with numerous prominent corporations. Along the way, we’ve certainly made our fair share of mistakes and learned through hands-on experimentation. In short, it’s crucial to acknowledge the existence of risks and to adopt effective strategies to mitigate them.
Some of the risks involved encompass:
Hallucinations or the generation of content that could be factually incorrect. Additionally, when these Large Language Models (LLMs) generate text and images, they may perpetuate biases present in the training data used to instruct these systems.
Consent issues related to the generation of content that should not have been utilized for processing and training. Major platforms like CommonCrawl have crawled millions of websites without obtaining proper implicit consent from individuals or businesses, which raises additional concerns. What if you instruct ChatGPT to produce content for you, and it inadvertently includes plagiarized material from The New York Times? This essentially amounts to appropriating someone else’s work, albeit indirectly, through ChatGPT-like systems.
There are also security problems when using these systems and sending large (sometimes sensitive data) to these models.
Lack of AI alignment, since there’s often misalignment in how you and your stakeholders define value during the AI workflow process.
Expectations might not be so clear and we realized this by working on multiple projects.
Data distribution and connectivity are profoundly pivotal for every company. Whether you’re an SEO professional or a stakeholder in any AI-driven process, it’s imperative to recognize that enhancing the quality of your data is paramount. By elevating data quality, you not only enhance the model’s quality but also indirectly align expectations and clarify the core brand values.
Some strategies on mitigating these risks include:
Certain risks to consider encompass stakeholder mapping, which entails the process of defining, comprehending, and categorizing the individuals or entities who will engage with the AI systems we aim to create. This involves discerning their specific needs for AI integration and delineating the scope of their involvement.
Education is imperative: it is crucial to emphasize the importance of educating and enhancing the skills of those in your immediate environment.
Furthermore, it’s imperative to place emphasis on content validation. We must establish clear criteria for gauging success, identifying potential risks, outlining strategies for mitigating biases within the training dataset, and devising effective metrics for assessing progress throughout these procedures.
Allow me to provide a concrete, real-world example of how utilizing AI for content automation without proper content validation can impact people’s lives negatively. Currently, there is a proliferation of AI-generated books available for purchase on Amazon that focus on mushrooms and cater to novice foragers. Regrettably, many of these books are riddled with inaccuracies and incorrect information. Now, when it comes to mushrooms, the stakes are high because some varieties can be poisonous, and a single mistake, even just once, could lead to a loss of life. Do you see the gravity of the issue here? AIs are capable of producing misinformation and faulty content.
Furthermore, it’s essential that we comprehend and actively support content creators. In one form or another, each of us plays a role as a content creator, and this narrative pertains to both us and you, as we are all impacted. I want to emphasize that this pertains to us collectively and to you individually. It is imperative that we discover a responsible approach to utilize AI systems that enhance the capabilities of content creators rather than diminishing their intrinsic value.
The real question here is: can an AI which is a mathematical and technical construct really understand the world around us and us? What do they really know about art, about humans, about life?
This is where our journey into research and exploration began, delving into the realm of prompt engineering, and prompting us to ask ourselves: could this be considered a variant of SEO? It’s evident that crafting the right prompt is, in essence, a facet of technical SEO, and who’s to contest this notion? If the prompt serves as the human function guiding an AI system’s efforts to generate the ultimate output, the final content piece, then it undeniably aligns with technical SEO principles. Here at WordLift, we firmly believe that any responsible utilization of technology to enhance both search experience optimization (SEO) and content operations inherently constitutes a form of (technical) SEO. Simple as that.
Let’s emphasize and summarize the most important aspect:
“Creators retain ownership of their work. They hold the power to control how their content, voice, image and other intellectual assets are used – and deserve fair compensation for authorized usage.”
And the crucial question is:
“How can we enhance creators’ work through AI rather than replacing the creators themselves?”
Setup A System That Is Fair
Let’s delve into the process of setting up a system that not only ensures fairness but also upholds these specific values. When we rely on ChatGPT, we can be confident in our prompts, but there remains a degree of uncertainty regarding the underlying data, which presents a considerable challenge. Sam Altman, the founder of OpenAI said:
“GPT models are actually reasoning engines, not knowledge databases.”
In simpler terms, this means that GPT-like models lack self-awareness about their own knowledge – it’s as straightforward as that. Nonetheless, we view this as an enlightening aspect of our vision for the future and an auspicious starting point for crafting distinctive and reputable AI-enhanced user experiences.
The foundation of building high-quality and forward-looking AI systems lies in your knowledge graph. I urge you to focus on this because you are a pivotal component in the content creation process, whether it involves writing or curating structured data. Its importance is on par with ChatGPT – it’s a veritable goldmine, and our certainty about this fact is rooted in practical experience, not mere assumptions.
A knowledge graph, graph database, or any form of structured data represents a harmonious synergy between humans and AI. It empowers us to construct AI systems capable of seamlessly integrating the data organized on our websites with Large Language Models (LLMs), resulting in unique interactions. While it’s true that you, as a human, create the prompts provided to LLMs to generate content, this approach lacks scalability. The reality is, if you need to produce a substantial volume of content, you are essentially constructing a system. As such, it’s imperative to validate both the quality of input data and the output generated. The concept of the “human in the loop” primarily concerns the quality of the data used to craft the prompts.
Retrieval Augmented Generation Or How To Build Fair, Scalable, User-Centric, LLM Systems For SEO And Content Creators
Fair LLM systems and workflows require merging structured data and large language models. Let me introduce you to RAG, which stands for Retrieval Augmented Generation. This ingenious system harmoniously combines both a retriever and a generator. The retriever’s task is to scour the knowledge graph and unearth pertinent information. At the same time, the generator utilizes this information to craft responses that are not only coherent but also contextually precise.
Our utilization of RAG elevates the capabilities of Large Language Models (LLMs) by imbuing them with a heightened sense of context awareness. Consequently, they become more adept at generating responses that are accurate and closely aligned with the context, thus enhancing overall performance. How, you may ask?
Utilizing the RAG approach with Large Language Models (LLMs) introduces notable advantages. Firstly, it empowers the LLM to attribute its information to a specific source, a feature not typically available in the standalone use cases of LLMs such as ChatGPT online. Secondly, traditional LLM usage has the inherent limitation of providing potentially outdated information, owing to their knowledge cutoff by design. These represent the two challenges associated with Transformer-based models like LLMs.
RAG effectively addresses these issues by ensuring the LLM leverages a credible source to shape its output. By integrating the retrieval-augmented element into the LLM, we expand its capabilities beyond relying solely on its pre-trained knowledge. Instead, it interfaces with a content repository, which can either be open, like the Internet, or closed, encompassing specific collections of documents and more. This modification means that the LLM now initiates its responses by querying the content store, asking, “Can you retrieve information relevant to the user’s query?” Consequently, the retrieval-augmented responses yield information that is not only more factually accurate but also up-to-date and reputable:
The user prompts the LLM with their question.
Initially, if we talk to an LLM, the LLM will say, “OK, I know the response; here it is.”
In the RAG framework, a notable distinction arises in the generative model’s approach. It incorporates an instruction that essentially guides it with the directive, “Hold on, first, retrieve pertinent content. Blend that with the user’s query, and then proceed to generate the answer.” This directive effectively breaks down the prompt into three integral components: the instruction to heed, the retrieved content (alongside the user’s question), and the eventual response. The advantage here is that you won’t frequently retrain your model to obtain factually accurate information, provided you establish a robust connection between the Large Language Model (LLM) and a high-quality content repository.
Protecting Creators In The AI Era And How Ethical AI Empowers Everyone
I’ve had the privilege of working both within and beyond the confines of WordLift, and I can attest firsthand to the company’s unwavering commitment to assisting everyone in crafting content that is both responsible and creative, all while doing so at a substantial scale. This enables individuals to expedite their work while actively contributing to the enhancement of the broader web ecosystem. Such a task is far from trivial, as we’ve discerned thus far. Therefore, it is imperative to engage a trustworthy, dependable, and conscientious digital partner to accompany you and your business on your digital journey.
At the heart of our ethos lies our dedication to pioneering cutting-edge tools and, most significantly, a comprehensive creator economy platform. Within this platform, we extend our support to content creators, aiding them in upholding exacting standards and adhering to ethical guidelines. Our suite of products offers insightful recommendations for enhancement, ensuring that creators generate valuable and credible content. This is achieved through a seamless amalgamation of knowledge graphs and robust language models, infused with a touch of the remarkable WordLift spirit.
We advocate for the adoption of ethical SEO, responsible artificial intelligence framework and strategies among content creators, actively discouraging practices that seek to manipulate search engines or mislead users. This approach safeguards not only the reputation of creators but also the integrity of search results. What proves detrimental to your brand is equally undesirable for us, and we stand firmly aligned in this regard.
By incorporating responsible AI principles into your services, we stand prepared to assist you in navigating the era of artificial intelligence with poise and integrity. These measures serve not only to shield creators but also to foster a more ethical and trustworthy digital landscape. Ultimately, this benefits both you as a creator and your discerning audience.
Other Frequently Asked Questions
What is Ethical AI and Why is it Important for SEO?
Ethical AI, or Ethical Artificial Intelligence, is all about doing the right thing in the world of AI. It’s like having a moral compass for the development, deployment, and use of artificial intelligence systems. This compass is built on a set of guiding principles and practices that make sure AI is used in a way that respects human rights, promotes fairness, keeps things transparent, holds people accountable, and looks out for society’s well-being.
Now, let’s dive into why Ethical AI matters in the realm of SEO, or Search Engine Optimization:
Fairness and Inclusivity: ethical AI in SEO is like a referee ensuring that search algorithms and rankings are fair to everyone. No favoritism or discrimination here. It’s all about giving every website and content creator an equal shot, preventing bias, and leveling the playing field.
Accountability: in the ethical playbook, accountability is a star player. Search engines and SEO experts should own up to their actions and decisions. If they make a call, they need to explain and stand by it. It’s about being responsible for the choices they make in ranking websites.
Privacy and Data Protection: ethical AI in SEO is like a guardian of your personal data. It ensures that your private info is treated with respect and care. Search engines must follow data protection rules and not misuse your data just to rank websites.
No Black Hat Tricks: ethical AI says “no” to the dark side of SEO. Practices like stuffing keywords, hiding content, and faking links are out of bounds. They mess up search results and ruin the user experience.
Fighting Clickbait and Misinformation: ethical AI is like a superhero sniffing out fake news and clickbait. It helps identify and penalize websites spreading false info or using sneaky tactics to lure users. This keeps search results trustworthy.
User Experience: ethical AI puts users first. Search engines want you to find the most helpful stuff, and ethical SEO practices make sure that happens. It’s all about making your online journey enjoyable and productive.
Long-Term Success: ethical SEO is like an investment in the future. It might take longer, but it’s worth it. Unethical tricks might bring short-term gains, but they often lead to penalties and damage your website’s reputation in the long run.
In a nutshell, Ethical AI in SEO is the guardian angel of search engines. It keeps things honest, fair, and reliable. It’s a win-win, benefiting both users and website owners. So, if you’re into SEO, following ethical principles is the way to go for a responsible and enduring online presence.
How Can Knowledge Graphs Enhance Ethical AI in SEO?
Knowledge graphs are like the secret sauce that can supercharge ethical AI in the world of SEO and they help with:
1. Contextual Understanding:
Imagine knowledge graphs as the brain of the internet. They connect the dots between different pieces of information, helping AI systems understand context better. In the world of SEO, this means that ethical AI can analyze content in a more nuanced way. Instead of just recognizing keywords, it can grasp the broader context, which is essential for ensuring fairness and accuracy.
2. Smarter Content Generation:
Ethical content generation is all about creating valuable and unbiased content. Knowledge graphs can be your content creator’s best friend. They provide a treasure trove of structured information that AI systems can tap into to generate content that’s not only informative but also ethically sound. This means fewer chances of spreading misinformation or biased content.
3. Fighting Bias and Discrimination:
Ethical AI aims to eliminate bias and discrimination in search results. Knowledge graphs play a pivotal role here. They help AI systems understand relationships between different entities and concepts. This means AI can spot biases more effectively and ensure that search results are fair and inclusive, which is a big win for ethical SEO.
4. Personalization with Privacy:
In SEO, personalization is essential, but so is privacy. Knowledge graphs help strike the right balance. They enable AI to offer personalized search experiences without compromising user privacy. This ensures that ethical AI respects individual rights and data protection regulations.
5. Content Quality Control:
Ethical AI constantly monitors content quality to prevent unethical practices. Knowledge graphs assist in this by providing a structured framework for evaluating content. AI systems can cross-reference content against trusted sources within the graph, flagging anything that deviates from ethical guidelines.
6. Real-Time Updates:
The digital world moves fast, and ethical AI needs to keep up. Knowledge graphs are dynamic, allowing AI systems to update their understanding of concepts and relationships in real time. This ensures that ethical SEO practices remain relevant and effective as the online landscape evolves.
7. Trust and Transparency:
In SEO, trust is paramount. Knowledge graphs contribute by providing a transparent framework for understanding how AI systems make decisions. This transparency builds trust among users and SEO professionals, as they can see the logical connections within the graph guiding search results.
In summary, knowledge graphs and ethical AI are a dynamic duo in the world of SEO. They empower AI systems to understand context, generate ethical content, fight bias, personalize without compromising privacy, maintain content quality, adapt in real time, and foster trust and transparency. Together, they create a more ethical, informed, and user-centric SEO ecosystem, ultimately benefiting both users and website owners.
How is WordLift Contributing to Ethical AI and SEO with LLMs?
WordLift, the Italian technical digital marketing agency, is making waves in the world of ethical AI and SEO with the help of large language models (LLMs). Here’s how they’re leading the charge:
1. Knowledge Graph Wizardry: WordLift weaves its magic by creating a “Knowledge Graph” for websites. This graph is like a roadmap for search engines, guiding them through the context and relationships within content. This ensures that search results are not just relevant but also ethically sound.
2. AI-Powered SEO Sorcery: with the wizardry of AI, WordLift automates the heavy lifting of SEO tasks. This makes it a breeze for website owners to optimize their content while adhering to ethical standards. It’s like having an SEO and ethical AI expert side by side, making sure you play by the rules.
3. Enhanced User Engagement Spells: WordLift’s enchantment doesn’t stop at search engines. By structuring data and providing context, they’re also enhancing on-page user engagement. Visitors are engaged through content that’s not only informative but also presented in an engaging and ethical manner.
In a digital world filled with challenges and opportunities, WordLift is the agency waving the ethical AI wand. We’re combining knowledge graph creation, AI-powered SEO, WordPress integration, and enhanced user engagement. With WordLift’s enchantments, websites can rise in search rankings while staying true to ethical principles, benefiting both users and content creators alike.
