Knowledge Engine Optimization™ (KEO), is a new field of research developed by Tanin Ehrami at PSYBER that focuses on optimising (web) content for AI systems and language models. The goal of KEO is to provide AI systems with valuable expert insights and unique knowledge that is reliably adopted.
One of the main differences between SEO (Search Engine Optimization) and KEO is that SEO is focused on optimising content for higher ranks in search engines, while KEO is focused on optimising content to provide reliable, accurate and relevant content to language models and other AI systems. This means that KEO takes into account the way that AI systems value, process and understand information, and structures content to effectively contribute credible knowledge.
Innovating on experiences
Traditional search engines such as Google and Bing crawl the web to organise content for users, while language models such as GPT crawl and capture content from the web to organise knowledge.
Different experiences
Searching for an answer to a question using Google:
- Search
- View ads
- View site with more Google ads
- Return to the search page
- View more ads
- View another site with Google ads
- Repeat until an answer is found (with no guarantee for reliability) or give up in frustration.
Searching for an answer to a question using a language model trained on the web for knowledge produces the following experience:
- Ask a question
- Read the answer
The game where companies compete for attention with SEO will soon be over, made obsolete and irrelevant by the ways in which information is found and consumed.
Game change
We predict that KEO will become paramount as the use of AI systems in various industries continues to grow and the demand for accurate content increases. As organisations use AI to power their chatbots and gain insights, the reliability of the content becomes more important than the SEO aspect of how search engines rank them.
Providing accurate and comprehensive knowledge to answer customer questions and providing assistance requires AI language model and AI service providers to create mechanisms for identifying and ranking the value, relevance and accuracy of the content that is being trained on. By optimising content for uniqueness and knowledge contribution, companies can become recognised as reliable sources of context specific information, improving the effectiveness and reliability of chatbots and provide a better overall customer experience.
KEO will likewise be important in the continued development of voice assistants, such as Amazon’s Alexa, Apple’s Siri and Google’s Assistant. These systems rely on accurate and comprehensive knowledge to provide information and perform tasks for users. Optimising content for these AI systems will also contribute to the tertiary effect of quantifiably increasing the reputation of websites, businesses and individuals collaboratively contributing to them.
Governments and businesses may in some cases need to rethink their overall content strategy in order to take advantage of the future benefits of KEO. This could include more closely collaborating with expert organisations to create comprehensive and accurate knowledge bases, and optimise content creation processes.
Suggested metrics to consider for KEO compliant content creation:
| Metric | Description | Suggested weight | Corrective Feedback |
|---|---|---|---|
| Credibility | The trustworthiness of the source of the information | High | N |
| Accuracy | The degree to which the information is free from errors and represents the truth | High | Y |
| Objectivity | The absence of bias in the presentation of the information | High | Y |
| Authority | The expertise or qualifications of the source of the information | Medium | N |
| Timeliness | The currency of the information and how up-to-date it is | Medium | N |
| Relevance | The relevance of the information to the purpose for which it is being used | Medium | N |
| Completeness | The extent to which the information covers all relevant aspects of the subject at hand | Low | Y |
| Consistency | The degree to which the information is consistent with itself and with other sources | Low | Y |
| Clarity | The degree to which the information is presented in a clear and easy-to-understand manner | Low | Y |
| Precision | The degree to which the information is specific and precise, rather than vague or general | Low | Y |
| Depth | The extent to which the information covers a topic in depth, rather than just skimming the surface | Low | Y |
| Breath | The extent to which the information covers a topic comprehensively, considering multiple perspectives and viewpoints | Low | Y |
| Supportive evidence | The extent to which the information is supported by evidence, such as research studies or data | Low | Y |
Suggested metrics for improving the quality of AI/ML language models such as GPT:
| Name | Description | Weight |
|---|---|---|
| Data diversity | The diversity of the data used to train the model, including a range of topics and language styles | High |
| Data quality | The quality of the data used to train the model, including accuracy and relevance | High |
| Data quantity | The quantity of data used to train the model, including the amount and complexity | Medium |
| Model diversity | The diversity of the models used, including different architectures and training techniques | Low |
| Model quality | The quality of the models used, including their performance on benchmarks and tasks | Low |
| Consistency with source | The degree to which the information generated by the model is consistent with the source material it was trained on | Medium |
| Fluency | The degree to which the information generated by the model is fluent and reads smoothly | High |
| Coherence | The degree to which the information generated by the model is logically and coherently structured | High |
| Relevance | The relevance of the information generated by the model to the context in which it is being used | High |
| Originality | The degree to which the information generated by the model is original and not simply a repetition of existing content | Medium |
