Introduction

My name is Yana. I am one of the co -founders at the company called Yai.

And I'm just setting my alarm so that I don't go overboard with the time because I know Stefano will restrict.

What Yai Does: Turning AI Drafts into Trusted Content

So at Yai, what we do, we transform AI drafts into trusted content. So we basically work with the teams that use AI heavily for content production. And typically they come from regulated industries.

and we help them through making sure that this content that they produce is trustworthy it's verifiable and also it's suitable for AI search and today

I'm going to show you from the very beginning you can see I'm logging in and I beg that it works but of course Google needs to throw an extra authentication at me just when I'm doing the demo anyway just confirming that's

me logging in from this beautiful place and I also want to thank you all for choosing to be here today to learn instead of maybe going for a pint in a pub and I understand it's hard to compete with a beautiful afternoon like this so thank you for being here right so this is the

Product Walkthrough: Content Lift Interface

interface of our tool which is called content lift it's called like this because we basically help to leave the content to the point where it can be usable and what we do here we

can either copy paste text or upload the file and I'm just gonna drop a file quickly it's from one of the banks out there nothing confidential just something I lifted from the website and the next step for me is to choose which metrics I want to work with during this evaluation process and there is a bunch of them but

Metric 1: Sources & References (Fact Checking)

today I'm going to focus on our key metrics which are called sources and references and AI search optimization so sources and references effectively a fact -checking metric so what we're going to do here we are going to fact check this document for facts and sources that have been used in this document and then

we are going to optimize the same document for AI which means LLMs so a lot of companies want to come up in searches when people are working with ChatGPT or Gemini or the likes of this LLMs as opposed to just coming up in Google search.

So this is basically SEO for AI age. And the reason I just selected these two metrics and the reason why we developed them in this

tool and why they're linked is that effectively LLMs are favoring fact dense content. LLMs like when we have facts that are verifiable and that could could be actually traced to credible sources.

So after we developed the first metric, which was fact checking a while back, yes, it was just last year, but it feels like it was ages ago,

we realized that AI search is the next frontier and it's very logical because of how LLMs like this content to be. And I'll show you and you'll see it in a minute.

it so what our tool does it's an agentic tool so which means there are multiple agents running on the background working with the content piece so the moment I put it in and I'll just beautify it slightly so you can see it's a bit better and you can see the structure with all the h1s h2s and so on so what

the agents do on the background they work through the document understanding in what are the sources, what are the facts in this document, and you can already see them being highlighted here, just outlined really, and they are verifying these facts.

So the way they're verifying the facts, they're searching the body of knowledge, in this case it's the Internet, they're searching the Internet for these facts to see what are the sources that will either confirm or deny what we're saying in this content piece and I will show you in a

moment what are the real facts what has been highlighted as potentially being false so this is the whole idea that we will have it here and another agent is checking the sources so once we were able to determine that the fact is

correct we need to trace it back to a source that will actually confirm it or the source that will deny it and because it's an agentic system similar to what Emma was showing before it does take a little bit of time so like you would work with Claude or another agentic system you will typically give it a task

and then maybe switch to do something else or go and make yourself a cup of tea But I can already show you some examples, which is great.

So the first example is of accurate fact. And in case of accurate facts, there is not much to see here apart from the claim analysis and the sources.

So the recommendation will be that there is nothing you can do with it. It's accurate based on provided results. But if you wanted to double check, you can go into sources and see what sources confirm this fact.

so in this case it's an article from a third party and in case the facts are inaccurate we can actually we here we have some options first of all we

analyze this claim and in this case the claim is that Trust Bank offers customers a free fractional stock worth up to five hundred dollars not five $5 ,000 as is stated here it's a mistake that I intentionally planted so to give credit to the bank they didn't have this as a mistake on the website but for the

purpose of this demo that's what I've done and straightaway the tool highlights this saying that this is not correct the official information is that this stock needs to be worth 500 not 5 ,000 and recommendation is to consult the official website to get up -to -date, most accurate information.

It also gives me the sources. Again, I can go and check that information on the third party website or on the bank's website itself.

And most importantly, and unlike many other fact checking tools out there, it gives me a correction. So all I need to do is accept this correction and straight away the sentence gets amended to show the correct figure.

It can do it with any inaccurate facts. Again, I can just review it, the claim analysis, the source and the correction and accept or reject it. So the power is always with a human decision maker in this case, which is important when we talk about trust.

Even in the HVI, we do accept that there there must be a human in the loop it's important at the stage where we are at the moment but i'll accept this change and i'll save and continue because we have another interesting part of the

Metric 2: AI Search Optimization (SEO for the LLM Era)

demo which is our second metric and in this case this is ai search optimization so because we now have accurate facts in this article they are all verified they all have correct sources we can now

work further with this little piece of content of course it's not as long as typical piece and that's for the purpose of the demo I basically made this article much much shorter otherwise we would take much longer than 15 minutes

and Stefano will be not happy but you can absolutely work with full -blown articles which are thousands words long or even longer it will just take a a little bit more time because the agents will be running fact checking on more sentences, more facts. And with AI optimization, it will also take longer to optimize longer text.

But here, what I

start with is choosing the main prompt or the primary question that this content answers. I have options. If I don't know what exactly I want the content to answer, I can just pick one of those and this will be can I trade fractional US stocks and ETFs

through Trust Bank because this is what the piece is about that makes sense you can select multiple target prompts but again I'll just go with one for now to speed things up and right now the tool is doing its magic again so we have a

bunch of agents analyzing the semantic clarity of the text checking whether or not there are direct answers evaluating content structure and headings and so on basically looking for the blueprint that we know that llm likes and favors when it comes to written text which helps to surface this text through llm answers when we are working with llms prompting them to give give us information in this case for example about a fractional trading of u .s stocks if we were

talking to ChatGPT or Gemini, as a brand I would like my content to come up in the searches because it's not enough anymore to come up in Google on page one because more and more of us are working with LLMs and this is only beginning of this trend.

We still have less people asking LLM questions than asking Google questions but it's not there for for a long time.

Readiness Score and Evaluation Blueprint

So the AI search optimization is ongoing and you can see here that first thing we do we give you an AI search readiness score which in this case is 53%.

The golden standard is 100 % but we do not suggest that it's necessarily the purpose of going going through this process.

We regard 70 % as a good enough optimization and that usually is the purpose of going through this process and it's very quick because we have the blueprint which I mentioned and you can see it here.

So there are five main areas that we are checking the content against. It's prompt alignment, it's snippet strategy, engagement and clarity, brand authority and AI retrieval optimization and under each of them we have a number of areas

that either pass or fail and we also have a number of areas that may need attention.

So when this happens we also provide the checks and of course this being live demo not everything has worked but I did think about it and if it doesn't give me what I want in a minute I will go to my plan B and show

you how it needs to look because what's important is not just knowing what's wrong with it it's important to know how to actually fix this and when we do fix fix things.

Applying Fixes and Re-evaluating the Content

And now I'm jumping to my plan B. So when we know how to fix things, we basically provide the fixes.

So fixes, you can see here on this slide, I was working with the same text, so it doesn't change anything materially. But when I go through the checks, I understand what's wrong.

And then I understand how to fix it because I'm looking at the fixes and and I can accept them one by one, or I can accept all of them in one go.

So if I click on the fix again, similar to what I was showing you in the fact -checking flow, it will allow me to understand what is being fixed.

And the text is effectively changing a lot on the left -hand side. You can already see that this looks quite different from what we had in the in the app here so this is a much shorter text it doesn't have sections like too long didn't read or key takeaways or anything

like that so this is what we know LLMs want and we basically provide it through the series of fixes to this content that we are suggesting so again you can see here we introduce the frequently asked questions section and the ending and beginning we have the too long, didn't read.

So once we went through this and accepted the changes, again we can reject changes, what happens is we re -evaluate the score and for that I just click the button re -evaluate it and in this particular case it went up from 52 % to 72 % which is a much higher readiness score it's maybe you know not up to some someone's standards and you can by all means spend more time editing it which is which is fine but it's much better than the baseline that that we had originally and after that I

Exporting, Continuing Edits, and Human-in-the-Loop Control

can basically accept what the tool has created and either download or export it to Google Drive and keep working with this text where I need it and it's also

important to mention that I can always make changes on the left -hand side here so once I have my text but I identify what else I might want to do to it because I had the ideas from what it's showing for me here things that you know could be done better like concise answer focused formats or things like that I can as a human editor just go in almost on time and make changes here so I

actually did set myself a buffer for 10 seconds so there we are thank you so So there we are, we have our optimized content piece.

Conclusion and How to Try the Tool

We can go ahead and publish it on the website or continue editing it and I thank you for your attention.

You can connect with me, you can check the app, it's very simple, you go to app .yai .digital and there is always a free trial that I encourage you to have a go at and you can contact me

either on email or on LinkedIn I'm very happy to connect I'll give you a second to take snapshots of this because my surname is long you're not gonna remember it or put it very quickly in the search browser so yeah happy to chat

afterwards happy to take questions afterwards thank you very much