Would love to see a full website search implemented into Elements. Automated not manually curated.
1 Like
Yeah, it would be nice to see something like Pagefind https://pagefind.app built-in to Elements.
Pagefind can be used as a standalone, binary (command line) executable as the app itself is written in Rust, is open source and can be run without npm/node. The majority of the work (beyond integrating the binary itself) would be creating a customizable Pagefind component for the search UI that would be flexible enough to meet most users’ needs.
Just out of interest sake @dan, any chance 3rd party components will be able to include binaries such as Pagefind, or in the case of Node/NPM dependencies, the ability to run external commands (if the dependencies are installed) to extend the functionality of Elements? For instance, executing the deploy command for Cloudflare Pages via wrangler to automatically deploy a site to Cloudflare once it’s been built/published. (S)FTP is so retro. 
1 Like
Yeah with a blog and a CMS to take full advantage of its features 
2 Likes
I know you guys are probably tired of reading my GenAI comments, but here is another use case, in my opinion, for a GenAI component specifically tailored and tuned for searching the website. In my opinion, it is a great example of a living RAG component that will assist people with finding answers on your site, not just another page link. Think, Perplexity for your specific website.
1 Like
The main reason for search to a link, beyond helping the reader learn more is Site stickiness. The more pages a site visitor hangs around to read the higher your site ranks. I don’t want AI generating answers that can not be edited by me first and indexed and by the Search Engines.
A GenAI search component would not stop search engines from indexing your site.
If you want stickiness, do you really want a search engine since they helps people to skip your other pages and potentially go the an only a page with the information they want? Wouldn’t stickiness mean you want them browsing around?
If we don’t provide a great information finding tool for our web sites, third parties (e.g. Perplexity) will do it and they won’t go to our site at all.
Don’t people come to our websites for information? If they don’t find it quickly they leave.
Wouldn’t a website that gives you your answer quickly and accurately something that is likely have repeat visitors?
Hopefully, a component for Elements that gives us a GenAI answer engine would be tunable and will present answers in a way that serves both the visitor and the owner of the website.
Lowes and Home Depot (in the USA) are two sites that I expect heavily depend on their search engines. They both have similar pricing and products. Assuming they are both giving good service, if one provides me with a GenAI search tool to replace the current legacy style search, I will gladly use the newer one. I generally hate Amazon, Walmart, Lowes, Home Depot, etc site search engines. Better than nothing, but not something to write home about.
1 Like
Well, I think there is a bit of a misunderstanding here.
I want search engines to index my polished content. An AI-generated answer that is a temporary response rendered in the browser DOM is not indexable and has not been reviewed for accuracy by the webmaster.
Second, website stickiness refers to users staying on my website and interacting with its data, creating logs that search engines can read. An AI-generated response in the DOM will not register as part of my website’s data, so it will not contribute to metrics like time on site, which build stickiness.
Good, fresh, and unique content presented in my voice, with well-researched information that visitors expect, will naturally create stickiness for most visitors who have the time to stay on the internet.
Regarding product websites like Home Depot and similar stores, I don’t think Elements is the software package people will choose for creating such a retail store. I see it being used as a store platform, yes, but more likely for mom-and-pop shops rather than big box stores. Could AI be helpful in small stores? Yes, probably, but how do we train website visitors to create prompts that generate better results? What is simpler than a keyword or keyword phrase?
Just my thoughts. Loving the conversation. 
I think you are right that there is a bit of confusion here.
I’m not suggesting that the AI supply the content and thus not provide your polished content. I’m suggesting that it provide a better answer using your content. In fact, a good component will enable you to provide only a link to the exact page you want the person to go to. However, there are two sides to a search. Traditional search relies on key word search. It sucks. A GenAI search engine would be able to take natural language chat and know you website so that it could respond correctly. Sort of the difference between chatting with ChatGPT and trying to chat with Siri.
Nothing prevents you from building log data from GenAI conversations if you are in control of the chat. I’m not sure how a local search component is going to help your SEO. The search engines (not your local search) crawl your site and build their own index according to their own algorithms.
Yes! Good, fresh, and unique content presented in your voice with well-researched information is exactly what your visitors expect. How is does a local GenAI trained and tuned component prevent that? I argue that it helps you and your visitors to find the amazing content while they are on your website.
Regarding Home Depot and similar, I never suggested they would be using Elements or anything like it. I used them as an example of a very powerful legacy website search engine. The kind that makes it very hard to find exactly what I am looking for and an if they had a powerful natural language GenAI that knows their inventory (that amazing content) then it would much more helpful than the current system.
Why would you train website users to create prompts? You, the website creator, would train the AI how how to search your site, how to respond, what to respond to, what links to return or not return, etc. The natural ability of the GenAI to understand a visitor’s query (in most languages of the world) would just magic to them. But you are in control.
Let me use another hardware store example of the difference a GenAI search component can make over a keyword one. Say I have a clogged drain and I’ve heard that I have to be careful what chemical drain cleaners I can use so I don’t hurt my pipes. I could simply tell a GenAI search, “I have a clogged drain and my pipes look like they are plastic. What drain cleaner can I buy to use?” A keyword search will require me to look know about PVC, drain cleaners, reading through all of the descriptions to see that is safe to use on PVC for each one in the link. It is a pain in the a**.
Maybe not the best example, but I hope the meaning makes it to you.
And I love the conversation! I wish more would do this. Thank you.
Agreed, but I suggest building more top notch quality content is a better use of my time and will create better results than me investing lots of time teaching my AI how to respond. UNLESS this is automatically created, then I see the possibilities so long as the results sends them BACK to a linked page on my website.
Also if the AI could report to me a history log of what the site visitors were looking for, that would give me the opportunity to create new or tweek existing content.
This is a good example, yes. But I suspect for a very small percentage of Elements’ user base.
Can you provide examples of any of these ideas working today?
You must not be using ChatGPT, Claude, or some other LLM today…at least that is my assumption based on your comments. I use GenAI tools all day long. I assure you that there would not be the time cost you are suggesting to set up a GenAI search component.
I would suggest–if you it available to you–go to Perplexity.AI and use that search for a bit instead of Google, Bing, etc. If you need tips on how to use it, perhaps we can set up a more direct method of communications…but I suspect you’ll be fine with the provided examples. That is an example of a GenAI tool trained as a generic internet search search engine.
I use ChatGPT for about 5 hours per day. NOT for search. I have no issues finding exactly what I need with DuckDuckGo.
What Exactly is involved with training the AI for search on a small website? If you have a sample of a website, simple basic website, that uses AI, that’s what I would like to see. Not something that has a huge team and budget to put it all together.
What I’m asking you do to is put on a salesman’s hat and sell the concept to me. Prove the ROI works.
I used to use DuckDuckGo all the time. Now I use a mixture of Perplexity and ChatGPT. The difference is remarkable.
It is good to hear that you use ChatGPT. When OpenAI introduced custom GPTs I built one for a local college website to test its ability. It took me an hour to build in several features without any help from the college. It was directed to only use the information to be found on the college’s website, never making up information, always providing links to the source information, only responding in a way that reflected the religious college’s Christian values, etc.
They were heart broken when I deleted it after letting them play with it for a few weeks. But I was not going to maintain it so I gave my training to them and recommended that they create their own. I don’t know what stage they are with making their own.
Because the GPT at that time required you to have a paid OpenAI subscription, they only used the one I made internally with their recruiting team to get answers to questions from prospective students and parents. I’ve not had cause to build another one.
I did build out a custom GPT for a manufacturing company that they use internally for their support personnel. I loaded it with their extensive installation guides, user guides, technical documentation, parts lists, etc. While not a web site, it is a search over specific information. Their response time to their users called in questions about just about anything dripped from around 30 minutes average to less than a minute. The difference is the ability for the GenAI to understand and infer meaning behind a query and match that up with the best answer(s) in the defined knowledge (web site or a bunch of PDF and Word documents).
I’m not going to sell you on a need you don’t see. I will tell you that natural language understanding and the ability to infer meaning and context is very often much better than keyword search, in my experience.
I can easily imagine an Elements component that lets you choose the LLM of your choice and pre-engineer it to respond as you like for getting answers from your website would be very useful–and not require a team or huge budget. But it would empower a single person to provide what looks like a huge team/budget to put together thus greatly increasing the value that a Rapidweaver/Elements developer can provide to clients.
Sorry, I don’t have specific live examples that I can feed to you. I would say build your own. Create a custom GPT. Limit it to a very specific website…maybe one that you made and know intimately. By the way, why would a simple basic website have any need for a search engine? I suppose if it has a large blog component, but then it is no longer small, just not a complicated design.
Sorry guys for intervening in your conversation, you know i can’t help but say stupid things: @jscotta you could make one about the weather? Because i don’t know about you but here it’s all wrong all the time for many months. Notice it’s interesting to watch our media announce a lethal heatwave for next week every week for three months. 4 days ago we were told 29°C the day before yesterday, almost all good. weather forecast : we had 19°C (sorry for those who are really experiencing excessive heat at the moment). So I tell myself that this may be an opportunity for your business. 
@jscotta More seriously, could you share your or a way of training a bot that seems more relevant to you than the others?
Oh dear really its 48°C to 53°C hear we have had no warnings in Saudi Arabia and this is just normal weather. LOL
2 Likes
[offtopic]
Ah, but that’s dry heat. Still no fun and dangerous, but very different.
Over here in north western Europe, 29C is humid heat due to our sea climate. The air becomes so humid that sweat no longer evaporates and you run the risk of overheating quite easily. And we’re not used to it either, so that’s an added factor.
I can usually stand the heat in southern Europe much better than the heat over here in the north west.
[/offtopic]
Cheers,
Erwin
1 Like
Good morning, @Bruno ! I take exception. Guaranteed that if we polled readers, I would come out on top for stupid things posted, in this forum! 
I’m not sure I understand your question. Are you asking about how to prime a chatbot, in general? How I prime a specific chatbot? How I use RAG (Retrieval Augmented Generation) with my primed chatbots?
I’d love to help if I can.
Scott
[offtopic]
Yea, in Austin, Texas (lived there for nearly 25 years) we had summer weather regularly up to about 45C with 90-95% humidity. My wife, who was living on the coast of the Gulf of Mexico as my fiancé, thought she knew what hot was until we got married and she moved to Texas.
1 Like
@jscotta Hi, Thanks a lot for your response. In fact I tried to use a fairly “traditional” methodology for a brand new model 
. But I find the use of elasticsearch and Bert complicated to implement. I did not find a Workflow satisfactory in terms of preparation time. If it’s not a top secret recipe I’d like to know how you do it. BTW is it so obvious to use elastic search and Bert for this purpose? (Psychologist psychological doubt 
)