AI on the web
A general language model does not know your terms and conditions, price list or guides. A chatbot over your own content first finds the related sections of the website for a question and only then asks the model with them – the answer is then based on your texts and can link to them. The approach is called RAG (retrieval augmented generation). The example needs nothing more than PHP and MySQL on ordinary hosting.
A visitor searches for “getting my money back”, but the article is called “Withdrawal from the contract”. An ordinary word search does not find it, a search by meaning does. How to calculate article embeddings, store them in MySQL and use them for searching as well as for “similar articles”.
Alternative text (the alt attribute) is read to blind users by their screen reader and is displayed when the image does not load. On older websites it is often missing for hundreds of photos. AI can suggest it – how to do it in bulk and why the suggestions need to be checked by a person.
Language models translate fluently and can shorten a text into a perex or a description for search engines. On a website, however, the translation must not break the HTML or change the links, and the summary has to fit the length. How to prompt for it and how to check the result in PHP.
Comments under articles bring spam, advertising and insults. AI can pre-sort them: approve ordinary comments right away, send disputable ones for review and set obvious spam aside. Leave decisions with consequences to a person, though. How to set it up in PHP and MySQL.