A brief analysis of a web page not based on artificial intelligence.

[fernando-jose-silva]

Is your feature request related to a problem? Please describe.

Imagine that a relatively inexperienced user opens a page that never opens, for example, the Google page.

They can logically navigate with the arrows and get to know the site; however, beginner users have little patience, and after a few arrows down, they consider that they haven't reached their goal and give up.

Now imagine other scenarios, pages like Gmail, GPT chat, news.

Describe the solution you'd like

My proposal is that when a new page is opened, NVDA performs a small analysis of the page, analyzing the types of controls and where they are located, and suggests some ideas to the user:

Examples:

On Google, on the results page, it would be something like:
"Page with several headers at level 3, it might be interesting to navigate through them."

In Gmail, in the inbox:
"There is a large table, each row of the table has a selection box, if desired you can investigate if there is useful information."

In GPT chat:
"If the user is in the edit box, ask something:"
"The tip could be:
"Above the current focus there are several headers alternating between levels 5 and 6, you can return to them to investigate with xshift 5 and shift 6."

This is considering that NVDA does not know the context; if it makes inferences from the URL and text, this could be much more powerful.

This is without using AI, which isn't readily available in NVDA yet; with AI, it becomes much more powerful.

Describe alternatives you've considered

No response

Additional context

I'm not a programmer, I'm venturing into requesting AI functions, testing, and modeling functionalities to run on my NVDA.
Proof of concept.
I couldn't figure out how to manipulate focus, nor how to discover the types of elements on the current page without breaking NVDA.
So I developed a script that periodically searches for a newly added text block on the current page, analyzes it, excludes some blocks, and reads it to the user.
And with this rudimentary approach, I achieved interesting results in the GPT chat and automatic reading of messages being written on the screen before the final result, automatic reading of at least a portion of articles displayed on a page, and information that new items will appear on the page even when the page doesn't inform NVDA of anything. I'm posting the rudimentary script written by GPT and refined with my tests and requests for refinements. Remember, I don't want my script to be integrated into NVDA, but rather to show that something more powerful can be done by real developers.

init.py

[seanbudd]

This issue shares similar problems with #19200 (comment), see that comment for a suggestion on how to proceed