Screenshot is only useful if it is about Emoji, Images or Mermaid charts rendered by ChatGPT,
in all other cases the text is better. Why? Because it’s indexable.
How I came up with it
If we’re crawling URLs, we can as well analyze their source code.
More insights
This will most likely always disappoint with GPT-3/3.5 from the first tests.
I’m so used to looking at source code that it doesn’t bother me.
Neither one made any reference to a robots.txt or xrobots headers (or meta robots), and I’d certainly say that making sure they did (as it does pertain to data security) would be an improvement.
I’m not web design novice, so I’m curious what other “aspects” would be useful other than “security, speed, and comments”. Maybe a list or queue for common aspects of websites that would be interesting to other professions might be a cool feature to make this tool even more powerful for the average user.
Absolutely not. This is like having a set of bathroom scales to test your weight, and someone suggesting you could use that to diagnose illnesses, in place of MRIs, CT scans, Ultrasound, and years of study.
I have conducted experiments using various combinations of three variables, as well as permutations involving no variables or 1 to 2 variables. However, one of the major factors that requires attention is the issue of crawlability. This variable has a significant impact on the outcome and is strongly correlated with source code length and quality, among other factors.
In my opinion, if the source code is excessively lengthy, it may result in an incomplete analysis due to throttling or capping, leading to a compromise in the quality of output and results. In summary, the prompts I have used have proven to be effective within their crawlable limits. However, exceeding this limit results in a margin of diminishing returns as the program is unable to access anything beyond the sum of its parts, rather than the sum total.
