When saving as a JPEG, you should let the image itself dictate the quality rather than reusing settings. The same settings for one image may produce significant loss on another.
Basic Settings
Quality: the overall quality of the output JPEG. The lower this value is the more compression you'll attain, at a cost to visual quality. Generally you'll want the lowest quality with no significantly noticeable loss. Start at the default and adjust to the best compromise for the image.
You shouldn't usually go above 95, as the size grows significantly for no noticeable quality gain. Likewise, you shouldn't go below ~25 as banding may occur.
The “Show Preview in image Window” option creates a new layer with how the image will look with the current settings. It updates in real-time as you change the settings. You can zoom or hide as usual to get a better look at the quality.
Advanced Settings
These can be uncollapsed by selecting the box next to “Advanced Settings” and are saved for the current session, along with most other saving settings.
Lossy
These settings alter the actual image to help with compression. Changing these will change the overall subjective quality of the image.
Smoothing:
Setting the smoothing option to a non-zero value will smooth out the image slightly. This reduces fuzzy artifacts from compression, and helps with the compression. A setting of 0.10-0.15 removes a good portion of the artifacts without smearing edges.
Subsampling:
Gimp allows three subsampling modes.
Subsampling discards more colour information (which is harder to discern) compared to luminance.
Subsampling improves image compression at a cost to image quality, sometimes a significantly noticeable loss (such as on red), or no noticeable difference at all. Subsampling on an image where it shouldn't be used will smear the detail, most noticeably on sharp edges. If you notice significant loss with the default, you can turn it off by selecting "4:4:4", which is technically no subsampling.
DCT Method:
The 'Floating' DCT method produces slightly better results than the 'Integer' method with a slight cost to speed. 'Fast Integer' should only be used where speed is imperative.
(DCT) Discrete_cosine_transform Discrete Cosine Transform is generally a mathematical method that helps reduce the amount of unnecessary detail and aids in lossy compression methods.
Lossless
The following settings can losslessly be set; they don't affect the image's quality in any way.
Optimize:
Optimizes the images table for a gain in compression. Unless speed is a concern, there's no reason to turn it off.
Comment:
If you wish, a comment of your choosing can be set for the image, such as the quality settings, or a synopsis of how the image was made.
Save EXIF Data:
Some cameras save information about the image, such as when it was taken or even if the flash was on. If you have no need for this information in the image you can safely turn it off.
When saving an image that has no EXIF data present the option will be grayed out.
Save Thumbnail:
The JPEG format allows for a thumbnail of the image to be saved into the actual file itself. Most software using a thumbnail can generate one of its own and saving your own is unnecessary, as well as adding several kilobytes to the image size.
Progressive:
Selecting Progressive will change the encoding to display the image at increasingly higher quality levels until the image is fully loaded. Progressive encoding also benefits the image's compression.
Leaving this option unchecked will switch to Standard encoding, where the image is displayed in rows from top to bottom.
Restart Markers:
Restart markers are useful when data corruption can occur; the image will only be corrupt up until the next marker. Setting this higher increases the frequency of restart markers at a cost to filesize.
Without restart makers, if any corruption occurs, the entire file can be compromised.