Ad fraud has dropped a bit.

That’s the good news from ad intelligence provider Integral Ad Science  (IAS) in its most recent quarterly Media Quality Report about ads in the desktop and mobile web.

Ad fraud dropped from 14.1 percent of all ads to 11.2 percent. The company said that these fraud numbers include non-human bot traffic as well as any impression that couldn’t be viewed by a human, such as ads that are stacked on top of each other.

This is the overall picture. When it’s broken down, it shows that networks and exchanges have significantly different fraud levels than publishers. The level is 14.1 percent for networks and exchanges, a slight dip from 16.5 percent in Q1. For publishers: 3.7 percent, a snip down from 4 percent.

Publishers can have more control over which ads appear, senior vice president Kiril Tsemekhman told me via email, while networks and exchanges often employ programmatic platforms that have “a greater risk of fraudulent impressions.”

And then there’s the not-good news.

Ad viewability is down, from 49.4 percent to 44 percent of all ads served. This refers to the Media Research Council’s minimum viewability standard that half of the ad’s pixels have to be in view for at least one whole second.

In other words, according to Integral, 56 percent of all online ads are not on the screen completely enough or long enough for actual humans to see.

On networks and exchanges, viewability is 39.9 percent (down from 41.8 percent in the Q1 report) and 50.1 percent for publishers (compared to 52.8 percent in the last report). The Q1 report had shown a slight improvement from the preceding quarter, so the viewability metric appears to be bouncing around a bit.

Brand Risk is the metric used in the report to judge where ads are shown. It’s currently 12.2 percent overall, about the same as in the last report.

Tsemekhman said this means that “out of all impressions sampled, regardless of buying channel, 12.2 percent of impressions were flagged on moderate to very high risk content.”

Risky content can include adult content, alcohol, hate speech, illegal downloads, offensive language, and violence. Most advertisers only care about a few of these, he said, so the Brand Risk score is “not alarming.”

The report also includes a score called TRaq, or True Advertising Quality Score. It is a combo of all the scores in the report — viewability, fraud, and Brand Risk. It now sits at 604 overall for networks, exchanges, and publishers, on a scale of 250 to 1000. The TRaq score is not far off from the last quarterly report.

Almost no sites score anywhere near 1000, Tsemekhman told me. Sites are considered high quality if they are in the 750 to 800 range, and a fraudulent site is 250.