Facebook has found more flaws in its ad stats, and is making changes to give advertisers more accurate numbers.
This time Facebook said it will tweak how it estimates the potential audience size for new campaigns, an area where advertisers have been saying the figures seemed off.
"We are updating how we calculate the numbers that appear in our estimated reach tool -- found when creating an ad," Facebook said in a blog post on Friday.
"We're improving our methodology for sampling and extrapolating potential audience sizes," Facebook said. "This will help to provide a more accurate estimate for a given target audience and to better account for audiences across multiple platforms"
Facebook is also updating how it reports the number of reactions to live videos -- when people click emojis to register their opinions about the content. And it said it found an error in how it counts likes and shares on posts.
It's the third time since November that Facebook has disclosed reporting errors on metrics that advertisers rely on to plan or measure their campaigns.
Facebook previously found that it had inflated how long people were viewing videos on average and overstated how many people were visiting brands' pages monthly.
The errors dripping out have shaken many people in advertising. They say Facebook has not done anything nefarious, but that the problems erode trust.
"It's freaking everybody out," said one top ad agency exec, speaking anonymously because of a close relationship to Facebook. "It's hard to explain to clients."
Still, Facebook is not alone in presenting numbers to advertisers, and then facing skepticism, a number of marketers said. Other platforms also do their own reporting, and they control what companies get access to their data.
Facebook has been making changes, allowing third-party measurement companies like Moat and Integral Ad Science to double-check some of the numbers. Some have questioned whether those agreements go far enough, however, citing Facebook's continued control over the data the companies receive.