Precision
UPDATED DEC 11, 2025
What is precision
Precision measures the accuracy of our enforcement actions on Facebook or Instagram. Specifically, it estimates the percentage of content we actioned for violating our policies that were correctly identified as violating. This metric reflects how effective our systems and reviewers are at distinguishing violating content from non-violating content.
Another way to think of precision is: out of all the content we removed for policy violations, how much of it truly violated our standards? High precision means that most actioned content was correctly identified as violating, while low precision indicates that a significant portion of actioned content was actually non-violating.
How we measure precision
Precision is estimated using samples of actioned content from across Facebook or Instagram. We calculate it as: the estimated number of actioned items that were truly violating, divided by the estimated number of total actioned items. For example, if the precision on Facebook was 98% to 99%, that would mean of every 10,000 pieces of content actioned for this policy, 9,800 to 9,900 on average were incorrectly identified as violating. While precision can be very high, even a small number of mistakes can have a significant impact on people and creators.
Why we measure precision of actioned content
We estimate how accurately we enforce our policies because we want to ensure that our actions are fair and correct. A piece of content that is incorrectly actioned can negatively affect people’s experience and trust in our platform. Measuring precision helps us understand and improve the effectiveness of our enforcement systems and reviewer decisions.
A high precision number means that our enforcement is accurate and that we are minimizing the impact on non-violating content. Even a small decrease in precision can correspond to a large number of mistakenly actioned items, due to the scale of enforcement on our services.
How we use sampling to estimate precision
We estimate precision by sampling actioned content on Facebook or Instagram. To do this, we review samples of actioned content and label them as truly violating or not violating according to our policies. Using the portion of these samples that were truly violating, we estimate the percentage of all actioned items that were correctly identified. We express this uncertainty by quoting a range of values, for example by saying 9,800 to 9,900 out of every 10,000 actioned items were correctly identified for violating a policy. This range reflects a 95% confidence window. This means that if we performed this measurement 100 times using different samples each time, we expect the true number to lie within the range 95 out of the 100 times.
Caveats
- The people who apply labels to our samples sometimes make mistakes, including labeling violations as non-violating or vice versa. This typically doesn’t, but could impact the precision measurement. For this reason, we may have two people review a sample to ensure accuracy in our labeling, and if there is ever a disagreement, we have a third person act as the tiebreaker.
- The current precision measurement covers the majority of enforcement actions on Facebook and Instagram against our Community Standards. We are actively working to ensure all aspects of the platform are adequately represented in this metric.
- To generate a representative measurement of global precision, we sample and label content in the multiple languages for Facebook and Instagram and are confident this approach provides a representative global estimate and are continually working to expand coverage of the metric.