Background: Conversations about serious illness care goals are often inadequate and fail to address key elements of high-quality discussions. Methods: As part of a cluster-randomized trial of a multi-component communication intervention, we conducted a retrospective chart review of 147 deceased oncology patients to assess frequency and timing of documentation of end-of-life (EOL) conversations; charts of a subset of 20 intervention and 20 matched control patients underwent detailed review to assess quality. A systematically developed abstraction tool was used by two blinded researchers. The tool contained 25 elements reflecting four EOL conversation domains: goals/values, illness understanding/prognosis, EOL care planning, life-sustaining treatments. Results: Of 153 intervention patients, 44.4% died (n = 68); of 161 controls, 49.1% died (n = 79). Significantly more intervention patients than controls (92.7% vs 74.7%, p = 0.006) had at least 1 documented EOL discussion before death; intervention conversations occurred 3 months earlier (median 147 days vs 62 days, p = 0.003). 59.4% of intervention conversations were documented in a retrievable EHR location compared to 10.2% of controls (p = 0.001). In the detailed review, 85% (n = 17) of intervention and 40% (n = 8) of controls had at least 1 documented discussion about values/goals (p = 0.0001), with an average of 3.6 of 8 elements (0.7 of 8 for controls) (p = 0.0003). 85% percent (n = 17) of intervention and 30% (n = 6) of controls had at least 1 documented discussion about prognosis (p = 0.0014), with an average of 2.5 of 7 elements (0.5 of 7 for controls) (p = 0.001). 85% of intervention (n = 17) and 55% of controls (n = 11) had at least 1 documented discussion about EOL planning (p = 0.009). 55% of intervention and 30% of controls had at least 1 documented discussion about life-sustaining treatments (p = 0.20). Conclusions: The intervention resulted in more, earlier and better documentation of serious illness care goals. Across 3 of 4 domains of quality, intervention patients had more detailed information about their EOL care preferences, demonstrating that the intervention results in more patient-centered, comprehensive discussions. Clinical trial information: NCT01786811