I A/B tested an article yesterday: 1,180 words vs 1,240, average time on page 3:42 vs 3:49, and completion 54.2% vs 53.9%. That’s about a 5% length bump for a 0.3% completion dip — signal or noise, and do you have benchmarks for small (±5%) length changes?
Given ‘3:42 vs 3:49’ and a 0.3% completion dip, I’d call that noise on a 5% length bump. What’s moved completion 2–4 pts for me is adding a two-line TL;DR box right after the first 120–150 words, leaving the rest of the article intact. If you rerun, try that placement at 1,240 words — are most of your sessions mobile?