It's just A/B testing, not "AI". It's done the same way as digital ads. You come up with 3 different titles for the same link, and serve them to 3 random audiences, and figure out which one got more engagement.
Google and Facebook of course do offer "ML-based conversion optimization", but it's widely regarded as hoopla by ad buyers.
If those frameworks for A/B testing do any statistical analysis or automation I think they are easily classifiable as AI. I wouldn't be surprised if some publications at the top are using much more sophisticated software than that. I could only speculate to what level or with what success ML has been integrated into those pipelines.
Yes, A/B tests are evaluated using T-Test or the Mann-Whitney U statistic. They have existed for a long time and never been classified as "AI". Although with that buzzword currently creating a gold rush, I have no doubt that every half-assed platform will be using that terminology going forward.
Coming up with interesting headlines that entice readers without misrepresenting content is actually pretty difficult--especially at publications that value clever headlines.
Google and Facebook of course do offer "ML-based conversion optimization", but it's widely regarded as hoopla by ad buyers.