Smells like bad experimental design. Shouldn't the face-to-face subjects also be completing written tasks in session as a control? There are entire schools of psychotherapy that require in-session activities.

If not, then they should compare standard talk face-to-face therapy with online CBT "chat."

As an addendum: I think social sciences are categorically separate from harder sciences (good social scientists understand this) and attempts at laboratory/population experiments are typically laughable. They're usually just done because that's what the people and organizations that fund the departments want.