The basic notion of this type of counterfactual can be seen in an interferometer.
In an interferometer, a laser beam is split and then recombined to form a interference pattern.
Place a detector in the middle of one of the dark bands in the interference pattern and the detector will detect nothing. Then block one of the two beams - thus eliminating the interference pattern. Now the detector will detect many photons. But the photons it detects are photons that did not hit the block, but could have. They are evidence that you have blocked one of the beams - but without ever having touched that block.
By leveraging and exaggerating that effect, you can have a communication channel that does not contain particles that carry your message.
so basically they are using indirect evidence that did not come from the photons to infer the state of the photon
The full text of the article is behind a paywall - but I can look at the figures.
The experiment is set up so that most of the time, no photon is transmitted across the channel. But that is always possible. If a photon did cross the channel, it would either be reflected or not - depending on whether Bob wants to transmit a zero or a one. Some of the photons that stay on Alice's side are affected by the possibility that they could have been reflected by Bob.
A key concept here is that photons do not behave as baseballs behave. A baseball has a trajectory, it is always somewhere. A photon has a start and end point, but it is not correct to think that it simply makes a bee line between those points.
From the diagrams, it looks like the difference between a "1" and a "0" was an 8% difference in photon intensity. So they must have been using dozens of photons per bit. It would be interesting whether this caused the average number of photons that actually transited the channel to exceed the number of bit that were transmitted.