Property descriptions are the piece AI took over most completely, and they were also the cheapest piece to begin with. If a provider's value was writing 120 words about granite counters, that value is mostly gone.
What hasn't collapsed is the work where the output depends on judgment about that specific market and that specific agent. Deciding the pricing story to tell a seller who thinks their house is worth 15% more than the comps support. Choosing which three neighborhoods an agent should farm and building consistent material for them over a year. Building a listing presentation that answers the fee question a seller will actually ask, in the words that agent can say out loud without sounding rehearsed. Photography and video, which are still labor and equipment.
There's also the plain fact that most agents don't do it. An agent can write a decent description with a free tool and still not send a newsletter for eight months, because they're in the car showing houses. Providers sell follow-through as much as craft, and follow-through hasn't been automated.
The honest constraint on the provider side is that AI compressed the low end, so a provider whose whole offer is content volume competes with something free. The ones holding up moved toward strategy, produced things AI can't make, or got embedded in the agent's systems.
One thing worth knowing from your side of the table: a description that reads well tells you the agent bought good tools. It doesn't tell you the comps in that package were chosen honestly, and the pricing page in a listing presentation is a sales document.