Open Loop/Guides
Yes — if you are willing to review what it gives you.
That is the whole question. The tools are good now. They are not the problem. The problem is the pile of candidates on the other side, and who goes through it.
| Tool, self-serve | Us | |
|---|---|---|
| Reads the episode | Yes, AI | Yes, AI |
| Finds candidates | Yes, well | Yes, well |
| Who culls the list | You | Us |
| Who fixes bad timings | You | Us |
| Says an episode is thin | No | Yes |
| Knows what you posted last week | No | Yes |
| Monetisation checked | You | Audio and captions |
| Answerable if a clip is wrong | Nobody | A person |
| Your time per episode | An hour or two | Five minutes |
| Cost | Little to nothing | Hundreds a month |
Clipping tools used to work on signal: volume spikes, speech pace, silence gaps, keyword hits. It was fair to say they did not understand the conversation, because they did not.
That is no longer true. Current tools run on language models. They read the transcript, follow the argument, and can genuinely identify a surprising claim or a story that works without its setup. If you see a clipping service claiming AI "just matches patterns", they either have not used one recently or they are hoping you have not.
We are not going to make that argument, because we would lose it — and because we run the same first pass ourselves.
The culling. A model asked for ten clips returns ten. It has no mechanism to return three and tell you the rest of the episode was flat. Someone has to be willing to deliver less.
The context it cannot have. That you posted something too similar last week. That this guest asked not to be quoted on that. That the clip is fine but makes you sound smug. That you are launching something in a fortnight and this cuts against it.
Accountability. If a bad clip goes out, a person made that call and can be asked why. Nothing is answerable for a tool's output except the person who pressed publish — which is you.
The hour you get back. The most honest framing of the whole service: you are buying somebody else's attention, not a capability you cannot otherwise get.
A reasonable middle path: run a tool on your back catalogue for volume, and put a person on current episodes where quality matters.
If a clipping service tells you AI cannot understand your podcast, ask when they last tried one. We use it, we say so, and what we charge for is what happens after it runs.
Bring an episode to a 20-minute call. If a tool would serve you better we will say so.