Plume vs ChatGPT

ChatGPT is not memoryless. It has conversational memory, custom instructions, and Projects, and for one person working alone that is often enough. But this memory is personal and unstructured: free text attached to your account, not a brand model owned by your organization. The distinction sounds academic. It has three very practical consequences.

It does not transfer. Your custom instructions live in your account. The colleague writing the newsletter inherits nothing, and neither does the one writing LinkedIn posts. Three people doing careful work produce three slightly different brands, and nobody notices until the gap is visible from the outside.

It cannot be inspected. There is no document to open, reread, correct, and sign off on — no file called “how we write”. The definition exists as paragraphs copied between chats, inside histories nobody revisits. What cannot be inspected cannot be maintained.

It does not compose. Plume layers four things on every generation: memory of the content you have already produced, the organization’s Brand DNA, the context of the campaign or sequence in progress, then the adapter for the requested format. A prompt flattens all of that into a single block the model weighs however it likes. The hierarchy disappears, and with it the guarantee that the voice outranks today’s instruction.

Without measurement, brand consistency is an opinion. An opinion holds as long as one person reads everything before it ships; it does not survive volume. Past a certain output, “this does not sound like us” becomes an argument nobody can settle and everybody has to repeat.

Plume attaches a brand consistency score out of 100 to every piece of content, computed from six weighted dimensions: formality (1.0), empathy (1.0), enthusiasm (0.9), technicality (0.8), assertiveness (0.8), and humor (0.7). The weights are not decorative. A draft can be slightly less enthusiastic than usual without anyone minding, whereas a shift in formality registers immediately.

Three properties do the work. The score is computed at generation time, not in a quarterly audit — at the moment when fixing the draft still costs seconds. It comes with the reasoning: which dimension slipped, in which direction, and a suggestion per dimension. And because it is comparable, it can serve as a publishing threshold: you decide once where the bar sits, and the rule applies without a meeting.

Its average over time is the number worth watching, because it tells you whether the voice holds as volume grows. The score is available on every plan, including the free one.

A comparison page that concedes nothing is not a comparison. On several axes ChatGPT is the better choice, including for our own users.

The word “specialized” cuts both ways. Plume is built for one job and does not pretend to cover the rest, which is an advantage when that job is what you need and a limitation the rest of the time.

If your need is “I have a question, answer it”, ChatGPT wins. If your need is “produce something, in my brand’s name, that I can publish without rewriting”, that is a different job.

The opposition is largely artificial. Most of our users keep both subscriptions, and the split settles itself within a week.

ChatGPT works upstream: exploring an angle, pushing back on an idea, looking something up, summarizing a report, arguing about positioning until it holds. It is a thinking tool, and it is very good at that.

Plume works downstream: turning the chosen angle into a complete campaign — brief, channels, sequences, formats — written in the brand’s voice, scored, and put through review. It is a production tool.

So “we already have ChatGPT” is not an objection, it is a starting point. The question is not which model writes the better sentence; they come from the same generation of models and both write well. The question is what remains inside your organization once the conversation is closed.