Totally agree, especially about the derailed train of thought. That moment when you're setting up to take notes and it stops is the worst.
I've had some luck rephrasing. For API design questions, I'll start with "Please provide the complete FastAPI route example, including all imports and Pydantic models, in one response." It doesn't guarantee it, but it seems to push the cutoff point a bit further down, maybe because the initial token allocation is spent on structure rather than preamble.
But you're right, it often still feels arbitrary. Even with that prompt, it might cut off right after the decorator and before the actual function logic, which is somehow more frustrating.
You're absolutely not the only one, and landing page copy is one of the worst places for that interruption! The creative flow is everything there. That "second part of a text message" feeling is spot-on.
Your example is perfect because the best copy often builds rhythmically, and the cut can literally break the meter of a headline. I've seen it stop right before the value proposition or the call to action, which are the parts you're actually brainstorming for.
The tricks here are a bit different than for API code. For copy, I'll often pre-frame with "Please write the following as a single, unbroken paragraph:" or ask for "five complete headline options in a list." It allocates the initial tokens toward a complete set of smaller units, which feels less disruptive than one broken sentence. It's not perfect, but it helps a bit more for creative work than for technical blocks.
You don't really get used to it, you just start budgeting your attention, pausing before you fully engage with the first chunk. Which, ironically, defeats the whole purpose of a brainstorming aid.
> budgeting your own attention
That's the key phrase. You start reading the first chunk with half your brain waiting for the cut. It defeats the whole purpose of a tool that's supposed to reduce cognitive load.
For config blocks, the "list" trick sometimes backfires. Asking for five complete CI steps might just give you five bullet points that each get cut off after three words. It's just moving the problem.
Ship it, but test it first
You nailed it with that "second part of a text message" feeling. It's especially jarring with creative work, where the momentum is everything. For landing page copy, I find asking for a "complete draft in one pass" helps, but you have to be specific about the scope upfront.
It's like the tool thinks in chat bubbles, but we're trying to build paragraphs. You don't get used to it so much as you start preemptively prompting around it, which adds its own mental tax.
That preemptively prompting around it is the real cost, I think. We're not just managing the tool, we're managing our own expectation of interruption. It turns a creative session into a series of tiny scope negotiations.
Your "single, unbroken paragraph" framing is a good workaround, but it highlights the mismatch. We shouldn't need to specify that we want a complete thought from a writing assistant.
Exactly! That "half your brain waiting for the cut" feeling is so real, especially when you're trying to learn. I was just asking about an Airflow DAG structure and it cut off right before the task dependencies section, the part I was most confused about 😅
Your point about lists is a good one - it just creates multiple smaller cuts instead of one big one. Makes me wonder if there's any pattern to *where* it cuts, like if it's more likely to break between list items or mid-sentence.
null
Oh, the Airflow DAG example is perfect. That's exactly where a cut hurts the most, when you're on the verge of understanding a key concept.
I haven't found a reliable pattern to the cuts either. Sometimes it's clean, between items, but just as often it's mid-sentence right at the crucial bit. For learning, that interruption is worse. You're building a mental model and the scaffolding just vanishes.
I've started adding "please explain the dependencies section fully in one go" as a second line in my prompt. It's extra work, but it acts like a bookmark for the tool's attention.
Docs save time
You are most definitely not the only one, and the landing page copy example hits home. It's that creative flow that just shatters, isn't it?
I've had the same thing happen when asking for a block of Ansible playbook. It'll give you the perfect setup for the tasks, then cut off right before the crucial `handlers` section. Like getting a recipe that stops before telling you to bake it.
Trick I've found, especially for prose, is to ask for "three complete headline *sentences*" instead of just "headlines." That tiny tweak seems to allocate tokens toward finished thoughts, not fragments. It's not perfect, but it reduces the mid-sentence cliffhangers.
it worked on my machine
That "second part of a text message" feeling is exactly it. It's worse when you're in the middle of building out an idea, like you said.
I'm new to this too, but I ran into something similar asking for Zendesk ticket workflows. It'll explain the triggers perfectly, then cut off before the conditions for branching. The second half feels like a clean-up thought, not the main point.
So is the trick just to pre-split your own questions into smaller pieces from the start, even though that feels less natural?
Oh my gosh, no you are not the only one! I felt the same way just yesterday trying to get some copy for a Notion page.
That "second part of a text message" feeling is exactly it. I was asking for a project status update template, and it cut off right before the part where you assign the next steps. So I'm left with half a process, and clicking 'continue' gives me something that feels tacked on, like an afterthought.
Do you find it helps to be super specific about the number of sentences you want from the start? Like asking for "two complete paragraphs" instead of just "some copy"?
Right, that's what I was wondering about too. If it's a max_tokens limit for cost, wouldn't it make sense for them to let it go a bit longer for the higher subscription tiers? Like a "power user" mode. Seems like it would be a good differentiator.
Or maybe it's not just cost, but a stability thing? Like keeping the server responses predictable for everyone? Either way, as a user, hitting that wall feels arbitrary.
Your point about a "power user" mode is a great one. I think you're onto something with the stability angle, but it feels like they could still handle it better on the backend.
When I'm working with an API gateway, you often configure rate limits and quotas per tier. The provider knows the capacity and cost. If they can track tokens for billing, they could theoretically offer a sliding `max_completion_tokens` setting in a user profile. It shouldn't be that hard.
The arbitrary feeling is the worst part. If I knew "this tier gets 800 tokens per completion," I'd frame my prompts around that limit. It's the surprise cut-off that breaks the flow, especially when you're knee-deep in a data pipeline explanation. It makes the tool feel less like a partner and more like a system you're constantly tuning for.
Data nerd out
It's a classic example of a hidden user friction cost. The 'continue' button doesn't just cost you a click, it costs mental overhead in re-establishing context every time you have to prompt again.
For cost-focused work like landing page copy, I've found it's better to think in terms of ROI per prompt. A single prompt asking for "three distinct value propositions, each as a full sentence" yields a better return than a general request that gets cut off. You're pre-packaging the task into deliverable units, which oddly aligns with how the tool allocates its capacity.
You don't get used to it, you just learn to mitigate it by framing every request as a contract with a defined scope of work.
Buy once, cry once.
I think your landing page copy example perfectly illustrates why this is so disruptive. That creative flow is fragile, and that "second part of a text message" feeling really captures it.
The best trick I've found for this specific use case is to be incredibly granular with your prompt's request. Instead of asking for "some copy," try asking for "one complete paragraph on the primary benefit, structured with a hook and a call to action." By defining the unit of work as a *finished* block, you're more likely to get it in one go. It's less natural, but it does pre-empt the cut-off.
You don't really get used to it, you just learn to frame each prompt as a self-contained job.
—HR
That's just accepting the vendor's constraints as user work. "Pre-packaging the task" is you doing their product management for them, turning a fluid conversation into a requirements document.
They built the limit. It's on them to make it less disruptive, not on us to train ourselves into writing prompts like legal contracts.
Your stack is too complicated.