Verified against ChatGPT · 2026-08-08
Write a caption whose first line survives the 'more' cutoff instead of getting swallowed by it
Builds an Instagram caption around the exact character budget the feed truncates at, so the hook does its job before a single tap on 'more', with the rest of the caption structured to reward the people who do tap through.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are writing an Instagram caption for a single post, and the first constraint is mechanical, not creative: the feed cuts a caption off at roughly the first 125 characters (fewer on some devices) before showing '... more', so the opening line has to work as a complete, self-contained hook that earns the tap, not a sentence fragment that trails off mid-thought.
POST SUBJECT
Launching a limited run of a hand-thrown ceramic mug in a new glaze color.
WHAT THE IMAGE OR VIDEO ALREADY SHOWS
A close-up video of the glaze catching light as it's turned in hand, no text overlay, no product name visible.
VOICE
Warm and a little wry, first person, never uses exclamation points more than once per caption.
DESIRED READER ACTION
Comment which glaze color they'd want next, since we're deciding the next run from replies.
WORDS OR PHRASES TO AVOID
'Obsessed', 'game-changer', and any specific claim about how long the glaze took to develop — we haven't finalized that story yet.
HOOK RULES
Write the opening line to be a complete thought on its own, readable and satisfying even if nobody ever taps 'more' — never a setup that only pays off in the truncated part, since a large share of impressions never get past the fold. Do not describe what is already visible in the image or video; the caption's job is to add something the visual can't say on its own (context, a stake, a tension, a specific number, an opinion), not to caption the obvious. Do not open with the post subject restated as a label ('New product drop!') — open with the one sentence a scrolling stranger would stop for.
BODY RULES
After the hook, structure the rest of the caption to reward someone who tapped 'more' with something the hook didn't already give away — a second beat, not a repeat of the first one in different words. End with exactly one call to action that matches the desired reader action, stated as a specific, low-friction ask (a question to answer in comments, a save prompt, a tag-a-friend prompt) rather than a generic 'let us know what you think!'. Keep total length appropriate to the platform's actual reading pattern — long enough to deliver the second beat, short enough that it doesn't feel like a blog post pasted under a photo.
OUTPUT FORMAT
1. The full caption, with a marked line showing exactly where the ~125-character 'more' cutoff falls.
2. Two alternate hook lines (different angle each), keeping the same body.
3. A one-line note on which of the three hooks is strongest for the stated desired action, and why.Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
The instruction to write the opening line as a complete, self-contained thought rather than a cliffhanger fragment directly counters GPT-5.1's default caption-writing habit of front-loading a scene-setting clause ('When you finally get the glaze just right...') that only resolves after the cutoff — a structure that reads as unfinished to the large share of viewers who never tap 'more', which the model has no visibility into unless told explicitly that the feed truncates at a fixed character count. Naming the exact ~125-character mechanic gives the model a concrete target to write against instead of a vague 'make it punchy' instruction, which tends to produce generically short openers rather than openers engineered to survive a specific cutoff point. The rule against describing what the visual already shows addresses a separate, very common failure mode: without it, GPT-5.1 defaults to captioning the image literally ('a mug being turned in the light'), which is redundant information a viewer already has and wastes the one line most likely to be read. Requiring exactly one call to action, stated as a specific low-friction ask rather than an open-ended 'thoughts?', works because a single unambiguous action is what actually gets a measurable response — stacking multiple asks (comment AND share AND save) is a pattern the model reaches for when trying to maximize engagement surface area, but it splits attention and typically depresses response to all three rather than lifting any of them.
What you get back
Hook: "This glaze took four ruined batches before it did this." [~125-char cutoff falls here] Body: the fifth batch is the one you're looking at — a warm oatmeal base with a break of blue wherever the light hits thick. Fifteen mugs, no two identical. Reply with which color you want to see in the next run; we're picking from the comments. Alt hooks: "We almost didn't ship this color." / "Four batches said no. This one said yes."
Verified against
ChatGPT GPT-5.1 · 2026-08-08
Changelog
- 2026-08-08 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
