Asking for a Format
Tessa asks the model to pull the key details from five guest reviews so she can paste them into her tracking spreadsheet — and gets back a chatty essay. Accurate, well-written, and completely unusable: she needed rows and columns, and she got prose, because prose is what fits when nobody says otherwise. The model can deliver almost any shape you can name — bullets, tables, numbered steps, a fixed template — but the shape must be asked for. This page turns that into a habit, and the habit into the beginning of something bigger.
First, the easy version: say the shape. "As a table with columns for tour, rating, and main complaint." "As five bullets, one sentence each." "As a numbered checklist I can print." The model obeys explicit shape instructions well, and the difference between an answer you read and an answer you use — paste into the tracker, hand to a colleague, print for the wall — is usually one sentence of shape.
Show the Shape
For anything beyond a simple list, the last page's lesson applies with full force: show the shape. Paste a skeleton — the column headers, plus one filled example row — and say "complete it like this." The model finishes visible patterns more faithfully than it builds structures from descriptions, and the gap widens as formats get particular. Tessa's spreadsheet ask becomes: the header row, one worked example from a review she did by hand, then the five new reviews. The output drops into her spreadsheet without edits.
Think of the difference between asking someone to "write up the details" and handing them a pre-printed form. The form decides the shape; the content lands where you need it. A skeleton is a form the model fills.
Templates You Reuse
Here is where the habit becomes a practice. Tessa summarizes reviews every week — same task, same wanted shape. So she stops improvising: the instruction, the shape line, and the worked example get written once, saved in a document, and reused every Monday with only the new reviews swapped in. A fixed ask plus a fixed skeleton is a template, and a template makes the model's output land the same way every time — the closest thing to consistency that Chapter 1's controlled chance allows. The content still adapts to each week's reviews; the shape holds.
Do not skip past how much this one move buys. Consistent shape is what lets outputs stack — this week's summary comparable with last week's, a colleague's Monday matching Tessa's. Chapter 5 builds a whole discipline on this page's saved document, and by Chapter 8 the template becomes something Tessa did not see coming. Plant the habit now: a task you will repeat deserves a template, not a fresh improvisation.
When the Shape Drifts
One honest limitation, so it never surprises you. Over a long output — say, forty reviews summarized in one go — the format tends to wander near the end: columns merge, the verdict phrase grows into sentences, bullets sprout sub-bullets. Nothing is broken; the model is assembling token by token, and over a long assembly the pattern's grip loosens. The fixes are mechanical: work in smaller batches, and re-show the skeleton when you start a new one. Ten reviews at a time in a shown format beats forty in a drifting one — and takes less total time than fixing the drift by hand.
- "The model will pick a sensible format on its own." It picks a plausible one — usually prose, because prose is what fits when nobody says otherwise. Sensible-for-your-spreadsheet must be asked for, or shown.
- "Formatting requests are cosmetic." Format is what makes output usable — pasteable, comparable, stackable. The gap between an answer you read and an answer you use is usually one sentence of shape.
- "Once I ask for a format, it holds forever." Over long outputs the shape drifts near the end. Smaller batches with a re-shown skeleton beat one long drifting run.
- "Saving prompts is for power users." A template is the office's boilerplate letter, reborn: written once, reused every Monday. If you will repeat the task, improvising the ask each time is the amateur move.
- Asking for shape is the bridge from chatting to working — output that drops into a spreadsheet, a document, a workflow without hand-reshaping.
- The template habit planted here is the seed of Chapters 5 and 8: repeatable prompts become checkable output, and checkable output becomes automation. It starts with one saved document.
Knowledge Check
Why did Tessa get a chatty essay instead of spreadsheet rows?
- The model cannot produce tables unless the product's spreadsheet mode is switched on
- Nobody asked for a shape, and prose is what fits when nothing says otherwise
- Guest reviews are too unstructured to summarize as spreadsheet rows
- The chat box can only display flowing paragraphs of text, never tables or columns
For a particular format, what does this page say works best?
- Describe the format in words, naming each column and what it holds
- Ask for the format twice, once before the reviews and once after them, to reinforce it
- Paste a skeleton with headers and one filled example row, and say "like this"
- Accept the prose output and reshape it into rows yourself each week
What turns a one-off ask into a template?
- Saving the fixed instruction and skeleton, and swapping in only the new inputs
- Scrolling back to last week's chat and copying your old messages
- Asking the model to remember this format for all future Mondays
- Buying a chat product that comes with a built-in template feature
Forty reviews in one run, and the format wanders near the end. What is the fix?
- Repeat the format instruction in capital letters at the top for emphasis
- Report the drifting columns to the provider as a bug worth fixing in the model
- Work in smaller batches and re-show the skeleton at the start of each
- Switch to a model from another provider that holds long tables together better
You got correct