Give the model context before you ask

What it is

Most people open a chat and ask a question straight away. This approach flips the order. You paste the material you want worked on, tell the model what it is, and let it read before you ask for anything. The reading pass becomes the foundation for every question that follows.

It matters because a model can only reason about what sits in front of it. If you ask for feedback on a landing page before the page exists in the chat, the model invents a generic page and critiques that. Once the real text is loaded, its answers point at your actual sentences, your actual gaps.

When to use it

Reach for this pattern when the task depends on a specific body of text:

• Editing or tightening a draft you wrote
• Reviewing a contract, policy or report for gaps
• Writing copy that must match an existing brand voice
• Comparing two versions of the same document
• Answering questions about a long thread or transcript

Skip it for open-ended brainstorming where no fixed material exists. There is nothing to prime with, and a blank page is the right starting point.

How to run it

1. Paste the full text in one message. Do not summarise it yourself. Trimming removes the details the model needs to catch problems.
2. Label it in one line. Say what the text is, who it is for and what stage it is at. For example: "This is a second draft of a welcome email for new subscribers to a meal-kit service."
3. Ask for a read-back before any critique. Request a short summary of the main claims and the intended reader. If the summary misses the point, your label was unclear. Fix it before going further.
4. State the task in a separate message. Be concrete: "Flag any sentence that promises a delivery time we have not confirmed." One task per message keeps the answers sharp.
5. Ask for evidence with each finding. Require the model to quote the line it is reacting to. That makes the output checkable and stops vague advice.
6. Iterate on the same thread. Follow-up questions inherit the loaded text, so you never paste it twice.

What a good result looks like

A useful response names specific lines and explains why each one is a problem. It separates what the text says from what it assumes. It does not rewrite everything at once unless you ask it to.

Here is a small example. You paste a three-paragraph product description for a desk lamp and label it as copy for a furniture retailer's site. You ask for a read-back. The model says the lamp is aimed at home-office buyers and that the main claim is adjustable brightness. You then ask it to list any claim a shopper could not verify from a photo. It returns two lines, quoted, with a note that neither is backed by a specification.

That is the payoff. You get comments tied to your words, not a generic essay about product copy. The model has read the room before it speaks, and every later question builds on that shared ground.