How to Make Claude Opus 5 Less Verbose

To make Claude Opus 5 less verbose, specify the answer’s shape and length in your prompt, then show a short example. Anthropic’s Opus 5 prompting guidance↗ (opens in a new tab) distinguishes visible response length from thinking effort: lowering effort does not reliably make the answer shorter.

Start with the response you actually need: a decision, a short explanation, or a finished document. Tools like fenbs keep your approved response instructions as reusable project guidance, so each new session can read the same preference.

What should you tell Opus 5 first?

1. Check the selected model label. Use the instructions that match the model you are actually using, rather than assuming every Opus version behaves alike.
2. Name the reader and the job. For example, say that a project lead needs a decision before a meeting, rather than asking for an unrestricted analysis.
3. Set a visible output budget: one recommendation, three supporting points, and a short statement of uncertainty.
4. Show one answer that has the amount of detail you want. Keep its subject separate from the real task so Claude copies the form rather than the facts.
5. Try the prompt on a real question. If it omits something necessary, change the requirement instead of simply asking for an even shorter answer.

A useful limit describes what fits inside it. “Keep it short” leaves Claude to decide whether your topic deserves a sentence or a page. “Give me the recommendation, its main reason, and the next action in under 120 words” supplies both a structure and a stopping point.

Answer my question in under 120 words.
Start with the recommendation.
Then give up to three supporting points.
Include a caveat only if it changes the recommendation.
Ask one question only if missing information prevents a useful answer.
Offer more detail in one sentence when it would help.

Why can a lower thinking setting still produce a long answer?

Thinking and communicating are different activities. A difficult calculation may need careful work but only a short conclusion. An easy request to write a detailed report can require a long deliverable. Treat your desired answer length as a separate requirement from how deeply the task needs to be analyzed.

Keep accuracy requirements explicit. For a spreadsheet check, ask for the incorrect cells, the correction, and whether the total now matches. A short explanation can still name the evidence. Do not trade away necessary checking merely to reduce the amount you have to read.

The broader Claude concision guide covers everyday response preferences across Claude workflows. Here, the useful adjustment is to give Opus 5 separate limits for its answer, its updates during work, and the files it produces.

How do you shorten updates without losing visibility?

When an assistant uses tools, several small explanations can add up to more reading than the final result. Describe which events deserve an update. A change in direction, a meaningful finding, or a blocked dependency is usually more helpful than repeated announcements of the next routine action.

Before starting, state the intended action in one sentence.
During the task, update me when you find a material issue, change direction, or need my input.
Keep each update to two sentences.
At the end, state the outcome, what you checked, and any remaining limitation.

For example, “The totals match; two invoice dates are missing” gives you information you can act on. “I will now carefully review the data to ensure everything is correct” adds little once the task is underway. Include your preferred version as a positive example, then judge whether the next update meets it.

Save the version that works in fenbs Skills and prompts, with a note that it governs communication during work. That keeps it distinct from the instructions for the final report.

What if the document is still too long?

Give the deliverable its own size and contents. “Write a report” can reasonably invite background, method, findings, alternatives, and recommendations. If you need a one-page decision memo, ask for that object and list its required parts. Do not expect a short chat preference to explain every document you might request.

• For a decision memo: state the decision, two options, the recommendation, and one unresolved risk.
• For a work handoff: state what changed, what was verified, what remains, and where to resume.
• For an email: name the recipient, desired action, key facts, and maximum length.
• For a technical explanation: name the reader’s knowledge and the specific question the explanation must settle.

Ask for an outline first when the document has substantial scope. Remove sections you do not need before requesting prose. If an answer is already long, say “Reduce this to 180 words while preserving the recommendation, deadline, and unresolved dependency.” That gives revision a concrete test.

Where should you keep the preference?

Use a current prompt for an exception and persistent instructions for a repeatable preference. Anthropic’s personalization guide↗ (opens in a new tab) distinguishes account-wide instructions, project instructions, and task-specific skills. Choose the scope that matches your need rather than placing a client’s report format into every conversation.

Remove conflicting directions. An older instruction to explain every step can compete with a newer request for a short answer. Anthropic’s prompting best practices↗ (opens in a new tab) recommends explicit output constraints and examples. A small, consistent instruction set is easier to evaluate than several overlapping demands.

Do not switch models before trying the output requirement on the same task. If you are deciding between model families, the Sonnet and Opus comparison covers that separate choice; a model change alone is not a precise length instruction.

How to do this with fenbs

1. Create a free account, free for one person with no credit card, and add the project that needs shorter responses.
2. Save a prompt with separate instructions for final answers, progress updates, and documents. Include one approved example and explain when each format applies.
3. Connect Claude through the integration guide, then ask it to read the project context and retrieve the saved prompt before working.
4. Review a real result. Update the saved instructions to preserve anything important that was omitted, and save progress before ending the session.

fenbs does not change Opus settings or guarantee a word count. If you only need a shorter reply once, a direct follow-up is enough. Its value is keeping the reviewed guidance available across sessions and supported assistants.

Judge the result by whether you can act on it without asking for missing essentials. The AI prompts hub has more reusable formats; reducing wordiness focuses on editing unnecessary phrasing inside an answer.