AGENCYSCRIPT
CoursesEnterpriseBlog
đź‘‘FoundersSign inJoin Waitlist
AGENCYSCRIPT

Governed Certification Framework

The operating system for AI-enabled agency building. Certify judgment under constraint. Standards over scale. Governance over shortcuts.

Stay informed

Governance updates, certification insights, and industry standards.

Products

  • Platform
  • Certification
  • Launch Program
  • Vault
  • The Book

Certification

  • Foundation (AS-F)
  • Operator (AS-O)
  • Architect (AS-A)
  • Principal (AS-P)

Resources

  • Blog
  • Verify Credential
  • Enterprise
  • Partners
  • Pricing

Company

  • About
  • Contact
  • Careers
  • Press
© 2026 Agency Script, Inc.·
Privacy PolicyTerms of ServiceCertification AgreementSecurity

Standards over scale. Judgment over volume. Governance over shortcuts.

On This Page

Pass One: Define the Job Before You Write AnythingWho Reads the Output and What Do They Do With ItWhat Would Make This Summary FailPass Two: Write the Instruction BlockLead With the Task and the ReaderSeparate Instructions From SourcePass Three: Add Length and Format ControlsGive a Concrete LengthChoose a Structure That Matches the JobPass Four: Add Fidelity ConstraintsName What Must Survive CompressionForbid Invention and SmoothingPass Five: Run It and Read Against the SourceCheck for Omission FirstCheck Tone and SpecificsPass Six: Tighten and Save as a TemplateConvert Fixes Into Permanent RulesSave the Prompt With a LabelA Worked Example of the Six PassesPasses One Through Three in ActionPasses Four Through Six in ActionAdapting the Workflow to Your PacePutting the Passes TogetherFrequently Asked QuestionsCan I skip the planning passes once I am experienced?How do I handle a document that is too long to paste in one go?What if the summary is accurate but still feels generic?Should the verification read happen every time?Key Takeaways
Home/Blog/Build a Reliable Summarization Prompt in Six Passes
General

Build a Reliable Summarization Prompt in Six Passes

A

Agency Script Editorial

Editorial Team

·December 9, 2021·6 min read
prompting for summarization qualityprompting for summarization quality how toprompting for summarization quality guideprompt engineering

Most advice about summarization prompts is a pile of tips with no order to them. You read them, nod, and then sit in front of a blank prompt box with no idea what to type first. This guide fixes that by giving you a sequence. You will move through six passes, each building on the last, until you have a prompt that produces summaries you can trust without re-reading the source every time.

The workflow assumes you have a specific document and a specific reason for summarizing it. That context matters, because the best summary prompt depends entirely on what the summary is for. We will start by pinning that down, then add structure, constraints, and a verification step.

Work through the passes in order the first few times. After a handful of documents, you will internalize the sequence and run it in your head.

Pass One: Define the Job Before You Write Anything

The most common mistake is opening with "summarize this." Resist it. First answer three questions on paper.

Who Reads the Output and What Do They Do With It

A summary that feeds a decision is different from one that feeds a filing system. Write down the reader and the action. "The account lead, to decide whether to escalate" is a job. "Summarize" is not.

What Would Make This Summary Fail

Name the failure you most want to avoid. Dropping a deadline? Sounding more certain than the source? Burying the one objection the client raised? Knowing the failure mode tells you which constraint to write later.

Pass Two: Write the Instruction Block

Now open the prompt. Start with a plain instruction that encodes the job from Pass One.

Lead With the Task and the Reader

Write a single sentence: "Summarize the document below for [reader] so they can [action]." This puts the purpose first, which orients everything the model does afterward.

Separate Instructions From Source

Add a clear divider before the source text, such as a line reading "Document:" on its own. Models follow instructions more reliably when they can tell where the directions end and the material to compress begins.

Pass Three: Add Length and Format Controls

A summary without a length target tends to drift toward whatever feels natural to the model, which is rarely what you need.

Give a Concrete Length

State a number. "In no more than 120 words" or "exactly five bullet points." Concrete targets are enforceable; "keep it brief" is not.

Choose a Structure That Matches the Job

Decisions and action items read best as a numbered or bulleted list. Narrative context reads best as a short paragraph. Pick the structure that the reader can scan fastest given what they will do next.

Pass Four: Add Fidelity Constraints

This pass is where summary quality is won or lost. You are telling the model what it may not do.

Name What Must Survive Compression

List the elements that cannot be dropped: "Preserve every name, figure, date, and commitment." This protects the specifics that models tend to discard in favor of themes.

Forbid Invention and Smoothing

Add two rules: "Use only information present in the document" and "Match the level of certainty in the source; do not resolve ambiguity or add confidence." The second rule stops the model from rewriting a hedged report into a confident one.

Pass Five: Run It and Read Against the Source

Now execute the prompt, but do not accept the output blindly.

Check for Omission First

Skim the source and list the points you would have kept. Confirm each one appears in the summary. Missing material is the quality problem people notice last and regret most.

Check Tone and Specifics

Verify that numbers match exactly, that no claim is stronger than the source, and that nothing was invented. If anything is off, you have a fix for the next pass rather than a reason to start over.

Pass Six: Tighten and Save as a Template

Use what you learned from the read-through to make one or two targeted edits, then preserve the result.

Convert Fixes Into Permanent Rules

If the summary dropped the budget, the permanent fix is a line that says "always preserve budget figures." Each correction becomes a reusable rule, so the next document of this type starts from a stronger prompt.

Save the Prompt With a Label

Store the finished prompt with a name like "client-call summary" so you are not rebuilding it from scratch next week. A small library of labeled prompts is the real payoff of this workflow.

A Worked Example of the Six Passes

To make the sequence concrete, walk through a single document, a forty-minute client call you need to recap for a teammate.

Passes One Through Three in Action

Pass one: the reader is your teammate, who will run the next call, and the failure to avoid is dropping any client request. Pass two: you write "Summarize the call below for the teammate running the next session so they can pick up where we left off," then add a "Transcript:" divider. Pass three: you set "no more than 150 words" and choose a bulleted list because the teammate will scan for action items.

Passes Four Through Six in Action

Pass four: you add "list every client request, concern, and commitment; use only what appears in the transcript; match the client's level of certainty." Pass five: you run it, then skim the transcript and notice the model dropped a passing comment about a budget cap. Pass six: you add "preserve any mention of budget or pricing" as a permanent rule and save the prompt as "client-call recap." The next call starts from a prompt that already knows your failure modes.

Adapting the Workflow to Your Pace

The six passes are training wheels. Once the sequence is familiar, you will collapse the planning passes into a few seconds of thought and run the constraint and verification passes almost automatically. The order never changes, only the speed. Resist the temptation to drop the verification read as you get faster, since that is the pass that catches the silent omissions, and it is the one experienced users most regret skipping when a summary turns out to have missed something.

Putting the Passes Together

Run end to end, the six passes take a few minutes the first time and under a minute once the template exists. The discipline is in the order: job before instruction, instruction before format, format before constraints, and a verification read before you trust the result. Skipping the early passes is what produces the unreliable summaries most people blame on the model.

Frequently Asked Questions

Can I skip the planning passes once I am experienced?

You can compress them, but do not skip the thinking. Experienced users still decide who the reader is and what failure to avoid; they just do it in seconds rather than on paper. The structure stays the same even when it gets faster.

How do I handle a document that is too long to paste in one go?

Split it into sections, summarize each with the same prompt, then run a final pass that summarizes the section summaries. Keep your fidelity constraints on every pass so specifics survive the chain rather than eroding step by step.

What if the summary is accurate but still feels generic?

Generic usually means your length target is too tight or your "what must survive" list is empty. Add specific elements to preserve and give the model enough room to include them. Specificity comes from naming what matters, not from asking for more detail in the abstract.

Should the verification read happen every time?

For anything that informs a decision, yes. For low-stakes personal notes you can relax it. The read against the source is cheap insurance, and most quality failures are caught in that single step.

Key Takeaways

  • Define the reader and the failure to avoid before you write a single instruction.
  • Lead the prompt with task and purpose, and separate instructions from the source.
  • Set a concrete length and a structure that matches what the reader will do.
  • Add fidelity rules: preserve specifics, forbid invention, match the source's certainty.
  • Read the output against the source, then convert each fix into a permanent rule and save the prompt.

To go deeper, see the underlying principles in our Prompting for Summarization Quality: Best Practices That Actually Work, turn the workflow into a reusable model with A Framework for Prompting for Summarization Quality, and keep yourself honest using the Prompting for Summarization Quality Checklist for 2026.

Search Articles

Categories

OperationsSalesDeliveryGovernance

Popular Tags

prompt engineeringai fundamentalsai toolsthe difference between AIMLagency operationsagency growthenterprise sales

Share Article

A

Agency Script Editorial

Editorial Team

The Agency Script editorial team delivers operational insights on AI delivery, certification, and governance for modern agency operators.

Related Articles

General

Prompt Quality Decides Whether AI Earns Its Keep

Prompt quality is the single biggest variable in whether AI delivers real work or expensive noise. The model matters, the platform matters — but the prompt you write determines whether you get a first

A
Agency Script Editorial
June 1, 2026·10 min read
General

Counting the Real Cost of Every Token You Send

Tokens and context windows sit at the intersection of AI capability and operational cost—yet most business cases treat them as technical footnotes. That's a mistake that costs real money. Every time y

A
Agency Script Editorial
June 1, 2026·10 min read
General

Rolling Out AI Hallucinations Across a Team

Most teams discover AI hallucinations the hard way — a confident-sounding wrong answer makes it into a client deliverable, a legal brief, or a published report. The damage isn't just to the output; it

A
Agency Script Editorial
June 1, 2026·11 min read

Ready to certify your AI capability?

Join the professionals building governed, repeatable AI delivery systems.

Explore Certification