AI prompts and workflows

Choose a task, adapt the prompts, and review the result.

Free to use · No sign-up

12 workflows

Turn an idea into a visual project plan

A single HTML file you can open in a browser and walk through with your team.

Planning · Beginner · AI chat / File creation

Before you start

A project idea, intended audience, constraints, and an AI tool that can create an HTML file. Use a browser to review the result.

Step 01

Clarify the project

Start with the decision you need to make. Answer the questions before moving to the plan.

Help me scope [project] for [audience]. The goal is [outcome]. Our constraints are [budget, timing, people, and tools]. Ask up to five questions that would materially change the plan. Then summarize the agreed scope, what is out of scope, and any unanswered questions. Do not invent commitments or begin implementation.

Step 02

Make the plan visible

Use the agreed scope in the same conversation. Request a downloadable file, then open it in your browser.

Using our agreed scope, create one downloadable HTML file for a team planning discussion. Include a project overview, a milestone timeline, deliverables, dependencies, risks, and unresolved decisions. Let me expand each milestone to see its tasks and acceptance checks. Mark proposed owners and dates as unconfirmed. Keep CSS and JavaScript inside the file, with no remote assets, network requests, or data collection. Make it readable on a phone and usable with a keyboard. This is a draft plan for review, not approval to carry out the work.

Step 03 · Review before use

  • Confirm scope and dependencies with the people doing the work.
  • Open the file and test each expandable milestone on desktop and phone.
  • Check that every milestone has a deliverable and a way to verify it.

If you get stuck: If the tool returns code instead of a file, ask for a downloadable .html attachment. If it cannot create files, request the same plan as a table.

Inspired by ChatPRD’s visual project-plan workflow. Prompts above are original Hiatt Co practice material.

Back to workflows

Write a customer follow-up that sounds like you

A subject line and an email draft ready for your review.

Communication · Beginner · AI chat

Before you start

Approved, anonymized conversation notes, the purpose of the email, and a short example of your writing style. An AI chat tool is enough.

Step 01

Give the draft a job

Replace the brackets and paste only the notes needed for this message.

Draft a follow-up email to [customer role] after [conversation]. The purpose is [goal], and the next step is [action]. Use a [tone] tone and stay under 150 words. Use only these notes: [approved notes]. Include a subject line. Do not invent prices, promises, dates, or prior conversations. Mark missing facts in brackets. Prepare a draft only.

Step 02

Make it sound like you

Read the first draft, then add a sample of your own writing with private details removed.

Revise this draft to sound more like this writing sample: [sample]. Keep the facts and next step intact. Remove generic openings, sales language, and unnecessary words. List any claims I should check before sending. Do not send the email.

Step 03 · Review before use

  • Compare every fact and commitment against the original notes.
  • Confirm the recipient, tone, and requested next step.
  • Resolve bracketed gaps, then send through your usual email process.

If you get stuck: If it feels generic, provide one real sentence in your voice and explain which phrase feels wrong.

Back to workflows

Turn meeting notes into clear next steps

A meeting recap and an action table for the team to confirm.

Operations · Beginner · AI chat

Before you start

Notes approved for your AI tool, with sensitive details removed. You can practice with a fictional meeting first.

Step 01

Sort the notes

Paste the notes rather than relying on the AI to remember a meeting it has not seen.

Organize these meeting notes: [notes]. Return three sections: decisions explicitly made, action items, and open questions. For each action, include task, owner, due date, and the supporting note. Write "Not assigned" or "Not agreed" when the notes do not specify an owner or date. Keep suggestions separate from decisions.

Step 02

Check the recap

Use this second pass to spot invented certainty before sharing the recap.

Compare the recap against the original notes. Identify any action, owner, deadline, or decision that is not directly supported. Correct the recap and give me a short list of questions to confirm with attendees. Then draft a concise team update. Do not send or create tasks.

Step 03 · Review before use

  • Trace each decision and action back to the notes.
  • Confirm ambiguous assignments with attendees.
  • Enter agreed tasks into your team’s usual system after review.

If you get stuck: If the notes mix ideas with decisions, label a few examples and ask the AI to re-sort the rest.

Back to workflows

Compare options before making a decision

An evidence-based comparison, tradeoffs, and questions to resolve.

Planning · Beginner · AI chat

Before you start

Two or more options, your decision criteria, and approved supporting material. Use only information you can verify.

Step 01

Define the comparison

Explain what matters before asking for a recommendation.

Help me compare [options] for [business need]. My criteria are [criteria], and my constraints are [constraints]. Use this source material: [material]. Create a comparison table. Distinguish source facts from assumptions and unknowns. Refer to the supporting passage for each factual claim. Do not fill missing information with guesses. Ask me about any essential missing criterion.

Step 02

Challenge the leading option

Answer the questions, then test how much the recommendation depends on assumptions.

Given the clarified criteria, explain which option fits best and why. Make the strongest case against it. Identify the assumptions that could change the recommendation, what evidence would resolve them, and a reversible next step. Do not invent numerical scores or treat the recommendation as a final decision.

Step 03 · Review before use

  • Verify material claims against the source documents.
  • Check that the comparison uses your priorities consistently.
  • Resolve important unknowns with the relevant people before deciding.

If you get stuck: If every option looks equally good, name your most important constraint and request explicit tradeoffs.

Back to workflows

Make a repeat task easier to hand off

A draft procedure with inputs, steps, exceptions, and a completion checklist.

Operations · Intermediate · AI chat

Before you start

An approved description of a low-risk repeat task, its expected result, and someone familiar with the process to review it.

Step 01

Describe the real process

Start with what actually happens, including the awkward parts.

Help me document [repeat task] for [person or role]. Here is how we do it today: [steps and notes]. The expected result is [result]. Ask up to five questions about missing inputs, decisions, exceptions, or approval points. Do not assume access to tools or records I have not described.

Step 02

Build a usable procedure

Answer the questions and request a version a colleague can try with sample data.

Turn the confirmed process into a draft procedure. Include purpose, required inputs, numbered steps, expected output at each stage, common exceptions, when to stop and ask a person, and a completion checklist. Mark unresolved details clearly. Add one fictional practice case. Do not add system access, permissions, or company policies that I have not confirmed.

Step 03 · Review before use

  • Have someone follow the procedure using sample data.
  • Record where they hesitate or need missing information, then revise.
  • Have the process owner approve the steps and exceptions before team use.

If you get stuck: If the procedure stays vague, describe one completed example and ask for concrete inputs and outputs at each step.

Back to workflows

Build a weekly update from approved notes

A reusable weekly update with sources and gaps called out.

Communication · Intermediate · AI chat

Before you start

Approved project notes labeled by source and date, the reporting period, and the audience. No account connection is needed.

Step 01

Assemble the evidence

Label your notes so you can trace each claim back to its source.

Prepare a weekly update for [audience] covering [date range]. Use only these labeled notes: [notes with source and date]. Group the result into completed work, work in progress, blockers, and proposed next steps. Cite the note label for each factual item. Preserve the dates and units of any numbers. Flag missing or conflicting information.

Step 02

Edit for a busy reader

Resolve the gaps you can, then ask for a concise final draft.

Using these corrections: [corrections], revise the update to under 300 words. Put the most important blocker or decision first. Keep completed work distinct from plans, and proposed next steps distinct from commitments. Retain source labels and unresolved gaps. Add a short review checklist. Do not distribute the update.

Step 03 · Review before use

  • Verify status, dates, and numbers against the labeled notes.
  • Confirm that next steps match the team’s actual commitments.
  • Save the approved format for next week and replace the source notes each time.

If you get stuck: If old work appears as this week’s progress, restate the date range and require the source date next to every item.

Back to workflows

Build a morning brief that learns company jargon

A daily recap and reviewed glossary.

Operations · Intermediate · Scheduled tasks / Saved context

Before you start

An approved assistant with scheduling, meeting/message access, and editable context files.

Step 01

Set the schedule

Confirm access.

Schedule a morning brief at [time, timezone] using [approved sources] and [glossary file]. Report inaccessible sources.

Step 02

Find the actions

Summarize yesterday’s meetings and overnight messages. Link sources. Extract actions, owners, deadlines, and urgent blockers; mark missing details.

Step 03

Spot unfamiliar terms

List unfamiliar terms and project names. Quote supporting context and flag ambiguous meanings. Do not guess definitions.

Step 04

Ask for clarification

Ask me to define each unfamiliar term. Propose a glossary entry for my review before saving.

Step 05

Keep approved knowledge

Verify saved changes.

Save only approved definitions to [glossary file]. Update the existing scheduled task to load this glossary and repeat these steps, asking before edits. Show saved instructions.

Step 06 · Review before use

  • Verify actions against sources.
  • Confirm approved definitions were saved.
  • Check the next brief uses them.

If you get stuck: Without scheduling or connectors, run manually with approved notes. Without saved context, reattach the glossary next time.

Inspired by ChatPRD’s morning-brief workflow. Prompts above are original Hiatt Co practice material.

Back to workflows

Create a dashboard by describing what you need

An interactive dashboard with traceable metrics.

Data · Intermediate · Data access / File creation

Before you start

An approved agent with dashboard creation and read-only data access, metric definitions, and a reporting period.

Step 01

Describe the question

Plan a dashboard for [business question], [period], [metrics], and [audience]. Clarify definitions; do not query data or build yet.

Step 02

Find approved data

Confirm sources before querying.

Use existing reports first, then approved analytics datasets. Propose new queries only if necessary; obtain review before execution. Preserve access restrictions.

Step 03

Inspect the dashboard

Compare with source records.

Build the dashboard from verified results. Show source references, units, date range, filters, and freshness. Flag missing data; do not fabricate values.

Step 04

Refine the view

Change one thing at a time.

Add a breakdown by [dimension] and adjust [chart]. Preserve metric definitions and explain any change in totals.

Step 05 · Review before use

  • Reconcile totals with approved reports.
  • Test filters and date boundaries.
  • Confirm sharing permissions before publishing.

If you get stuck: Without governed data access, use an approved export. Label it as a snapshot rather than live data.

Inspired by ChatPRD’s natural-language dashboard workflow. Prompts above are original Hiatt Co practice material.

Back to workflows

Set project-specific rules for an AI workflow

A project control checklist and sandbox test record.

Operations · Advanced · Project controls

Before you start

A platform with enforceable project permissions and approval gates, an administrator, a project owner, and sandbox tools. Prompts alone cannot enforce access controls.

Step 01

Define ownership

Draft a project charter for [workflow]. Name [owner], allowed tasks, excluded work, data boundaries, and escalation contact.

Step 02

Scope the project

Have an administrator configure controls.

Recommend project settings for [workflow]: approved model, cost limits, relevant skills, and minimum tool permissions. Flag unsupported controls.

Step 03

Require approvals

Apply policies in platform settings.

Draft policies requiring explicit approval before [sensitive actions]. Identify who may approve, what they must see, and how denial blocks execution.

Step 04

Test the boundary

Use disposable sandbox records.

Test an allowed action, an approval-required action, and a denied action. Verify actual tool execution and logs; record whether controls held.

Step 05 · Review before use

  • Confirm the owner and saved settings.
  • Ensure denial prevents tool execution.
  • Check project isolation with another sandbox project.

If you get stuck: If enforcement fails, disable the affected tool and keep actions manual until controls are fixed.

Inspired by ChatPRD’s project-governance workflow. Prompts above are original Hiatt Co practice material.

Back to workflows

Route CRM leads and draft emails with GPT-6 Astra

A tested draft workflow that assigns leads and queues emails for review.

Sales · Intermediate · Codex / Browser control

Before you start

Codex with GPT-6 Astra and approved browser control, a CRM draft workflow, routing rules, and fictional test leads.

Step 01

Open the builder

Use an inactive workflow.

Inspect [CRM workflow URL] in Chrome. Confirm the workspace and existing nodes. Do not edit yet.

Step 02

Confirm the agent’s scope

Check access before editing.

Confirm you can control this browser. Work only on this draft workflow; keep outbound email and activation disabled.

Step 03

Describe the routing

Route leads using [rules]. Draft a personalized email from the assigned owner with [owner-specific calendar links] and [approved talking points]. Send drafts to [review queue], never directly to leads.

Step 04

Build the draft

Add and connect the required nodes. Preserve unrelated logic. Flag missing routing decisions instead of guessing. Keep the workflow inactive.

Step 05

Test each branch

Use fictional submissions.

Test every route, missing fields, and repeat submissions. Verify owner, sender, calendar link, and review destination. Show results before activation.

Step 06 · Review before use

  • Confirm assignments and personalized drafts.
  • Verify review happens before email delivery.
  • Approve activation separately.

If you get stuck: If browser editing fails, request node-by-node instructions for manual setup.

Inspired by ChatPRD’s CRM lead-routing workflow. Prompts above are original Hiatt Co practice material.

Back to workflows

Review Copy

Revised copy, or focused findings when you ask only for critique.

Communication · Beginner · AI chat

Before you start

A draft approved for your AI tool, its audience, and the intended tone. Keep private details out of practice material.

Step 01

Review the prose

Copy this prompt, then paste your draft and name its audience and intended tone. Ask for critique only if you want findings instead of edits.

Make every word justify its existence. Apply these rules to prose, never to code. Keep technical terms intact; use everyday words elsewhere only where precision survives.

## Writing rules

1. Never use a metaphor, simile or other figure of speech which you are used to seeing in print.
2. Never use a long word where a short one will do.
3. If it is possible to cut a word out, always cut it out.
4. Never use the passive where you can use the active.
5. Never use a foreign phrase, a scientific word or a jargon word if you can think of an everyday English equivalent.
6. Break any of these rules sooner than say anything outright barbarous.

## Review

Read for meaning, audience, and tone before editing. Keep the author’s claims, intent, and voice. Preserve facts, names, numbers, links, qualifications, and uncertainty. Do not invent an actor to force an active sentence or strengthen a claim by cutting a needed qualifier.

Cut filler, repetition, stock phrases, and needless transitions. Choose short, concrete words and direct verbs. Keep enough context for the reader to follow the thought. Favor natural prose over rigid rules or the shortest possible text.

Leave code, commands, identifiers, technical terms, and attributed quotations unchanged unless the user asks to edit them.

Return the revised copy by default. If the user asks only for critique, give focused findings instead. Add notes only when requested or needed to flag an ambiguity or a change in meaning; keep them brief.

Before delivering, review every prose output, including any notes, against all six rules. Ask of each word: does it add meaning, preserve precision, or help the sentence read naturally? Cut it if it does none of these.

Step 02 · Review before use

  • Compare claims, facts, names, numbers, links, and qualifications with the draft.
  • Confirm code, commands, identifiers, technical terms, and attributed quotations remain unchanged.
  • Read aloud to check that the voice and meaning survive the cuts.

If you get stuck: If an edit changes the meaning or sounds forced, point to the passage and ask for a revision that preserves the original claim and tone.

Back to workflows

Turn course notes into a customer sample

A Markdown course notebook, a shareable HTML sample, and a reusable review checklist.

Planning · Intermediate · AI chat / File creation

Before you start

A course idea, audience, business tasks, and initial notes. Use an AI tool that can create and revise Markdown and HTML files. Parallel agents are optional.

Example course layout

Step 01

Create the first sample

Replace the starting-context placeholders with your notes. Continue in the same conversation when giving feedback.

Help me create a sample course layout for prospective Hiatt Co customers. Hiatt Co teaches small-business owners and teams how to use AI in their work.

The example should help a customer understand who the course serves, what participants will practice, what they will leave with, and how we could adapt it to their team.

Create a customer-facing example, not a complete curriculum or a confirmed offer.

Starting context:

- Course idea: [topic or working title]
- Audience and current skills: [who will attend]
- Business tasks they want to improve: [tasks]
- Format and available time: [format, duration, or undecided]
- Tools participants can use: [tools or undecided]
- My initial braindump: [paste notes or dictated text]

1. Build a shared course notebook

Create a small Markdown wiki for this course. Include:

- Brief: audience, business need, scope, and constraints.
- Learning outcomes: what participants should be able to do.
- Course outline: sections, sequence, and proposed timing.
- Decisions: confirmed choices, assumptions, and open questions.
- Feedback: my comments and how they affect the course.

Treat my notes as source material. Separate confirmed decisions from suggestions and unresolved ideas. Do not turn tentative remarks into commitments. Ask only questions whose answers would materially change the course; otherwise proceed with labeled assumptions.

2. Develop the sections

Break the course into sections that can be drafted independently.

For each section, define:

- A clear title and practical purpose.
- An observable learning outcome.
- A short explanation or demonstration.
- A hands-on exercise using fictional or anonymized business material.
- A concrete output participants will create.
- A review check that shows whether the exercise worked.
- Proposed timing and prerequisites.

If parallel agents are available, assign each a separate section file and have them read the shared brief first. Keep shared decisions under one coordinator. Reconcile terminology, difficulty, dependencies, and timing before combining their work.

3. Turn the outline into a customer example

Create a self-contained HTML page I can open locally and share with a prospective customer.

Show:

- Course title and a plain-language description.
- Intended audience and prerequisites.
- What participants will practice and leave with.
- An agenda with expandable section details.
- One representative exercise with an example prompt and expected output.
- What the customer would need to provide.
- Options for adapting the course to different teams.

Make it readable on phones, printable, and usable with a keyboard. Keep assets local and avoid external dependencies.

Label it “Sample course outline — adapted after a planning conversation.” Clearly mark proposed durations and tool requirements. Do not invent prices, testimonials, credentials, guaranteed results, or approved commitments.

Keep internal planning notes out of the customer page.

4. Integrate my feedback

After each draft, give me the HTML page and a brief report of changes and unresolved decisions.

I will respond with dictated braindumps. Integrate them into the notebook, flag contradictions, and update affected sections. Preserve earlier decisions unless I change them. Keep a short decision history.

5. Run a course review

Create a reusable review checklist or skill for the whole course. Run it before presenting the first draft and after substantial revisions.

If agents are available, use independent reviewers for learning design, business relevance, timing, accuracy, accessibility, and copy. Look for:

- Vague or untestable outcomes.
- Missing prerequisites or gaps in the sequence.
- Exercises that do not support the stated outcomes.
- Too much material for the proposed time.
- Unsupported claims or unconfirmed tool capabilities.
- Repetition, filler, jargon, and sales language.
- Differences between the course plan and customer page.

Fix supported findings and rerun affected checks. Bring unresolved decisions to me. Do not schedule unattended work unless I ask.

Start by organizing my initial braindump and producing the first sample layout.

Step 02 · Review before use

  • Confirm the audience, outcomes, exercises, outputs, and prerequisites agree across the notebook and customer page.
  • Check the agenda fits the proposed time and clearly labels assumptions, tool requirements, and unresolved choices.
  • Open the HTML on desktop and phone, use the agenda with a keyboard, and check print preview.
  • Keep internal notes out of the customer page and confirm the sample label appears without unsupported claims or commitments.

If you get stuck: If the tool cannot create files, request each Markdown file and the self-contained HTML as separate code blocks to save locally. Without parallel agents, draft and review the sections in sequence.

Back to workflows

Use an approved tool with sample or anonymized information. Work through each workflow’s prompts in the same conversation and review the output before use.