Human
Keep the work with people, by choice. Some work calls for human presence, responsibility, empathy or direct experience.
Business example: a person holds a sensitive conversation with a customer who is considering leaving.
Discovery guideDiscovery · Your business AI practice
Leave with a business workspace, a clear aim and one useful result you have checked.
Understand the Five Ways and four parts of the Bolt Method. Then set up your AI and apply the Method to one piece of work. Work with your facilitator or take it at your own pace.
Start the guide → Resume where I left off →Have your Claude or ChatGPT sign-in, a charged device and one small business task ready. Keep this guide and your AI open alongside each other. No sign-in is needed for this guide.
These choices tailor the setup tips. Change them here at any time.
Bolt Method · Start here
Where does AI belong in the work in front of you?
Bolt Method™: Five Ways. Four parts. The Five Ways answer where AI belongs. Values, Judgement, Thinking and Habits shape how the work is done. Human accountability runs throughout.
These are equal choices, not levels to climb. A Way belongs to the work, not the person. One business project may use several Ways.
Keep the work with people, by choice. Some work calls for human presence, responsibility, empathy or direct experience.
Business example: a person holds a sensitive conversation with a customer who is considering leaving.
Use AI for a defined task under your direction. You choose the task, provide the relevant input and check the result.
Business example: AI groups approved customer comments into themes for you to review.
Delegate a bounded sequence, with limits and checks. Give the agent a clear outcome, the resources it may use and the points where it needs human review.
Business example: an agent works through approved customer feedback and prepares a report for review before anything is shared.
Develop thinking through sustained context and interaction. Work with AI over successive conversations to explore questions, challenge assumptions and refine your understanding.
Business example: an owner explores the possible causes of customer churn across sessions, using relevant context and checked handovers.
Coordinate people, AI partners and agents within a directed system. Different contributors have defined roles, with human oversight of how their work fits together.
Business example: a team coordinates customer conversations, feedback analysis, reporting and decisions as one retention effort.
The Method is independent of the AI product you use. It supports better thinking, decisions and outcomes, but does not guarantee results or replace expertise and human relationships.
This guide follows the Bolt Method V5 Training Guide: Five Ways, then Values, Judgement, Thinking and Habits. The setup instructions that follow help you put it into practice.
Part 1 · You embed
What must remain true, whichever Way you choose.
The five Values give you questions to bring to the work and the people it affects:
These require human interpretation. They help you examine choices and explain trade-offs; adding them to AI instructions does not guarantee that every answer will reflect them.
Part 2 · You judge
Decide what to trust, question, verify, reject or act upon.
Judgement is the human work of deciding what to trust, question, verify, reject or act upon. AI can produce a confident answer that still needs checking. Your own assumptions also deserve scrutiny.
The Bias Matrix helps you notice how human and AI bias can distort the work. Its aim is Tested Judgement: surface assumptions, question them and test them before deciding.
Tested Judgement does not mean bias disappears. Agreement between two AI responses is not independent evidence. The Five Ways help you decide where AI belongs in each task.
Part 3 · You apply
Frame questions, explore possibilities, make connections and develop ideas.
Define the Aim comes first: clarify the result you want, why it matters and what evidence would show success.
Once the aim is clear, move through:
You can return to an earlier Stage when new evidence appears. The downloadable skills later in this guide help you use this process on your own work.
For simple work, keep the Stages brief. For substantial work, move through them deliberately. Review what happened against your aim and return to Gather when new evidence changes the work.
Define the Aim helps you agree what you want to achieve and why. Results Cycle helps you carry that aim through Gather, Think, Decide, Do and Review. You will install and use these two skills in the practical sections below.
Part 4 · You practise
The repeated behaviours that carry Bolt Method into daily work and daily life.
The Habits help you bring the Bolt Method into everyday business work: express your thinking, give useful context, carry learning forward, choose effective tools and consider different perspectives. Discuss them with your facilitator, or use the business examples to connect each Habit to your own work.
Use voice to develop your thinking. Speak a first explanation, listen to or read it back and notice what is missing. Choose a safe setting where you can pay attention to your surroundings. Typing is equally valid when it suits you.
In business: An owner speaks a two-minute account of a difficult decision, then asks for the competing priorities to be separated.
Share relevant information about the aim, constraints, audience and evidence. Better context can improve the work. Use judgement about privacy, confidentiality and data sensitivity; more disclosure is not automatically better.
In business: A vague request for a proposal becomes a brief with a named audience, approved evidence, budget limits and success criteria.
Capture and reuse useful context. Keep dated briefs, confirmed decisions and open questions. Check what your platform actually saves and can retrieve.
In business: A participant saves a checked handover and supplies it at the start of the next session.
Keep your AI capability current and use what suits the work. Test a tool against a real task before adding it to your routine.
In business: A team improves its existing assistant brief before buying another service.
Bring sceptics, specialists and contrary perspectives into the work. AI-generated perspectives are simulations, not consultations with real experts. Check material claims and involve the right people.
In business: Before changing pricing, an owner examines customer, operations and finance perspectives, then tests assumptions with real evidence.
Choose your setup
Start with a sensible subscription. Choose the model and thinking effort to fit the work.
These are individual plans. Team and business accounts have different controls and allowances. Prices below are US dollar reference prices, checked on 18 September 2026. Check your local checkout for currency, taxes and the amount you will actually pay.
Free · $0
Suitable for trying AI. Limited usage; Sonnet and Haiku, but no Opus or Fable. Projects and skills are available, subject to account limits.
Pro · $20/month
Our recommended starting paid plan. More usage and broader model access. Fable requires usage credits on Pro.
Max · from $100/month
Higher allowances for frequent use. Available in 5× and 20× Pro usage options. Premium models can have their own allowances; this is not unlimited use.
Free · $0
Useful for everyday text conversations. Uploads, research and other tools have tighter allowances. Advanced paid model choices are unavailable. Projects are available.
Plus · $20/month
Our recommended starting paid plan: broader reasoning access and expanded file, research and other tools. Limits still apply. A cheaper Go plan also exists, with fewer advanced reasoning options.
Pro · $100 or $200/month
Higher usage and access to Pro models. New subscriptions and upgrades to the $200 tier are currently paused; existing subscriptions continue. The $100 tier remains available.
A model is the AI engine you are using. Our task examples below are starting suggestions: compare the result with your aim, rather than assuming the most expensive option is always best.
| Model family | Try it for | Trade-off |
|---|---|---|
| Haiku | Quick sorting, short rewrites, simple first drafts. | Fast and economical; use a stronger model for difficult judgement or multi-step analysis. |
| Sonnet | Everyday writing, planning, document analysis and business work. | A useful balance of capability, speed and usage. A good everyday starting point. |
| Opus | Complex analysis, difficult comparisons and challenging a substantial proposal. | Greater capability for demanding work; consumes more allowance. |
| Fable | Very demanding reasoning or long, multi-stage agent work. | High resource use and plan-specific access or credits. Not necessary for today’s first task. |
Current versions: Haiku 4.5, Sonnet 5, Opus 5 and Fable 5.1. Official model and consumption guide ↗
Free and Go use GPT-5.6 Luna. Eligible paid accounts use GPT-5.6 Sol for Instant and thinking levels. Pro choices include GPT-5.6 Sol Pro and GPT-6 Pro, powered by Astra.
Try everyday chat for a short rewrite; use reasoning for comparing options with several constraints. Reserve Pro for difficult, extended work. Terra is a balanced option in Work and Codex, rather than a selectable model in ordinary ChatGPT conversations. Plus includes Astra in Work and Codex, but not GPT-6 Pro in ordinary chat.
More thinking can help with complexity, but may take longer and use more of your allowance. It cannot replace missing facts or human checking.
Open the model menu beside the send button. Depending on the model, you may see Low, Medium, High, Extra High or Max effort and a thinking control.
Low or Medium suits routine work; High gives more room for a difficult problem. Extra High and Max are for demanding tasks where the extra time and usage are worthwhile. Fable 5.1 and Opus 5 always use thinking in Claude; effort and the thinking switch are separate controls.
Claude thinking controls ↗Instant: quick everyday replies. Medium: standard reasoning. High: more reasoning. Extra High and Pro: demanding work on eligible plans.
Plus includes Medium and High. Free and Go offer Think using Luna. Choose the setting in the model controls: typing “think harder” does not change the selected level on Plus or Pro.
ChatGPT thinking controls ↗Account, model and thinking are three different choices. Menus and availability change; your account’s controls and the linked official pages are the current reference.
Setup · 1 of 4
Start with the equipment and account you already have.
Prepare a 20-minute fundraising planning meeting for four volunteers. This is a fictional example, not a real fundraising result.
Check that you are signed into the intended work account. If sign-in fails, use account recovery or ask your facilitator. You can read the guide without an AI account.
A chat is one conversation with AI. A project groups related conversations, instructions and reference material around a purpose. You can start with a chat and organise projects later.
Setup · 2 of 4
Give your AI a clear set of instructions for how you want to work together.
A guide to the sequence, not a screenshot of your AI’s menus.
After the test prompt, the AI asks what the meeting needs to achieve before it drafts an agenda. You check that the Method text is still saved.
Use the compact Method if the instructions field is too short. If the setting is missing, paste the Method at the start of a dedicated chat and continue there.
Bolt Method brings together Values, Judgement, Thinking and Habits. Its guiding question is: Where does AI belong in the work in front of you?
Use this compact version instead.
Open a new, dedicated chat and paste the Method text as your first message. It applies to the material you supply in that conversation. Paste it again when you begin a new chat.
Start a fresh conversation covered by those instructions. For the chat-only route, paste the Method first. Then send:
Look for a question about what you want to achieve before the AI starts drafting. Explain your task and check its response against the instructions. For substantial factual replies, the text requests a confidence rating, a reason and what remains unverified. A rating is a rough self-assessment, not proof of accuracy.
Setup · 3 of 4
Set useful preferences, then create a home for related work. Allow about 30 minutes for this step.
A guide to the sequence, not a screenshot of your AI’s menus.
Project: “Community fundraising”. Brief: a proposed £5,000 target over three months, four volunteers and no confirmed donations. A target is not money raised.
If Projects is missing, use a dedicated chat and keep your project brief in a document. Ask your administrator about workspace restrictions before buying another account.
Describe your role and how you like to work. Replace the square-bracketed prompts below. Add the result alongside your Method instructions.
Find Projects or the equivalent workspace feature. Give it a clear name, check who can access it, and add the Method, your preferences and a project brief. Use this project for the business work you want to focus on.
Add one short, appropriate reference document or paste a brief. Name and date it. Ask the AI a question whose answer is in that material, then check the response against the original.
Inspect the settings available to you and learn what can be corrected or removed. Keeping automatic memory off is a valid choice. A project helps organisation but does not, by itself, establish a privacy boundary.
Setup · 4 of 4
Use Define the Aim to choose the result you want, then the Results Cycle to carry it through.
A guide to the sequence, not a screenshot of your AI’s menus.
“Prepare a one-page agenda for a 20-minute meeting that helps four volunteers agree three next actions, with owners and dates.” Gather first: confirm the target and what the team can commit to.
In Claude, keep the download zipped and use the Skills upload screen. In ChatGPT, use the MD conversation route described below. If upload fails, open the MD file and paste its text.
A skill is a reusable set of instructions that gives your AI a way of working. Your project holds the context for your work; a skill guides the process. You can work through this step with a facilitator or follow the instructions yourself.
Get clear about what you want to achieve, why it matters and how you will recognise success. The AI asks one question at a time, helps you consider your boundaries and offers wording for you to refine. You finish with an agreed aim before moving into planning.
Use it when: you are starting a task, feel unsure about the goal or need to change direction.
You leave with: your aim, the reason it matters, evidence of success and your non-negotiables.
Take your agreed aim through five Stages: Gather the relevant information; Think through the options; Decide what to do; Do the work; and Review what happened. The cycle helps you avoid rushing into action and makes learning part of the work.
Use it when: your aim is clear and you want to turn it into a useful result.
You leave with: a considered decision, completed work within the authority you give, and a review of the result and next step. You remain responsible for the choices.
Upload the ZIPs through the Skills screen. Adding them as project reference files or chat attachments does not install them as skills.
Menu names can change. Claude’s official skill installation guide ↗
Once the skills are available, continue to Define the Aim to use and test them on your chosen task.
Check you are in the right Claude account, try the website on a laptop, and check that code execution is enabled. If you use a work account, ask your administrator whether user-created skills are allowed. Upload one original ZIP at a time. If your browser opens an MD file rather than saving it, use its Save or Download option.
You can continue with the Markdown route below while access is resolved.
This uses the skill’s instructions in the conversation. It does not install a native skill in that AI. If the AI cannot read the attachment, paste the text instead.
Apply the Method · Start with the aim
Agree what you want to achieve and why, before planning or producing the work.
Choose one manageable business task. Clarify the intended outcome, why it matters, what would count as success and what must not be sacrificed.
“Prepare a one-page agenda for a 20-minute meeting that helps four volunteers agree three next actions, with owners and dates.” Success means a usable agenda; it does not mean the fundraising target has been achieved.
Check: the AI asks a focused question rather than immediately producing the finished work. Answer its questions, refine the wording and confirm the aim. Look for an indication that it used the skill where Claude shows this; a matching response alone does not prove installation.
Make the task smaller and ask the AI one question at a time. Keep refining the wording until it describes the result you want. You decide when the aim is clear enough.
Apply the Method · Results Cycle · 1
Collect the relevant information and build a shared understanding before reasoning on it.
Audience: four volunteers. Output: one-page agenda. Known: 20 minutes available. Assumption: a £5,000 target is realistic. Unknown: available volunteer hours. Ask about that before planning the campaign.
Start with the task, audience and desired result. Mark anything you do not know as unknown. Ask the AI which single missing detail matters most.
Bring forward your agreed aim. Give relevant context: the information, sources, constraints and unknowns that matter to this work. Keep facts separate from assumptions.
Check: it begins with the aim and the information needed, makes the current Stage clear and keeps you involved in decisions. Continue through Think, Decide, Do and Review as the work progresses.
If you need a smaller task, consider: a weekly work plan, a meeting agenda, a clearer business email or a short supplier comparison. Use information you are comfortable sharing. An anonymised or fictional example is enough to practise.
Paste your brief into a conversation in the relevant project. Ask the AI to explain what it understands and what is missing. Correct any errors, then agree when you have enough information to move to Think.
Apply the Method · Results Cycle · 2–5
Use what you gathered to make a deliberate decision, produce the work and review the result.
Agenda: purpose and target 3 minutes; evidence and constraints 5; compare options 7; agree actions, owners and dates 5. Check: 3 + 5 + 7 + 5 = 20. Suggested actions remain proposals until the team agrees them.
Make the task smaller. Check one important fact against its source, correct the brief and ask for a revision. A confident answer is not evidence.
Keep your agreed aim in view. You have started with Gather; now continue the Results Cycle.
Explore options, make connections and weigh the trade-offs. Question assumptions, including your own.
Choose what to do and why. The human decides; the AI informs. Make any limits or permissions clear.
Produce the agreed piece of work. Return to Think or Decide if the approach proves wrong.
Compare the result with your aim. Check what worked, what did not and what you learned. Return to Gather for the next cycle.
Revise the result and save the useful version. If you are still stuck, narrow the task or ask which missing information matters most.
Maintain Memory · Continue the work
Capture the working state so your next conversation can pick up where you left off.
Agreed: the meeting aim and 20-minute agenda. Open: volunteer availability and whether the target is realistic. Next: the organiser checks availability before the meeting. Save the agenda and brief with the handover.
Ask for a short handover if the result is too long. Read and correct it, save it, then supply it and the necessary source files in the next chat.
The Context Handover Skill (labelled handover in these downloads) creates a working brief for the next session. It preserves the decisions, reasoning, useful outputs and unresolved questions that matter for continuing the work.
A brief exchange may only need a short note. Use a full handover when decisions, reasoning or open threads would otherwise need substantial re-explanation.
ZIP: the ready-to-install Claude skill. MD: the same skill instructions as a readable Markdown file. These downloads tell your AI how to create a handover; your own session handover is produced when you run the skill.
Install the ZIP using the skill installation instructions, then enable handover in your Skills list. If you are using the Markdown route, attach the MD file to your conversation or paste its contents and ask the AI to follow it.
In the conversation you want to preserve, ask:
Read the result, correct anything inaccurate and save a dated copy. In your next conversation, attach or paste that checked handover along with any source material it needs. Explain what you want to do next.
You can also ask for a short handover, a handover for another AI platform or a downloadable Markdown handover.
Record what changed and what you learned. Keep estimates separate from achieved benefits, then choose the next piece of work and when to return to it.
A useful result. A clear next step. A practice you can build.
Graduation · Take your learning forward
Capture what you achieved and your next practical step.
Write your own record below. This guide does not read your AI conversations, so bring across the useful parts yourself. Downloads include these entries and your working brief.
Need a paper copy? Open the guide menu and choose Print this guide. Your entered notes are included.
Your summary is a Markdown text document you can keep, edit or bring into your next AI conversation. Record achieved benefits separately from estimates.