The Claude AI Money Bible Key Takeaways

Start on the free plan through Book 3, upgrade to Pro before the product module or Claude Code/Cowork tools, and skip Max entirely your first few weeks.

Choose the strongest model your plan gives you for any task that matters, and use the lighter models only for throwaway jobs, or you will get average output and unfairly blame the tool.

Treat any confident-sounding number, price, date, or "studies show" claim as unverified until you check it, because Claude can be wrong calmly, knows nothing after its cutoff without web search, and learns only what you tell it.

Run a quick sanity check whenever output feels shallow: confirm the model name in the dropdown, then compare the lightest and strongest models on the same prompt before trusting either.

After roughly two months of daily use, switch from monthly to yearly billing and hang onto the receipt for tax purposes.

Install Claude only from the official App Store or Play Store listing with Anthropic’s dark icon and orange starburst, and sign in with the same email you use on your computer; the two silent failure modes are lookalike apps and a split account.

Put the phone app on your home screen next to apps you open daily, since that placement is the difference between a tool you actually use and one you forget.

Install the desktop app on a personal computer so you get one-click file access and the doorway to Claude Code and Cowork, but when a work computer’s IT policy forbids it, a bookmarked claude.ai handles most of this book.

Match the device to the moment: capture ideas, photos, voice notes, and quick replies on the phone, and produce files, long outputs, and customer-facing work on the computer, none of which works well on the other device.

Keep a single project named Inbox for every random capture and sort it on your computer at the end of the week, so an idea taken on the move reliably turns into something usable.

Use the phone's dictation button to ramble through a problem aloud, then ask for the ramble to be compressed into a plan or decisions when you are done.

Create a Project for anything you will return to more than twice, and fill its instructions with who you are, who your customers are, your voice, three banned phrases, and one required ending.

Upload your menu, price list, or other reference documents to the Project's knowledge area so answers use your real details instead of invented ones; keep the set under ten files and replace outdated files rather than adding second versions.

Match Project size to what you produce: one per client or per subject, plus an Inbox for random captures, and keep the total around five to eight so none becomes forgettable.

Keep the five instruction lines as a reusable phone note, and every future Project setup drops from fifteen minutes to three.

Feed Claude the actual document, not what you remember of it, and treat the answer as preparation: for anything legally binding, carry it to a professional before you act.

Run the three questions as a fixed routine before you commit to any document: a plain translation, a hunt for every cost that could come due later, and a comparison with the standard shape of such paperwork. The third pass tells you which clauses are genuinely worth fighting over.

Ask for the contract's three harshest clauses rewritten the way you would prefer, so you walk into the negotiation with concrete wording instead of a vague objection.

Point the money question at subscriptions you already pay for and at any paper document handed to you in person; forgotten renewals and hidden charges surface there while you can still act on them.

Keep each photo to a single page, taken head-on and with good light, and verify every name and figure yourself, since a blurry capture is precisely the input failure that turns a careful reading into a guess.

Turn on web search and open time-sensitive requests with "Search the web..."; replying "source?" to any figure will either produce a link or expose that the number came from memory.

Ask for anything larger than a paragraph as an artifact: a real file you can copy, download, refine, and share, with one artifact per document.

When output is destined for a client, request Word or PDF and have the working version rebuilt as a clean, presentable document.

Write four lines into your personal preferences covering who you are, brevity, a concrete example, and no padding; unlike a Project, these apply to every chat.

Add a preference that Claude ask one question before answering, so it stops guessing your intent; verify it in a fresh chat outside any Project.

Add a permanent rule block to every AI writing request covering words and habits alike: no "delve," "unlock," or "elevate," no restating the question, no opinion-free on-the-one-hand passages, no three-item runs, no closing summary, no uniform sentence lengths.

Keep a private list of words that sound dead to you and append it to the block, since generic bans never cover what specifically reads as machine-made to your ear.

Test any generated passage by reading it aloud next to a paragraph you wrote yourself, then tune the bans until the machine voice disappears without the prose going clipped.

Give an AI three samples of your best casual writing, such as a long text to a friend, an email you were happy with, or a post that drew replies, and have it extract the patterns that define your style.

Fill in WHO, WHAT, SIZE, and SOUND before you send any request, because what you leave out is exactly what the answer will get wrong.

Keep the awkward context in the brief: a sanitized prompt flattens the output, while an honest problem like a late job or a price rise produces copy that actually works.

Add “Give me three versions: one safe, one bold, one unusual” to short requests so you are choosing between options rather than judging a single answer in isolation.

Save the five everyday prompt templates in a phone note, since most requests you write are a variation of apologizing, announcing, applying, complaining, or deciding.

After you get a strong version, ask for it at half the length: the elements that survive the cut are the ones you could not afford to lose.

Name the specific expert and your real constraints: a hiring manager who reads two hundred CVs a week will catch what "please review my CV" never does, and a plan built on your true hours, budget, and missing skills is one you can actually run.

Match the request to where the material lives: convert content that is right but stuck in the wrong format, translate expert-speak for a smart non-specialist, and let the assistant interview you one question at a time when the material is in your head rather than on paper.

Before anything goes public or any money moves, ask for every problem ordered worst to least damaging with a one-line fix each, and separately for the strongest case against your own plan; the criticism costs one message, and a strong objection may have just cancelled a costly mistake.

When you want options or volume, force variety by asking for more than you want in outputs that differ in kind, make the model bet on the best few to force reasoning, and if the first crop feels samey repeat the request with a demand for stranger answers; keep runs to fifteen items per request because quality slides in longer pushes.

Chain the ten shapes in a single afternoon: twenty ideas, a monthly plan, thirty captions in two batches of fifteen, and a harsh critique before publishing is work an agency would bill hundreds for, and every output should land in a dedicated templates project so it becomes reusable, then sellable.

Before asking for a revision, name the exact flaw; if you can't, request five guesses and correct the likeliest one.

Fix only one dimension at a time, and respect the order: structure, content, tone, length.

When three fixes have failed, abandon the thread and issue one fresh prompt that folds in everything learned.

Verify every number, date, name, law, or "studies show" claim in the output, and mark anything you cannot source; skip that burden for opinions, general advice, and known business facts.

Track minutes per finished piece from the start, and when a piece runs long, write down why; a missing Project detail is usually the cause.

Restrict each email to one action and one link, and make the subject line and opening sentence carry all the persuasion.

From one product, generate ten selling angles, test only two against each other, and ask a real customer which they'd choose.

Start a product description with the emotional payoff, then state the specs and the guarantees, and skip hollow labels like premium.

Order a sales page as problem, promise, proof, offer, price framing, doubts, and a single next step, with no "About us" opening and no hidden price.

Treat trust as the business's only compounding asset, so never make a claim you cannot prove.

Generate ten article topics from Google's autocomplete and AI suggestions, but commit only to those whose searches return real results, and favor cost questions for their buying intent.

Open every article with the direct answer, including a price range, then explain under simple headings, and close with a single practical next step.

Write at 800 to 1,200 words and take the time to fact-check each piece; three thorough articles will beat ten thin ones.

Spend ten minutes adding a real observation and a locally specific detail so the article passes both Google's machine-text detection and the reader's own judgment.

Assign one of the six post types to each weekday so your account balances selling, trust, proof, and engagement without ever needing a fresh invention daily.

Produce the month in three passes: a ten-minute list of thirty one-line ideas, captions written in two batches of fifteen, and a non-optional human pass that deletes flat lines and verifies facts before anything is scheduled.

Schedule the whole month in one sitting with free tools, leave two empty slots per week for live moments, and treat reliability as what makes the system work, not brilliance.

Prepare a printable answer sheet for the twelve most common customer questions, and write review replies under sixty words: specific, calm, and never argumentative, since the public reply sells to the hundred silent readers.

Offer review-reply handling in your content packages: the ten weekly minutes spent on replies lifts more weight off an owner than the posts themselves.

Use a single chat thread to generate all the formats from one article, so the tone and details stay coherent across posts, emails, and scripts.

Stagger the release of derived pieces and reword each one, rather than publishing them at once or pasting identical copy.

Read any video script aloud before using it, and save the multiplier prompt as a reusable template for future articles.

At the first client meeting, ask for their archives of old posts, newsletters, and reviews, then repurpose the best material while skipping anything that flopped.

Book a recurring two-hour weekly slot for the whole pipeline, and expect the process to tighten from three hours to two as it becomes routine.

Set your scope before you install: Claude Code is for modest projects that solve one problem, not for a consumer product people will use at scale.

Ask Claude chat for the official Anthropic install commands for your operating system, paste them into the terminal, and interpret red text as noise; a fresh line is your signal that installation finished.

Create a desktop folder called builds, navigate to it with cd Desktop/builds, and launch claude from there so every file it creates stays neatly in that folder.

Give your first job in ordinary language, specifying color scheme, layout, and features like a contact form, and you will get real files you can open, edit, and share.

Screenshot the broken page, describe the problem in plain words, and name the device you used; after a fix, reload on that same device before checking the other one.

Open any existing-site conversation with an audit: have Claude act as a customer, name the five blockers and the three highest-impact changes, and after backing up the live site fix only the complaint you were given.

When you are asked to improve a site, spend your effort on the essentials: put the phone number above the fold, say what and where, show real photos, list opening hours, and make one action impossible to miss.

Treat the five beginner errors as two-message fixes: paste the exact command or error text, ask for a plain explanation and the next step, and never delete and rebuild; keep a running error log so you can charge after ten entries.

When genuinely stuck, ask for the plan first, restore the last working version, shrink the problem to a single section in a separate file, and after three failed attempts switch to the simplest beginner alternative or walk away for an hour.

Use Cowork only for repetitive, multi-file, multi-step tasks and keep single-item questions in Chat.

Give a job one folder, never your entire desktop, and protect yourself by backing up originals and excluding passwords before launch.

For your first job, have Cowork ingest the customer emails and write customer-insights.md with the five common questions, three complaints, and five exact phrases, then verify its result against two of the source files.

Time the manual version of five files and let Cowork tackle the rest, so you can quote the minutes saved when you offer this as a service.

Build a reusable one-page template for recurring reports, then feed it fresh data with a top summary and a comparison to last month.

Review the generated report yourself before sending, add the covering email to the prompt, and check the output against last month's manual version.

Structure every quote from a template that includes exclusions and a validity date, open with the client's own words, and log accepted price points to learn what wins.

For messy spreadsheets, describe the desired outcome in plain language, work on a copy, spot-check the matching, and verify three random rows against the source.

Choose your first automation target by scoring recurring tasks on duration, frequency, and how much you cringe at them, then pick the dull, document-heavy one that needs no judgment; if the steps will not fit a numbered list, let Claude interview you to surface the hidden ones.

Save each routine as its own recipe-style file, kick it off with a single line, and append a correction step after every run; recurring mistakes fade out by the third or fourth run this way.

Let Claude read, triage, and draft from email and calendar but never send; reviewing before sending is non-negotiable, because one wrongly worded sentence sent in your name forfeits a customer.

Block twenty quiet minutes on Sunday evening for a fixed sequence: review unkept promises from the week, find real free hours, pick three priorities, run your routines, and print a one-page plan. You want to decide while calm, not in Monday's rush.

Compare planned versus done at week's end, and sustain that comparison for four weeks; the pattern it exposes will change your working habits more than any productivity book.

Treat an agent as a coworker with a goal, not a smarter chat: give it a task, tools, and permission to take multiple steps before reporting back.

Use agents for repetitive checking and scheduled multi-step work, but never for judgment calls about people, taste, or anything you cannot explain precisely.

Only build an agent for a task you've done manually at least five times, with exception-laden rules, a stopping point, and a human approval checkpoint.

Write the spec as five lines: goal, numbered steps, never, stop when, and ask me before, then have someone else read it to expose missing answers.

Put the "Ask me before" line at the center of your pitch, since it turns owners' fear of automation acting in their name into trust.

Build a Skill for one recurring job at a time, and write it like a small-business handbook, not a project description, so it activates only for its targeted task.

Choose your first Skills from three weekly tasks that always follow the same steps, and name each Skill by the job it does rather than by a version.

When you create a Skill, use two or three real past outputs as teaching examples, and after a day has passed, test it on a new case to find the missing rules.

If a Skill's output is wrong, revise the Skill itself rather than editing the output, and keep revising until two or three corrections are enough.

Over five weeks, add one Skill per week for the five high-value jobs (customer reply, quoting, content batch, weekly report, welcome) so most of your weekly busywork gets automated.

Build the research agent first, because it is the only one of the three whose output is pure information; failure costs you time, not trust.

Limit the watcher to two targets and set thresholds so high that a quiet week is just a dated no-change line; the system earns its keep through the morning routine it removes from your day.

Use the content approval session as a diagnostic: fifteen minutes with fewer than three changes means the voice rules are sound, while a half-rewrite tells you to fix the rules, not the writer.

Set an agent's permissions as the inverse of its worst plausible damage; anything touching money, customers, or reputation stays behind your final click.

Lead sales with what the agent will not do; the promise that nothing ever sends, posts, or pays without your approval turns an owner's fear into adoption.

Answer the five-question quiz honestly as your current self, not the person you hope to become, then write the chosen path somewhere visible and stick to it for 90 days before adding anything else.

To get paid fastest, start with freelancing or AI setups built on knowledge you already have; if steady repeat income matters more, choose content services; and if you can wait for money, pick digital products or Amazon books.

Use the quiz's three clusters to shortcut your choice: comfortable talking to strangers about money and have consistent hours means content or AI setups, avoiding cold calls points to platform freelancing, and wanting quiet work in small time slots leads to products or publishing.

The skills from Books 1 through 6 are ready to sell as finished items, but a stranger needs proof, so create one visible portfolio piece from a real business, even if unpaid, and show the best work first whenever someone asks what you do.

Skip waiting for confidence; if you left tasks unfinished, complete one task now because reading alone counts for nothing, and keep a single portfolio folder with your strongest item on top.

Narrow your offer to one service paired with one audience, so you can finish the sentence "I write [what] for [who] so they can [result]" before contacting anyone.

Run the 14-day plan as a timed sequence: produce three samples, assemble a proof folder, message ten acquaintances, follow up, and take the first reasonable job rather than waiting for certainty.

Treat a closed deal as the start of the relationship: confirm the scope and price in writing, acknowledge the job the same day, deliver a day early in a proper file, and include one useful observation plus a suggested next step.

Collect feedback deliberately after delivery: ask what they would change, follow up a week later, and when praise arrives, request two lines of written proof you can show future clients.

Structure your pricing around recurring monthly retainers at about $250 per client and cap your roster at five or six, so steady retainers replace the need to resell every month.

Offer every prospect a three-option menu with clearly distinct, deliverable tiers, and hand it over as a printed one-pager so the buying decision becomes choosing the middle option.

For each promising local prospect, create a concrete sample of the work in about an hour and a half, deliver it in person with no pitch, and let the prospect ask what else you do.

With a one-in-four conversion rate and a six-month stay at $250 a month, Fine figures each 90-minute gift is worth roughly $375, which argues for running only a few well-chosen prospects at a time.

Start with a deliberately small product (prompts, templates, guide, or checklist) aimed at one named buyer's specific problem; the point is to get a finished, listed item, because that first listing accelerates everything you sell afterward.

Build and package the product in a single afternoon: draw from the weekly tasks you dread, test every file yourself, and assemble a PDF with a how-to page and one bonus element that doubles its perceived value.

Set the price between $7 and $19, raise it after ten sales, and choose a platform that gets you live quickly (Gumroad now, Etsy later) rather than waiting for a perfect product.

Turn every buyer into a promoter: ask each one for a one-sentence review, then embed the product link in your email signature, social bios, and free materials so it keeps being discovered after launch.

Validate a niche by checking autocomplete demand and review counts: under 50 reviews is winnable, above 500 means move on.

Outline 10 to 14 chapters, each ending with a concrete task, and write with voice rules, a banned-word list, and personal experience.

Let autocomplete dictate both your seven keywords and your subtitle, which should repeat the exact searched phrase, state the benefit, and identify the reader.

Ensure the cover reads clearly at thumbnail size, since that is the first and often only impression shoppers get.

During launch, set an honest AI disclosure, pick three specific categories, price low for two weeks to build ranking, and raise the price only after five reviews arrive.

Sell a fixed-price setup package built around a one-page deliverables list, not "AI consulting"; take your first job free so you learn where the hours actually go and can point at a real example when quoting the next client.

Pitch by naming the business owner's private suspicion that they pay for AI and use almost none of it, then demonstrate on your phone for ninety seconds; handle the staff-replacement and data-safety fears up front, and keep every word free of technical vocabulary.

Structure delivery as a discovery meeting, a solo build from their real writing and weekly tasks, and a hands-on training session with a printed one-page guide; at the check-in two weeks later, ask what still eats their time and turn that answer into a monthly offer.

Make the income recurring rather than one-off: the natural follow-ons are content production, report generation, keeping their Skills updated, adding staff to the system, and small tools built from the earlier material.

Avoid building while the client watches, handing over the system without a printed guide, or skipping the follow-up; if the client is not using at least two pieces after two weeks, the training was too fast or the pieces missed their actual work.

Set your first-year prices from local market value and the time a job takes you now, not from your beginner status or your first-week speed.

Quote only round numbers you can say aloud without flinching, and stay silent after naming the price until the client responds.

Sell packages instead of hourly work, each with included items, exclusions, timeline, revision cap, and a single price, so speed no longer reduces what you earn.

Put a cancel-anytime clause in monthly retainers and state exactly what you need from the client; work that visibly delivers will rarely be cancelled.

Raise prices when you're fully booked, getting instant yeses, or have results to point to, and schedule a price review every six months so planned increases actually happen.

Draw up twenty names from your current contacts before looking anywhere else, then mark the ones who both benefit and are reachable without an introduction.

Send your first ten messages with one of the three relationship-based scripts: free practice for close contacts, a specific observation plus free example for businesses you use, or a referral ask for friends.

Keep every first message short and concrete: mention one thing about them, give something first, and end with an easy yes-or-no rather than a call request.

Follow up on a fixed pattern: the original message, a one-line check-in five days later, and a useful no-ask touch two weeks after that if the lead is warm, then stop.

Write your headline and opening lines around the customer's concrete result, name the exact trade you serve, and back it with real samples or labelled practice work rather than certificates.

Open every proposal with one specific detail from the job post, then two lines of what you'll do, one relevant sample link, and a question that invites a reply; keep it to five lines.

Compete as a newcomer by being among the first ten proposals, naming the business directly, and offering a small free sample to a worthwhile job.

Convert one-off jobs into repeat customers by delivering early with one unrequested extra, then suggesting the next paid step at delivery and asking for the review immediately.

Spend your first two months taking small jobs to build reviews, then raise prices and apply selectively while keeping two or three repeat customers as your base.

Lock in your first work session for tomorrow, and protect one hour per day so the sequence stays unbroken.

Set up in the prescribed order: accounts and apps, workspace, settings, the asking formula, your path and service, then a real piece of work for a known business.

Treat day seven as a deliberate pause or catch-up slot so you enter the next phase with a clear path, a defined service, and at least one deliverable.

If you miss days, never restart the week; jump back in at the correct day and keep the overall order intact.

Build everything in week one with week three in mind, because that is the week where paying customers or serious conversations actually appear.

Choose one real business you genuinely want to work with, then treat every exercise that week as if it were a paid assignment for them.

Complete each day's small deliverable in order rather than in parallel, because each item becomes one page in a portfolio you will physically show.

Leave the work slightly unpolished and specific to that business; a generic sample proves craft, while a concrete artifact proves you can solve their actual problem.

Use the 20-prospect list to define your week 3 outreach: mark the double-yes names and draft ten messages now, so the ask on day 15 is already written.

By day 14, you should have a portfolio folder, a one-page offer menu, and a contact list; when you meet the prospect, those three items are the proof.

Send the two-sentence gift with no ask, because the generosity itself is what earns attention.

Personalize each of your ten messages with one concrete detail and invite introductions from exactly two contacts.

Reply on the day you respond, share a free sample with anyone who writes back warmly, and have the two-minute demo ready before you pitch.

Send three proposals, chase non-responders with a short warm note, and close the week by handing your top prospect free work.

Log every outreach and judge the week by ten logged messages plus two follow-ups; a Friday check-in with another person holds you better than motivation.

Show the offer, not the explanation: run the two-minute demonstration, present three packages, state the price, and treat a hesitant pause as a reason to stay silent, not to cut the price.

After the price, a yes means confirm scope and terms in writing, a maybe means book one dated next step, and a no means ask for the name of someone else who could use it.

Hand over the job a day early, with a clean file, a short cover note, and one useful detail the client did not ask for; a late first delivery becomes the first thing they remember.

When you collect the money, stay in the room: ask what to change, request the two-line testimonial, suggest the next assignment, and log the customer before the month closes.

At day 30, let the numbers tell you which stage of your funnel is broken, and keep any signed agreement or payment, even a screenshot of the first one, as proof you can do this.

Send three genuine outreach messages and get one honest verdict on sellability before you let more study or polishing delay you.

Quote a low price you can defend, send at least ten personalized pitches, and turn a no into either a condition for yes or a referral.

Give yourself twenty minutes of steady daily action instead of waiting for intensity; call your portfolio done and treat a changed tool as something the AI can walk you through.

Keep fact-checking and judgment on your side when you use AI, then put your name on the work; counter the fear of mistakes by sending more updates and finishing ahead of schedule.

If you switch from services to publishing, expect progress to reset; the difficulty stays, so the real variable is contact, not readiness.

Let each customer fund the next: deliver so well that they refer you, then spend the freed effort on routines and ten new messages a month.

Raise prices with every new customer, and bring older clients up to the new rate at the six-month mark instead of working tired at falling rates.

By month three, customer three should cost you less time and effort than customer one; if it does not, you are growing without leverage.

Choose your enough number before you scale, and once you pass five customers, stop adding: raise prices, deepen services, or hire instead.

Follow the 90-day map in order, then hold a twenty-minute monthly count of messages, customers, money, and hours, cut the lowest-paying task, and make any second income stream wait until the first is dependable and aimed at the same audience.

Treat how you introduce yourself, the templates you reuse, and your fact-checking as permanent; fix cosmetic changes like new menu names or commands in a few minutes instead of chasing newer tools, because customers see results, not your choices.

Once a month, spend ten minutes asking what changed, which two changes would save the most time, and what you could now do better; take up an update only when it saves time every week or unlocks a sale you could not make before, and use each major release as a reason to contact past customers.

Keep a working vault of two hundred fill-in-the-blank prompts, and let honesty run through every category: lines that turn the wrong buyer away, urgency that is true, and admissions when a competitor is the better fit.

Sort each complaint before replying: you were at fault, the customer is wrong, or the request is unreasonable; write review replies for the readers who come later, and settle public arguments with one calm public reply and one private message aimed at closing, not winning.

Let agents and Skills draft, file, watch, and chase, but put a human checkpoint before anything goes public or money moves; build each Skill from interviews or good examples, test it on fresh cases, and repair failures at the instruction level rather than in the output.