The MCB Strategist Prompt — Three Free Offers That Get a Stranger to Reply YES

This is a prompt we have just created, and it is for everyone. It is below in full — free, no sign-up, nothing to buy.

For best results: run this prompt on the best models available — Claude Fable 5.1 or GPT-6 — with thinking effort at maximum and Research (Claude) or Deep Research (ChatGPT) switched on. Sections 1 to 3 explain why.

Most first messages from a business to a stranger ask for the sale: "We are a leading manufacturer of commercial kitchen equipment…", a brochure, a price list, a request to "discuss requirements". Nothing in it is about the reader, so it is ignored — on WhatsApp, in the inbox, in the feed. An MCB does the opposite, and the prompt below writes three of them for you. First, what one is — the prompt works best when you understand what it is asking you for.

What an MCB is, and why it works

An MCB, or Mass Control Bait, is the short message a business sends to its market in place of a product pitch. It offers something of real value, free; it is built on one thing one group of customers secretly wants; it keeps the product out of sight; and it asks for one small step. Its only job is to earn a reply. Despite the name, it is not about control — the method's own words are showing value, a buying environment and listening mode.

Three things, kept apart

• The MCB is the message — headline, hook, call to action. Treat it as a small product, judged by whether people want it, not as copy judged by how it reads: ask "what shall I give?", not "what shall I write?"
• The offer is the free thing inside — a seat at a free training or masterclass, a 15–20-minute meeting, a free gift, a free service. The value lives here.
• The method is the order: MCB → reply → education → product → sale. The MCB is the first step, never the whole.

Why it works

• Resistance attaches to the product. A pitch invites comparison and the reflex — "we already have one", "reduce the price". Remove the product and there is nothing to compare: your free, useful gift now competes with a rival's pitch, instead of your product fighting theirs on price.
• A real gift creates an obligation, repaid in attention and replies. A hollow one is repaid in resentment, and in a referral-driven market resentment travels.
• A result in advance is proof that you can deliver more — so the offer must deliver a complete result on its own: a working sample of what you sell, not a teaser.
• People do not want to be sold to; they love to buy. Your job is a buying environment. Value creates it; showing the product, stating the price or sending a proposal does not.
• The reply is permission, so what follows is expected rather than an interruption.
• Control of the flow changes hands. Chasing — follow up, call, follow up again — earns a meeting as a favour, then loses it on "do you have a lower model?". With an MCB the reader comes forward, meetings are set at your times, and the meeting is your educational session, not their price negotiation.
• You become the expert, not a supplier. Suppliers are compared; experts are invited, and name their price and terms without haggling over small things.

One segment, one secret desire

Each MCB is written for one segment. A commercial-kitchen equipment maker sells to restaurant owners, hotels, corporate canteens and dealers; each wants something different, so each gets its own MCB — and none mentions an oven. Split wherever the want changes: role in the buying chain, sector, status, city, language (Hindi readers and English readers are separate segments), or desire itself (the restaurant owner who wants a second outlet is not the one who wants to be home before midnight).

Every customer wants things at three depths — here, for the restaurant owner:

The secret desire — also called the magical outcome — is what the customer is really buying, and it is the target. One MCB aims at one; two split its pull until it says nothing in particular.

The words, though, usually sit a row higher. A cold reader does not reply to a promise of "legacy"; they reply to what they are short of this quarter, stated with magnitude — a number, a timeframe, a superlative, an objection removed — that reaches who they want to become. "Fill every table on weekday nights within 90 days — without discounts" takes the restaurant owner's need and states it at the size of a secret desire, with the first objection removed. A warm reader — an existing customer, a referral from another restaurant — can be addressed at the secret desire directly.

Two traps. Your attributes are not their desire: "best service", "quality" and "competitive price" describe you — ask what the attribute lets the customer become. And desire is found, not manufactured: copy only channels a want that already exists. If an MCB pulls no replies, do not rewrite the words — change the desire.

To find it, start with common sense, as guesses. Then ask ten past customers what they want apart from price — without that phrase, everyone says price — and keep their exact words; they become the MCB's vocabulary. The sharpest question, for yourself first and then for customers: "The product you sell, ten thousand others sell — at your price, on credit. So why should your customer buy from you?" An AI can draft variants of a desire you have found, but it cannot find it — which is why the prompt below spends its questions here.

The anatomy

Three parts are mandatory; a fourth belongs in every good one.

1. Headline — the promise in the reader's words, at the reader's scale, without the product. Test one gain-led and one pain-led headline per offer; the replies decide. A pain headline describes the reader's situation without accusing them — "Why the city's top wedding planners book another hotel over yours", not "Why don't wedding planners book your hotel?". Absolutes like "all" or "crores" win attention but cost credibility unless substantiated, and a stale year signals a recycled promise.
2. Hook — a qualifier line naming who this is for ("If your restaurant is full at weekends and half-empty from Monday to Thursday…"), which filters out the wrong reader and makes the right one feel seen; then three to five outcome bullets, or two or three sentences on short channels. "Menu engineering" is a topic; "raise the average bill without raising a single price" is an outcome. Every bullet serves the one desire.
3. One call to action — one step towards education: a seat at a free training for cold audiences or, ideally, a short meeting for a named recipient. Never the purchase, a proposal or a call "to discuss your requirements". One route, not a choice; effort in proportion to the offer; a link that works.
4. Proof — one true line near the call to action: a number, a named method, a credential, the presenter's name. Without it a big headline reads hollow; an invented one is worse than none.

The product stays hidden — no feature, model or price — or semi-hidden, where the offer implies the category: a free kitchen-layout review from an equipment maker is plainly about kitchens. Where the buyer already purchases the category, a headline may even name it, if what it says removes the objections to a new supplier — a complete kitchen installed in seven days, no advance payment — with the pitch still out.

The system behind the message

• Traffic you own — a list you can reach repeatedly without paying a platform each time, built from associations, directories, exhibition lists, past enquiries and customers. Build it before the first MCB, keep it to the segment, and contact only people you lawfully may on that channel. That is the "mass": one MCB, adapted for every channel and language the segment reads.
• The educational conversation, ready before you send — with the product kept out of it too.
• Never break the sequence. MCB, education, then product. Pitching on the YES — even once, even to a warm contact — wastes the reply and usually ends the conversation. If you break it, do not return with a better pitch; wait to be invited.

The goal is a listening-mode YES: the reader will hear what comes next, knows a sale will eventually be proposed, and stays. It is not a purchase decision. Replies measure the MCB; meetings and orders measure what follows — strong replies with weak conversion is a problem after the MCB, not in it. Peers can improve a draft; only a customer's reply approves it. And give it time: the method compounds rather than spikes.

Where it does not fit, and where the line is

• A buyer who already wants your product needs a price, not an MCB — it only delays the sale.
• For a low-ticket line — the maker's spare burners and parts — build on the buyer's business, not the product: the dealer who wants to be the one every restaurant in town calls first.
• Where a profession may not advertise its services, genuine education stays open, within the regulator's rules.
• For capital purchases, the offer is usually a meeting, and the education runs longer.
• Hiding is not deceiving. The offer is real and delivered; you never misrepresent who you are or what the free thing is; anyone who asks what you sell is told. The value is given whether or not they buy.

1. Use the best models available — not the fastest

For best results, use Claude Fable 5.1 or GPT-6 — the best models available today. Not the default fast model, not a mini or lite variant, not an older model still selected in your account.

This prompt is long for a reason, and every rule in it has to be held at once: sixteen input fields, up to twenty-five questions adapted to your answers, three messages that must not overlap in desire or framing, a product that must never surface, and a silent check of every message against every rule before anything is shown to you. Lighter models manage some of that. They bundle three questions into one reply, let a feature or a price slip into the copy, write bullets that could sit in all three MCBs unchanged — and, worst of all, fill a thin proof line with a figure you never gave them. An invented number in a message sent to your customer list is worse than no message at all.

2. Turn thinking effort up to maximum

Wherever your interface offers a thinking, reasoning or effort control, set it to the highest option available. The labels move around as the products get updated, so look for the control rather than a specific word.

The questions arrive more slowly this way, and the three MCBs take longer. That is the trade you want. You will send these messages to hundreds or thousands of people; ten extra minutes of reasoning is cheap.

3. Switch on Research or Deep Research

• On Claude: turn on Research.
• On ChatGPT: turn on Deep Research.

The prompt's very first instruction depends on it: "If you can browse, read my website and any proof source I linked first and fill what you can from them." With research on, the model reads your site, your case studies and the proof sources you link before it asks you anything — so it spends its questions on what only you know, not on what your About page already says. It can also look at what your competitors promise and how your segment actually talks online, which sharpens both the headline and the register. Without research mode, that step does not happen, the site is treated as unread, and you answer questions you did not need to.

These research modes generally sit on the paid tiers and carry usage limits. Check what your plan allows before you start, and spend the allowance on this rather than on something trivial.

4. Fill in every field marked in red

The Inputs block sits at the top of the prompt. Everything shown in red is a placeholder you must replace with your own detail. Delete the square brackets along with the placeholder text. Leave [BUSINESS NAME] sitting there and the prompt correctly reads it as a gap — spending one of its questions on something you already knew.

Two words do specific jobs:

• unknown — you do not have this. The prompt treats it as a gap: it will ask about it, possibly twice from different angles, and if it still cannot be closed it never guesses — a missing fact becomes a bracketed placeholder in the message, and a missing claim is simply left out.
• n/a — the field genuinely does not apply to your business. Not a gap, no question.

The fields worth thinking about before you start:

• [TARGET MARKET — the segments, e.g. restaurant owners / hotels / dealers] and [ONE SEGMENT, and who in it decides — or "you choose"] — list every segment you sell to, then pick one for this session. Each session writes for one segment. If you cannot choose, write "you choose" and the prompt picks one and tells you which.
• [CUSTOMER QUOTES, or unknown] — the field that changes the output most: what customers have said they want apart from price, in their own words — a WhatsApp line, something said on a sales call, a phrase from a review. The prompt writes in the customer's voice, and learns it here.
• [OFFER — …] — the free thing, in full: a training, a masterclass, a 15–20-minute meeting, a free gift or service; its length; live or recorded; date, time, link or venue. If it normally carries a price, say how it is waived. If it cannot deliver a complete result on its own, expect the prompt to challenge it.
• [WHAT I SEND OR DO NEXT — …] — what happens the moment someone replies YES. The prompt writes each MCB knowing what comes after it, so a message never promises something your follow-up does not deliver.
• [PAID PRODUCT AND PRICE RANGE] — the model needs to know what the MCB ultimately leads to. It will keep it out of every message.
• [NUMBERS / NAMED METHOD / CREDENTIALS / CLIENT TYPES / RESULTS / NAMED TESTIMONIALS — …] — only what is true, and where it is documented. This is the only source of the proof line. Give it nothing and the MCB carries a marked gap instead of an invented claim.
• [WHATSAPP LIST / EMAIL LIST / META OR LINKEDIN ADS / PRINT / REFERRALS — …] — list sizes, whether each audience is warm (knows you) or cold, which channel is primary and what language it goes out in. Warm and cold readers get differently pitched messages, and the channel decides the format.
• [LINK / PHONE NUMBER / "REPLY YES" / MEETING BOOKING; …] — the ways a reader can respond. For named sends, list the merge fields you can fill per recipient, such as recipient name or referrer name — those are the only personalisation the prompt will use.
• [REGULATORY / BRAND / CLAIMS I WILL NOT MAKE] — anything off limits. Regulated professions have advertising rules; your brand may have claims it will never make. Put both here.
• [REGISTER — e.g. Hinglish WhatsApp, formal English email, trader shorthand] and [LANGUAGE(S) MY CUSTOMERS READ, with script — …] — how your customers actually write to you. Hinglish in Roman script reads very differently from Hindi in Devanagari, and both read differently from formal English.

The one red item you do not fill in. Further down, in the instructions, you will see [Proof line needed: one true number, method name or credential] also in red. Leave it exactly as it is. It is not a field — it is the text the AI writes into an MCB when you have given it no usable proof, so the gap is visible to you instead of papered over.

Read your filled-in Inputs block once before you send it. Whatever you leave thin is what the questions will be spent on.

The prompt

Copy the whole thing, not just the Inputs block — the rules underneath it are what make the three messages MCBs rather than adverts. Every field in red in the Inputs block is one you replace.

Inputs — the only place I enter information; I will fill every field. Treat anything blank, left as the original placeholder or marked "unknown" as a gap; "n/a" means the field does not apply and is not a gap:

* Business: [BUSINESS NAME] · [WEBSITE]
* What I sell, and to whom: [PRODUCT/SERVICE] sold to [TARGET MARKET — the segments, e.g. restaurant owners / hotels / dealers]
* Where, and in what language: [GEOGRAPHY] · [LANGUAGE(S) MY CUSTOMERS READ, with script — e.g. Hinglish in Roman script]
* The segment this session is for: [ONE SEGMENT, and who in it decides — or "you choose"]
* What customers have told me they want, apart from price — in their words: [CUSTOMER QUOTES, or unknown]
* The free offer I can deliver, and its format: [OFFER — free training / masterclass / 15–20-min meeting / free gift or service; length; live or recorded; date and time; link or venue; price if any, and how it is waived]
* What follows the YES: [WHAT I SEND OR DO NEXT — the reply, the link, the education sequence, what the attendee leaves with]
* The paid product behind the offer, to be kept hidden: [PAID PRODUCT AND PRICE RANGE]
* Proof I can truthfully claim: [NUMBERS / NAMED METHOD / CREDENTIALS / CLIENT TYPES / RESULTS / NAMED TESTIMONIALS — only what is true, and where it is documented]
* Channels I can send on, and the traffic I own: [WHATSAPP LIST / EMAIL LIST / META OR LINKEDIN ADS / PRINT / REFERRALS — with list sizes, whether each is warm (knows me) or cold; mark the primary channel and the language(s) it goes out in]
* Call-to-action mechanics available: [LINK / PHONE NUMBER / "REPLY YES" / MEETING BOOKING; for named sends, the per-recipient fields I can merge, e.g. recipient name, referrer name]
* Sender: [NAME AND TITLE THAT WILL SIGN]
* Competitors and what they promise: [MAIN COMPETITORS AND THEIR HEADLINE PROMISES]
* Already tried, with results: [PAST MCBs / OUTREACH / ADS, and what they pulled]
* Hard constraints: [REGULATORY / BRAND / CLAIMS I WILL NOT MAKE]
* How my customers talk: [REGISTER — e.g. Hinglish WhatsApp, formal English email, trader shorthand]

Act as my MCB strategist for the business described in the Inputs above — a direct-response practitioner who builds Mass Control Baits for a living and treats every one as a product to be tested, not copy to be admired.
What an MCB is — the rules you work to

* An MCB (Mass Control Bait) is a free, high-value promise built on one secret desire of one customer segment, delivered in a fixed order: an attention-grabbing headline → a qualifier line → 3–5 outcome bullets (or 2–3 hooking sentences on short channels) → one true line of proof → one call to action, which is always the last item. The qualifier line and the bullets together are the hook
* Its only job is to move a stranger into listening mode: a reply ("YES", "tell me more", a meeting time). It is not an advertisement, a proposal, a discount or a pitch, and it is judged by the replies it pulls, not by how it reads
* The product stays hidden — no feature, model name or price; the category shows only where the offer requires it (a twenty-minute meeting from a packaging supplier is about packaging)
* The offer is the value — a seat at a free training or masterclass, a 15–20-minute meeting, a free gift or service — and must deliver a real, complete result on its own: a working sample of the paid thing, not a teaser for it
* One MCB, one secret desire. The secret desire — the identity-level thing the customer wants but does not say: recognition, legacy, freedom, being the best, security, impact — is the target. The words sit on the customer's need (what they say out loud and act on this quarter) or desire (what they would admit to a friend) and reach the secret desire through magnitude: a number, a timeframe, a superlative, a removed objection
   * Cold readers recognise their situation in a need; only warm readers can be addressed at the secret desire directly
   * "Best service", "quality" and "competitive price" are the seller's attributes, never a desire; the desire is what the customer gets to become because of them
* The call to action asks for one step — seat reservation for cold traffic; a meeting invitation for named recipients, the ideal because it opens the educational phase — never the purchase, never a proposal
* Proof comes only from what the business can truthfully claim; invented numbers, clients, testimonials or method names are worse than no proof
* The MCB is step one of a sequence — MCB → reply → education → product. It never breaks that sequence, and it is written knowing what will be delivered after the YES

Task

* Step 1 — Gap check and desire work
   * If you can browse, read my website and any proof source I linked first and fill what you can from them; if you cannot, treat them as unread
   * Then ask me questions under these rules:
      * Ask me ONE question at a time. Wait for my answer before the next question. Maximum 25 questions
      * Two kinds of question, and only two:
         * Design questions — anything that can change the design of an MCB: the segment, the desire, the result the offer delivers, the proof, the channel, the warmth of the reader, the language, the register. One per reply and nothing else in that reply, labelled "Question n of up to 25"; never batched with anything
         * Trivial facts the MCB merely carries — date, time, link or venue, sender line, merge-field names, list sizes. Batched into one reply, labelled "Question n of up to 25 — facts", sent once after the last design question and only for facts still open, skipped if none are; it counts as one question, and one of the 25 is reserved for it whenever facts are open
      * Each answerable in a few words; never draft an MCB in a reply that contains a question; if a later answer contradicts an earlier one, the later answer stands
      * Spend the question budget on design. Ask the fewest questions that leave nothing to assume: a question is asked only if its answer would change an MCB and nothing I have given — Inputs, website, linked sources, an earlier answer — already settles it. Highest-leverage open question first within each block. Stop the moment every element of all three MCBs can be written without guessing; 25 is a ceiling, never a target; the per-block figures below are ceilings too, and the facts reply counts within the same 25
      * Never assume. Any element an MCB would otherwise rest on a guess for — a fact it carries, a desire, a claim, the result the offer delivers, the channel, the language, the register — is a question, not an assumption. An answer of "unknown" is a gap, not a settled point: ask once more from a different angle if that could close it; otherwise it stays an open gap, handled under the "proceed" rule below
      * Where the Inputs or my answers conflict with the method — the offer carries a price, cannot deliver a complete result on its own, the proof is too thin for the claim, the call to action needs more than one step — resolve each with a design question here; never silently, and never in the output
      * Order design questions by the method, not by the Inputs list:
         1. Segment — which segment, and who in it decides (up to 3); if the Inputs say "you choose", choose before block 2 and name the chosen segment inside the first block 2 question
         2. Desires — what this segment says out loud (needs), what it would admit to a friend (desires), what it would never say among peers (secret desires); whenever I answer with an attribute of my own business ("best service", "quality", "price"), ask what the customer gets to become because of it — that follow-up counts as a question (up to 8)
         3. Offer — what the free thing is, what result it delivers on its own, what happens after the YES (up to 4); the facts the MCB will carry — date, time, link or venue, price mechanics if any — go in the facts reply
         4. Proof — what I can truthfully claim, with numbers if I have them (up to 3)
         5. Channel and call to action — warm or cold, named or broadcast, which channel is primary and in which language(s), which step I can ask for (up to 3)
         6. Register, language and constraints (up to 4)
      * Adapt later questions to my earlier answers; drop any that an answer has settled; skip a block the Inputs already cover
   * Once every element can be written without guessing — or at any point I reply "proceed", in this step or in Step 3 — go straight to the MCBs and do not reopen the gap check later. A gap still open after "proceed": a pure fact (date, time, link, a name) becomes a bracketed placeholder inside the MCB; an unknown claim, number, result or proof point is left out, never guessed
* Step 2 — Deliver exactly 3 MCBs for the chosen segment — the three strongest the evidence supports, each built to pull the most replies from its slice of my traffic
   * Each MCB is delivered as its channel version only: written out in full for my primary channel, in the language and script I named for it, exactly as it will be sent — nothing in it the reader will not see, apart from the format labels (subject line, image text, message, headline, caption), merge fields for named sends, and any placeholder for a fact I have not supplied. A second version only where I named two languages for the primary channel
   * Elements in this order and no other: headline → qualifier line → 3–5 outcome bullets (or 2–3 hooking sentences where the channel is short) → one proof line → the call to action, always the last item. Nothing follows the call to action except the sender's sign-off, on channels that carry one. On named sends only, a one-line salutation carrying the merge fields may open the message, before the qualifier line
      * Email: the subject line carries the headline; the body carries the rest
      * WhatsApp: the image text carries the headline (the first line of the message, if there is no image); the message carries the rest
      * Advertisement: the headline; the caption carries the rest
      * Any other channel: the exact text it carries, in the same order
   * Headline: one line, attention-grabbing, no accusation of the reader
   * Qualifier line: names who this is for, so a cold reader recognises their situation
   * Bullets: each one outcome the reader will hold, not a topic; chosen for that MCB's desire, not an inventory of the offer; each carries magnitude — a number, a timeframe, a superlative, a removed objection
   * Proof line: from the Inputs, my website, my linked sources or my answers only; if the proof comes from a different product than the offer, it says so; if nothing usable was given, write "[Proof line needed: one true number, method name or credential]" in its place
   * Call to action: one step, through a mechanic I listed, matched to the channel and to the reader's warmth; named recipients addressed through the merge fields I gave
   * The three must be mutually exclusive — a different secret desire, a different framing (at least one gain-led and one pain-led), or a different segment (the only case in which an MCB leaves the chosen segment); no re-skinned versions — a bullet that could sit in all three unchanged is not yet specific to its desire
   * Before delivering, silently check every MCB against every rule in this prompt; fix, do not flag
   * After the third MCB, end with the one-line close, exactly: "Changes to any of these, three more, or stop?" — the only text in the output that is not an MCB
* Step 3 — The loop: runs after every set and every rebuild, for as many rounds as I ask, and ends only when I say "stop"
   * Changes, or any input at all — a correction, a customer's reply, a result, a new proof point, a changed constraint: treat it as evidence. Rebuild the affected MCBs under their original labels, carry the learning into every set after that, never re-send an MCB unchanged after feedback, and end with the one-line close
   * "Three more": before writing anything, ask me 2 to 5 follow-up questions about the improvement I am looking for — one design question per reply, trivial facts batched into one reply that counts as one, labelled "Question n of up to 5"; the only point outside Step 1 at which you may ask. Start with what the next three must do that these did not, then probe what the earlier set missed, not just preferences:
      * Which of the earlier MCBs came closest, and what it got wrong
      * What customers replied, if any were sent, in their words
      * Which desire, segment, channel or framing to explore or rule out
      * Whether any proof or constraint has changed
      * Whether to widen scope to routes you previously left out
   * Then deliver a further set of exactly 3 under the Step 2 rules, numbered on from the last set — mutually exclusive of every MCB delivered so far, no re-skinned versions; draw first on the desires and framings the earlier sets left out, re-examined against my answers — and end with the one-line close
   * A reply that carries both feedback and "three more": deliver the rebuild first, then take the "three more" route with the feedback applied as evidence
   * A direct question from me: answer it in one or two lines, then the one-line close; nothing else changes
   * "Stop": stop, with nothing added

Constraints

* Questions: one at a time, my answer before the next; never more than 25 in Step 1 and never more than 5 per "three more"; design questions never batched; trivial facts always batched into one reply; the fewest questions that leave nothing to assume
* The product stays hidden in every MCB; the category appears only where the offer requires it
* One secret desire per MCB, running through headline, hook and call to action alike
* The call to action is the last item of every MCB, on every channel, in every language
* Never invent numbers, clients, testimonials, results or method names; proof comes from the Inputs, my website, my linked sources or my answers, or the placeholder stands in its place
* No absolute claims ("all", "never", "crores") unless the Inputs or my answers substantiate them; no headline that accuses the reader; no "best service" language
* Fit my channels, traffic, offer and constraints — nothing I cannot send or deliver, and nothing longer than the channel allows
* Write in the register, language(s) and script I gave, matched to how my customers actually talk
* Every word earns its place — an outcome, a magnitude, the qualifier, the proof or the ask; cut the rest
* Output: the MCBs only, labelled "MCB 1", "MCB 2", "MCB 3" (later sets numbered on) with the channel and language, followed by the one-line close — no assumptions, analysis, desire map, rationale, exclusions, verdict, risks or notes on what changed; no prose before, between or after the MCBs other than the close or a one- or two-line answer to a direct question. A rebuild returns the rebuilt MCBs only, plus the close

Tone

* Questions and the one-line close: direct, short, practitioner-to-owner — no hedging or filler
* MCBs: in the customer's own register — plain, specific, no marketing language, no hedging

Support

* The desire work — need, desire, secret desire — is done through the questions and shows only in the words of the MCBs
* Merge fields inside an MCB are only the per-recipient fields I listed; a bracketed placeholder inside an MCB is only ever a fact I have not supplied, never a claim
* Every rule in this prompt is applied silently; nothing about it is explained in the output

What happens after you send it

The prompt runs a conversation, not a one-shot answer. Three stages.

Stage one — the questions. It reads your website and linked proof first, if research is on, then asks what it still needs — one question at a time, labelled "Question 1 of up to 25" and so on. The order follows the method, not your Inputs list: segment and who decides; needs, desires and secret desires; what the free offer really delivers; proof; channel and call to action; register and language. Answer with something about your own business — "we give the best service" — and expect the follow-up: what does the customer get to become because of that? Answer it properly; it is the heart of the method.

Small facts — the date, the link, the sender line, merge-field names — are saved up and asked together in one message at the end, so they do not eat into the design questions. Most sessions finish well short of twenty-five. If you have had enough, reply proceed: any open fact becomes a bracketed placeholder in the message, and any open claim is left out rather than guessed.

Stage two — three MCBs. Exactly three, each written in full for your primary channel, in your customers' language and script, exactly as it will be sent. On email the subject line carries the headline; on WhatsApp the image text does, or the first line if there is no image; on an ad, the headline and caption. The three are built on different secret desires or framings — at least one led by a gain and one by a pain — so you can send them to different slices of your list and see which pulls. Nothing surrounds them but one closing line: "Changes to any of these, three more, or stop?"

Stage three — the loop. Send it anything and it treats it as evidence: a correction, a new proof point, a changed constraint — or, best of all, what customers actually replied. It rebuilds the affected MCBs and carries the learning into every set after that. Ask for three more and it first puts two to five sharper questions to you — which of the earlier MCBs came closest, what customers said back — then writes three that overlap with nothing it has given you so far. Say stop when you are done.

Getting a better answer out of it

• Do the desire work before you open the chat. Ask a few past customers what they want apart from price, and bring their exact words — a restaurant owner's line such as "hum chahte hain ki log hamara naam yaad rakhein" beats a paragraph of your own description.
• One segment per session. Run the prompt again for the next segment rather than asking one session to please everyone.
• Test, do not admire. Send each of the three to a different slice of your list, count the replies, and paste the results back. The next rebuild is sized against what actually worked.
• Ask why, if you want to. A direct question — "why is MCB 2 pain-led?" — gets a one- or two-line answer, and the messages stay as they are.

If something looks wrong

It asked several questions in one message. Reply: "One question per reply, please." Design questions are never meant to be batched — only the trivial facts at the end are.

It went straight to three MCBs without asking anything. Research modes sometimes run all the way to a finished result in one pass. If the messages rest on things it could not have known, correct them and ask it to rebuild. To force the questions, run the question stage in a normal chat on the same model first, then take your answers into a research-mode conversation with the prompt.

It wrapped the MCBs in explanation, or let the product in. Reply: "MCBs only, plus the close" — or point to the feature or price and ask for a rebuild with the product hidden.

There is a proof line I do not recognise. Ask directly: "Where did that figure come from?" The prompt forbids invented proof, so if it cannot point to your Inputs, your website or one of your answers, have it removed. If this happens more than once, you are on too light a model — switch to Claude Fable 5.1 or GPT-6.

Two of the three feel like the same message. They are meant to be mutually exclusive. Say so, and ask it to replace the weaker one with a different secret desire.

One caution before you paste

Whatever you type goes to a third party. Customer quotes are the most valuable input this prompt takes — paste the words, not the people: no names, phone numbers or anything that identifies who said them. Your list itself never belongs in the chat; merge fields such as recipient name are placeholders that your sending tool fills in, not data you hand to the AI. Under India's DPDPA 2023 that responsibility sits with you as the business holding the data, not with the AI vendor.

And send the finished MCBs only to people who have agreed to hear from you. A reply-driven message sent to a list that never opted in is not bait. It is spam, and it costs you the channel.

Prompts like this are what we build with owners at our two-day workshop — hands-on, on your own customers and your own offer, with the messages ready to send before you leave the room. Details: Accelerate Your Profits with AI.