AIMirrorv3

Adaptive Mirror — master product document.

Everything from the original Mirror Method thread and the Adaptive Mirror 2 expansion, merged, resolved, and marked up with the constraints that came out of the market and legal review.
Category
Conversation-to-Outcome Engine
Mechanism
The Mirror Method™
Status
Pre-build · decisions open
Supersedes
v1 thread · v2 expansion
00 How to read this

Three documents merged into one, with the disagreements resolved on the page.

This supersedes the original Mirror Method thread and the Adaptive Mirror 2 expansion. Nothing has been dropped — where the two disagreed, or where market and legal review changed the answer, the change is marked rather than quietly applied.

Source 1

Mirror Method v1

May 2026 thread. A personalised sales page assembled from a prospect conversation. Four mirror layers, JVZoo funnel, launch math.

Source 2

Adaptive Mirror 2

The expansion. Multi-channel input, universal profile, fifteen assets, nine modes, seventeen verticals, realtime-native architecture.

This doc

v3 — resolved

Both, merged, plus the competitive scan, the legal register, corrected launch math and a build plan with kill criteria.

Reading convention
Blocks like this one flag something that changed in v3 — a removal, a constraint, a correction or an addition that wasn't in either source document. Everything outside these blocks is your material, preserved and reorganised.

The seven decisions v3 makes

DecisionResolution
Emotional MirrorRemoved. Replaced by a Proof Mirror plus register-matching inside the Language Mirror. Emotional-state inference from free text is the highest-risk, lowest-value feature in the spec. §03
Budget MirrorRetained but hard-bounded: option hierarchy and presentation may change; the numbers may not. Price never varies by person. §03
Phone MirrorKept as a channel, removed from the launch funnel. Imported transcript delivers most of the value with none of the telecom exposure. §02, §15
Mirror MemoryKept, but derived-codes-only by default, opt-in for verbatim, and off entirely for health-adjacent verticals. §07
Positioning leadLeads on component re-composition, not headline personalisation. Headline generation is a funded, crowded category; re-composition is empty. §01
Vertical scopeSeventeen verticals retained as a map, but real estate, automotive and recruitment are reclassified as regulated markets needing their own legal workstream. §09
Launch mathMRR projection corrected — the v1 figure conflated MRR with ARR and was out by roughly an order of magnitude. §16
01 Positioning & category

One conversation. The exact proposal, page, quote, recommendation or next step that person needs.

Alternative positioning lines

Product category

Avoid describing this as an AI chatbot, an AI landing-page generator, an AI proposal writer, or generic personalisation software. All four are crowded, commoditised categories with an established price ceiling.

Conversation-to-Outcome Engine — an intelligence and asset-generation layer between a conversation and the next commercial outcome.

The branded mechanism stays: The Mirror Method™ — Listen → Understand → Mirror → Create → Route → Remember.

v3 · what to lead with, and why it changed
A competitive scan of roughly sixty products found the market has three architectures: asset factories (CSV in, batch pages out — Hyperise, Potion, Sendspark), runtime selectors (traffic signals pick between human-written variants — Unbounce, Webflow Optimize, Coframe), and runtime generators (an LLM rewrites copy per visitor — Fibr AI at ~$479/mo, Wix Adaptive Content).

Nothing takes a live conversation as input and produces page structure as output. But "AI writes your headline from a signal" is solved and funded. So the defensible claim is not the Language Mirror — it is the Offer Mirror and the Objection Mirror: changing which components exist on the page and in what order. Lead every deck, page and demo with that.

Two venture-backed teams shipped adjacent versions of this and abandoned them inside a year, both because they required customers to build and maintain separate microsites. Design constraint: the customer must build and maintain nothing new.

What it is not

Not a page builder

It sits on pages that already exist. No new URLs, no new funnel, no changes to an email sequence.

Not a chatbot

Three or four questions and it's finished. No persona, no name, no ongoing relationship. It asks, then it leaves.

Not autonomous copy

A human wrote every commercial sentence. The engine decides which of them this person sees. That distinction is the product.

02 The Adaptive Mirror model

Conversation layer → universal profile → Mirror Brain.

2.1 Conversation layer

Four interfaces, one backend intelligence.

☎ Phone

Phone Mirror

Inbound or outbound AI phone conversation over SIP.

🎙 Voice

WebRTC Mirror

"Talk to AI" — browser voice, no phone number required, no telecom.

💬 Text

Chat Mirror

Website chat or embedded text conversation.

📄 Import

Imported conversation

An existing human or AI sales-call transcript, processed after the fact.

The channel is not the product. The conversation is the data.
v3 · channel priority, reordered
Imported transcript should ship first, not last. It is the only channel that delivers value with zero adoption cost — no widget, no page change, no visitor behaviour to alter. That directly answers the failure mode that killed the two prior attempts at this mechanic. Chat second, WebRTC third.

Phone stays in the architecture and out of the launch. AI voice is treated as an "artificial voice" under FCC 24-17 with no live-agent carve-out; state mini-TCPA statutes stack on top (Pennsylvania SB 992 imputes vendor conduct to the hirer with a private right of action; California AB 2905 requires a live human to pre-announce the AI); and platform liability for a customer's violations is being actively litigated. BYOK does not shield a platform from that theory. Ship phone later, as a separate product, with its own compliance posture.

2.2 Universal Mirror Profile

Every conversation, on any channel, is converted into one structured profile. This single schema is what lets the same engine serve local business, agency, affiliate, SaaS, high-ticket, professional services and beyond.

SignalWhat AI Mirror understandsv3 handling
SituationWhat is happening now?store
ProblemWhat problem needs solving?store
Desired outcomeWhat result does the prospect actually want?store
MotivationWhy does that result matter?store
PriorityWhat matters most in the decision?store
TimelineWhen do they need the result?store
Budget sensitivityWhat financial constraints or ROI concerns exist?routing only
ObjectionsWhat is stopping them moving forward?store
PreferencesWhat do they prefer or want to avoid?store
AwarenessResearching, comparing, solution-aware, product-aware, ready?store
IntentLow, medium or high purchase/action intent.store
Decision criteriaWhat will make them choose one option over another?store
Exact languagePhrases and vocabulary worth reflecting.session only
SentimentSceptical, worried, excited, impatient, uncertain.routing only
Best next actionBuy, book, quote, compare, apply, learn, follow up.store
v3 · three fields get special handling
Exact language — session only. Verbatim prospect text is held in memory for the session and not persisted by default. Retaining it is the weakest link in the whole design: it is the field most likely to contain health, financial-distress or protected-class content, and the one enforcement precedent keeps landing on. Verbatim retention becomes an explicit, off-by-default, consent-gated setting.

Sentiment — routing only, never rendered. Use it to choose components; never quote or paraphrase it back. See the reflection-strength Silent mode in §03.

Budget sensitivity — routing only, never priced. It can decide which option leads. It can never decide what an option costs.

2.3 The Mirror Brain

The Brain takes the profile and answers ten questions:

  1. What does this person actually need?
  2. What information matters to them?
  3. What should be de-emphasised or removed?
  4. Which objections should be addressed?
  5. Which proof is most relevant?
  6. Which offer, package or product fits best?
  7. How much information should they see?
  8. Which asset should be created next?
  9. What CTA should that asset contain?
  10. What should be remembered for the next interaction?

Implemented as structured extraction plus an editable decision and routing layer — never an LLM freestyling the entire commercial journey.

☎ Phone🎙 WebRTC voice 💬 Website chat📄 Imported transcript Mirror Listener transcription + structured extraction Mirror Profile 15 signals · one schema Mirror Brain 10 decisions Router → asset engine MIRROR MEMORY — PROFILE PERSISTS, DISCOVERY NEVER RESTARTS THE CHANNEL IS NOT THE PRODUCT — ALL FOUR INPUTS PRODUCE THE IDENTICAL PROFILE
Fig. 1 — conversation layer through to router
03 The Mirror layers

Eleven independent personalisation passes over the same asset.

Each layer reads a different signal and changes a different thing. They run separately, so a weak answer on one doesn't degrade the others.

Layer What changes Defensibility
LanguageHeadline, bullets and explanations reflect the customer's natural vocabulary and register.Low — copyable
ProblemFocuses on the problem actually described, not the one the page assumed.Medium
GoalFrames the solution around the stated desired outcome.Medium
AwarenessControls the balance of education vs comparison vs direct offer.Medium
ObjectionSurfaces only the relevant handlers, FAQs and reassurance — and plants none of the others.Highest
ProofSelects the testimonial, case study, review or example that matches their situation.High
OfferSelects or emphasises the right service, package, product or bonus. The offer itself moves.Highest
TimelineAdjusts scheduling, urgency and implementation framing to their stated date.Medium
BudgetChanges which option leads, and how financing is presented. Never the number.Bounded
CTABuy, book, quote, apply, compare, watch, save, talk.Medium
FormatPage, proposal, report, recommendation, quote, comparison or summary — the asset type changes.Highest
v3 · one layer removed, one bounded
Emotional Mirror is out. It was in the v1 four-layer set as "page tone matches their emotional state". Removed for four reasons: inferring emotion from a free-text pain answer means inferring and potentially storing special-category data across any real population; it hands a plaintiff the argument that an AI vendor is processing sensitive prospect data; it drifts toward companion-chatbot statutes that carry a private right of action; and a system that presents as helpful discovery while optimising an emotional register for conversion is close to the fact pattern regulators have started describing as undisclosed steering.

It was also the least valuable of the set — tone shifts are the change a visitor is least likely to notice and the hardest to demo. What survives is register matching inside the Language Mirror, driven by the prospect's own observable diction rather than by an inferred internal state. Same visible effect, no inference, nothing sensitive stored.

Budget Mirror is bounded, not removed. Option hierarchy, financing presentation and which package leads may all change. The published price may not. Varying a price by inferred willingness-to-pay sits squarely inside the surveillance-pricing theory regulators are actively investigating — and it is the single claim your buyers will most want to make to their own clients in writing.

Reflection-strength guardrail

AI Mirror should not blindly repeat sensitive or awkward prospect language. Every asset declares a reflection strength per field:

Exact

Safe business language quoted closely. "the work that pays the bills"

Natural

Meaning preserved, wording polished. Reads written rather than transcribed.

Subtle

The concern is addressed without obvious repetition. Used for anything embarrassing.

Silent

Signal routes the asset and never appears in visible copy. Default for sentiment.

v3 · reflection strength is a safety mechanism, not a style setting
Treat Silent as the default for any field that could carry health, financial-distress or protected-class content, and make the escalation to Exact a deliberate, per-field decision made by whoever authors the pack. A single sensitive: true flag on the extraction object should force the whole asset down to a lower personalisation tier and suppress verbatim quoting entirely.
04 Mirror assets

A page is one output. The same conversation produces fifteen.

This is the core expansion from v1. The product is not a landing-page generator — it is an asset-generation layer, and the asset type is itself a personalisation decision.

Mirror Page

A personalised sales, recommendation or follow-up page.

Mirror Proposal

Current situation · desired outcome · recommended solution · scope · deliverables · timeline · relevant proof · pricing options · objection handling · approval step.

Mirror Quote

An individualised estimate experience. Strongest in home services and B2B services.

Mirror Recommendation

Answers: based on what you told us, what should you choose or do next?

Mirror Comparison

Services, packages, products, affiliate offers, properties or alternatives, weighted to the prospect's stated priorities.

Mirror Report

A diagnostic, assessment, audit or consultation report.

Mirror Action Plan

The conversation turned into a practical personalised roadmap.

Mirror Scope

A structured scope of work generated from a discovery conversation.

Mirror Offer

The most appropriate service, package or product, with the right framing around it.

Mirror Bonus Stack

For affiliates: the three to five bonuses most valuable to that specific buyer, out of twenty.

Mirror Booking Experience

The right service, the preparation information they need, relevant availability and CTA.

Mirror Application

Creates or pre-populates an application from the conversation.

Mirror Follow-Up

Personalised email, SMS, messaging-app message or follow-up page referencing what was actually discussed.

Mirror Recap

A polished prospect-facing summary of the conversation and the agreed next steps.

Mirror Seller Brief

Internal sales intelligence: lead summary, opportunity score, main needs, objections, decision criteria, buying readiness, recommended next action, CRM notes.

One conversation → multiple assets

The strongest single capability in the product. From one agency discovery call:

Prospect receives
  • Personalised proposal
  • Implementation roadmap
  • Matching case-study page
  • Approval or booking CTA
Seller receives
  • Discovery summary
  • Opportunity score
  • Objection list
  • Follow-up instructions
  • CRM-ready structured fields
Follow-up system receives
  • Personalised email and SMS sequence
  • Routing rules keyed to the actual objection
One conversation. Every sales asset you need.
v3 · the Seller Brief is the commercial insurance policy
Of the fifteen assets, the Seller Brief is the one worth over-investing in — because it is valuable at zero conversion lift. If personalisation delivers a modest +15% rather than something transformational, a structured intelligence report on every lead still justifies the subscription on its own. Almost nothing else in this market has a second reason to exist, and that is what protects the renewal rate.

It is also not a tenth mode (§05) — it is generated alongside every mode.
05 Mirror modes

Organise the engine by outcome, not by industry.

Industry is a content pack (§08). Mode is what the engine actually does. Nine modes cover the commercial surface.

ModeProducesWhere it earns its keep
Mirror SellSales pages, offer pages, adaptive sales experiencesThe demo asset. Fifteen seconds, no context needed.
Mirror ProposeProposals, scopes, pricing presentationsThe agency flagship. Turns a Tuesday call into a Tuesday-afternoon send.
Mirror RecommendProduct, service, package and affiliate recommendationsAffiliates and catalogue owners. Includes recommending not buying.
Mirror QuoteQuotes and estimate experiencesHome services and B2B. Competes on legibility, not on being cheapest.
Mirror CompareProducts, plans, services or optionsThe highest-converting affiliate format, written for the actual reader.
Mirror ReportAudits, assessments, diagnosticsThe foot-in-the-door for local consultants. Finally scales.
Mirror PlanRoadmaps and action plansDelivered with the purchase, it is the cheapest refund reduction available.
Mirror FollowFollow-up pages, messages, continuationsA continuation engine, not an email writer. See §06.
Mirror BookAppointments, demos, consultations, service bookingsFewer no-shows, and no first ten minutes re-running discovery.
v3 · do not sell nine modes
Nine modes on a sales page reads as feature bloat and makes a sharp product sound vague. Pick two per audience and mention the rest in one sentence. Affiliate: Sell + Plan. Consultant: Propose + Report. Course creator: Recommend + Follow. The full grid is in §10.
06 Intent-based routing

The engine should not build the same asset for every person.

This is what turns the product from conversation → content generation into conversation → understanding → asset selection → personalisation → next action.

High intent

Close

  • Proposal
  • Price and options
  • Approve / buy / book CTA
Medium intent

Resolve

  • Recommendation
  • Comparison
  • Proof and FAQ
  • Consultation CTA
Low intent

Educate

  • Education
  • Assessment or report
  • Action plan
  • Soft follow-up
v3 · routing is the most under-sold capability in the spec
Everyone in this market personalises copy. Choosing to build a different asset entirely — suppressing the proposal and issuing a diagnostic report because the prospect said "no timeline, I'm gathering information" — is a capability nobody else has and nobody else can screenshot.

Make the suppression explicit in the interface and in the demo. When the router withholds pricing, it should say it withheld pricing. A visible suppress([pricing, packages, close_cta]) is more persuasive than any amount of output.

Follow-Up Mirror — routing after the conversation

Follow-up is a first-class mode, and the route depends on what happened next.

What happenedWhat the engine sends
High intent, no actionConcise reminder plus the direct proposal, booking or payment CTA.
Medium intent, unresolved objectionThe case study, comparison, proof or FAQ specific to that objection.
Low intentEducation, recommendation or a low-friction next step — not another push.
Quote or proposal viewed, not acceptedA follow-up built around the sections actually viewed, the open questions and the known objections. Optionally an offer to continue by voice.
Prospect repliesThe reply becomes new Mirror data, updating the same profile, proposal and next-action logic.

A follow-up is not always another email. The engine may decide the right next move is a revised proposal, a simplified quote, a comparison page, a relevant case study, a short personalised video script, a booking page, another live conversation, or a "here's what changed since we spoke" page. It is a continuation engine.

Conversation Mirror Profile Mirror Asset Action, inactionor reply Follow-Up Mirror Next asset,call or CTA PROFILE UPDATES — CONTINUE UNTIL OUTCOME OR NURTURE
Fig. 2 — the closed-loop Mirror journey
07 Mirror Memory

The website should not forget what the phone call learned.

Memory is what turns a one-time asset generator into a persistent adaptive customer journey. It is also the single highest-risk feature in the specification, so it ships with constraints.

A roofing prospect previously said they had storm damage, were worried about insurance, and did not want to be pressured into a replacement. When they return, the engine should continue from that context — not open with "how can I help you?".

What memory powers

  • Returning conversations
  • Updated proposals
  • Follow-up pages
  • Email and SMS continuity
  • Phone-call continuity
  • Seller CRM context
  • Future recommendations

Example stored context

goal            10 qualified implant consults/mo
previous_attempt paid social
main_concern     lead_quality
budget_band      ~$3,000/mo
timeline         next_month
intent           medium_high
open_question    proof_of_qualification
v3 · memory constraints — decide these before a line is written
Cross-session persistence of verbatim prospect statements is where this product goes from "clever" to "liability", and the vertical list makes it concrete: a Dental pack whose signals include pain and anxiety is storing health data about an identified person.
  1. Store derived codes, never verbatim. The example above is correct — codes and bands, not sentences. Verbatim retention is a separate, off-by-default, consent-gated setting.
  2. Memory off by default in health-adjacent packs. Dental in particular needs a determination on whether the operator is a covered entity and whether you become a business associate. Until that exists, Dental should not be a launch pack.
  3. Single-session, transactional persona. No bot name, no simulated empathy, no relationship framing. Cross-session memory plus a human-like persona is precisely the combination companion-chatbot statutes describe, and those carry a private right of action.
  4. Visible and erasable. A returning prospect should be able to see what is remembered and clear it in one action. This is both a legal requirement in several regimes and, per the personalisation research, the version that converts better.

Seller-facing and prospect-facing, always in pairs

Prospect-facingSeller-facing
ProposalLead summary
PageOpportunity score
QuoteNeeds analysis
RecommendationObjections
ComparisonDecision criteria
ReportRecommended follow-up
Action planCRM fields
BookingSales coaching notes
08 Vertical Mirror Packs

The engine stays universal. Industry intelligence is packaged separately.

This is the architectural decision that makes every new vertical a content project rather than an engineering project — and it is also the recurring revenue and the defensibility.

Every pack contains

  • Discovery questions
  • Signal definitions
  • Routing logic
  • Allowed products and services
  • Offer rules
  • Proposal, page and quote templates
  • Objection library · proof library · FAQ library
  • CTA options
  • Follow-up sequences
  • Compliance guardrails

Candidate launch packs

  • Affiliate Marketing Pack
  • Agency / Consultant Pack
  • Roofing Pack
  • HVAC Pack
  • Dental Pack see §07
  • High-Ticket Pack
  • SaaS Pack
v3 · the three-layer split, stated as a rule
LAYER 3 — CAMPAIGN   owned by the customer
   their offer, links, proof, brand, their 3–4 questions

LAYER 2 — VAULT      owned by you, sold by niche
   master templates · slot maps · component libraries · fallback ladders

LAYER 1 — ENGINE     built once, never rebuilt
   conversation · extraction · routing · assembly · render · moderation
The engine knows nothing about dentists, roofers or affiliates. It knows slots, components, codes and tiers. A new vertical is a new pack. Never an engine change. If a vertical requires an engine change, the pack architecture is wrong.

Quantify the pack before promising the roadmap. A serious pack is roughly 58 components per template — 12 headline frames, 20 situation-tagged proof blocks, 6 offer configurations, 20 objection handlers — authored in two registers, across five templates. That is ~580 written pieces, about four weeks of one good direct-response copywriter, and it cannot start until the slot maps are frozen. Copywriting is the critical path, not code. This is the single most commonly missed line in a build plan of this shape.
09 Industry use cases

Seventeen verticals, mapped.

The signals each conversation surfaces, and the assets each one should produce. Treat this as a map of where packs could go — not as a roadmap commitment.

01

Affiliate marketing

flagship
Experience · goal · business model · budget · existing tools
Best-fit recommendationWhy it fits / where it doesn'tPersonalised bonus stack (3–5 of 20)Review / pre-sell emphasisWeighted comparison"Should I buy?" advisorPost-conversation page

The "Should I Buy?" advisor can credibly recommend not buying when fit is poor. It costs a percentage of sales and buys a list that opens the next twelve emails.

02

Local marketing consultants / agencies

flagship
Industry · current problem · desired outcome · prior failed channel · budget · timeline
Discovery call → proposalOpportunity reportScope of work30/60/90 implementation planFollow-up packSeller brief
03

Roofing

Storm damage · leaks · insurance concern · urgency · repair-vs-replacement uncertainty
Storm damage result pageInspection recommendationInsurance checklistRepair/replace comparisonQuoteProposalAppointment pageFollow-up
04

HVAC

Poor cooling · system age · repair uncertainty · financing concern · urgency
Recommended next stepRepair-vs-replace comparisonDiagnostic bookingOption-based quoteReplacement proposalFinancing informationFollow-up
05

Dental

health data — gate before launch
Desired treatment · cost sensitivity · pain / anxiety · event deadline · previous dental experience
Treatment information pageConsultation recommendationFinancing explanationRelevant patient storyBooking pageTreatment proposal

v3: pain and anxiety signals are health data about an identified person. Memory off by default, verbatim never stored, and a covered-entity determination required before this pack ships.

06

Solar

Electric bill · ownership horizon · battery interest · savings expectation · financing · scepticism
Suitability recommendationSavings explanationSystem optionsProposalQuoteFinancing comparisonConsultation booking
07

Landscaping / remodelling / home improvement

Project goal · style · deadline · budget · maintenance preference · must-haves
Project concept briefProposalScopeBudget optionsInspiration pageConsultation booking
08

SaaS / software

Team size · workflow · current tools · pain · governance concerns · desired outcome
Use-case pageDemo pathRequirements summaryRecommended planProposalImplementation planTrial CTA

A seven-person sales team worried about inconsistent follow-up sees the workflow, features and proof for follow-up automation and management visibility — not the full feature list. v3 note: this is Fibr's and Mutiny's home turf at $479/mo to $45k/yr with procurement attached. Serve it opportunistically; don't wedge here.

09

Coaches / consultants / high-ticket programmes

Current situation · desired transformation · previous attempts · blockers · urgency
Growth / strategy planProgramme-fit recommendationRelevant curriculumMatching case studyEnrolment proposalApplication or booking CTA
10

Webinar / training funnels

Intent tier from questions, chat or post-webinar voice
Personalised replay pageRecommendationFAQ pageOffer pageFollow-up sequence

High intent → offer and buy CTA. Medium → comparison, FAQ, proof. Low → replay, education, reminder. Replay traffic is the largest single wasted asset in info products.

11

Real estate

regulated — Fair Housing
Budget · bedrooms · commute · schools · location preference · timing
Buyer requirements briefShortlist criteriaProperty comparisonViewing planSeller intake summaryListing proposalMarketing plan

v3: "schools" and "location preference" as personalisation inputs is textbook steering exposure. Needs its own legal workstream before any pack is built.

12

Events / weddings

Date · guest count · style · budget · food and decor preferences · constraints
Event proposalPackage recommendationEstimateTimelineVenue / service comparisonDeposit CTA

Strong demo vertical: seasonality lets the engine volunteer bad news — "peonies, June only" — which is the cleanest available answer to "isn't this just AI slop?"

13

Automotive / dealership

regulated — ECOA if financing
Family / use case · budget · mileage · body style · priorities
Vehicle shortlistComparisonTrade-in / finance next stepDealer proposalTest-drive booking
14

Hospitality / travel

Party size · trip purpose · preferences · activities · budget · constraints
Stay recommendationPackage recommendationPersonalised itineraryRoom comparisonBooking offer
15

Recruitment

regulated — AEDT bias audit
Skills · experience · motivation · concerns · salary expectation · availability
Candidate-facing role briefGrowth opportunitiesInterview processRecruiter-facing candidate summary

v3: a candidate summary carrying a score is an automated employment decision tool. Published bias audits are required in some jurisdictions. This is a separate product, not a pack.

16

Professional services

Intake requirements where compliance allows AI-assisted intake
Intake summaryConsultation reportScopeEngagement proposalNext-step checklistFollow-up
17

B2B manufacturers / distributors

later market
Technical requirement · application · volume · specification constraints
Product recommendationConfigurationTechnical comparisonQuoteProposalSpecification summary
v3 · four of these are not verticals, they are legal projects
Dental, real estate, automotive finance and recruitment each carry a regime the other thirteen don't — health data, Fair Housing steering, credit discrimination, automated employment decisions. They stay on the map because the engine genuinely serves them. They come off the near-term roadmap until someone owns the legal workstream. Do not list them on a roadmap slide as if they were the same kind of thing as HVAC.
10 The three core audiences

Seventeen verticals is the map. Three audiences is the business.

Whoever we sell to buys the engine, not the vertical. These three are the ones with money, urgency and an existing habit of buying tools.

A1

Agency owner / local consultant

Discovery calls, proposals, retainers. Highest willingness to pay, longest cycle, and the only one with a real recurring-payment habit. Escapes the $47 ceiling.

A2

Affiliate marketer

Bonus pages, review pages, a list. Buys fast, refunds fast, competing against four hundred identical bonus pages. Cheapest acquisition.

A3

Course / digital product creator

Webinars, VSLs, applications. Their real enemy isn't conversion — it's refunds and non-completion. Best retention story.

Mode priority by audience

ModeProducesAgencyAffiliateCreator
Mirror SellPersonalised page●●●●●●●●●
Mirror ProposeProposal / scope●●●●●
Mirror QuoteEstimate / configuration●●●●●●●
Mirror RecommendWhat to choose●●●●●●●●
Mirror CompareWeighted comparison●●●●●●●●
Mirror ReportAudit / diagnostic●●●●●●●●
Mirror PlanRoadmap / study plan●●●●●●●●
Mirror FollowContinuation●●●●●●●●●
Mirror BookBooking●●●●●

The four layers, per audience

LayerAgencyAffiliateCreator
Language"tyre kickers" → "You're not short of enquiries. You're short of the right ones.""building every report by hand" → "Stop building every report by hand.""bought three courses, finished none" → "You don't need another course. You need to finish one."
ProofDentist sees the dental case at a matching practice size. Seven-figure logos suppressed.Five-client operator sees the four-client testimonial, not the $2M guru.Identity proof — the nurse retraining sees the student who was a nurse.
Offer"10 implant consults" vs "3 chairs full" → two different service packages.Bonus stack swaps 3-of-12 by stated goal.Curriculum re-orders; default tier flips from self-study to cohort.
Objection"burned on retainers" → fixed-scope clause. Price handler suppressed."another dashboard to learn" → setup call only. 1 of 14."I never finish anything" → completion mechanics. Lifetime access suppressed — it's a negative to this buyer.
v3 · the highest-leverage cell in the grid
Creator × Objection. For most course buyers, "lifetime access" is a benefit. For the buyer who just said they never finish things, it is a permission slip to procrastinate — so the engine suppresses it and renders cohort dates and completion mechanics instead. A static page cannot make that call in either direction. It's also the clearest single illustration of what the Objection Mirror does, and it belongs in the demo.

The long-form version of this section — three audiences across all four layers and all nine modes, with worked examples — lives in the partner kit.
11 Realtime-native architecture

The conversation is the generation process.

Adaptive Mirror must not wait for the conversation to finish and then generate. Mirror State, routing and asset composition update continuously while the prospect is still talking, so the finished asset can appear in a split second.

Provider-neutral engine layer

Support realtime engines through an adapter, never a hard dependency on one vendor.

RealtimeProvider
  connect()          sendAudio()        receiveAudio()
  receiveTranscript()  registerTools()  receiveToolCall()
  interrupt()        close()

implementations
  GeminiLiveProvider · OpenAIRealtimeProvider · FutureRealtimeProvider

Campaigns can eventually support Auto — the engine selected per conversation on channel, language, cost, latency, complexity and availability.

v3 · benchmark, don't commit
Named candidate models change faster than this document will. The adapter is the durable decision; the model list is not. Benchmark at least two realtime providers on the same scripted conversation and record time-to-first-token, turn-detection latency, tool-call reliability and cost per minute before writing anything provider-specific. Treat any model name in a spec older than a quarter as unverified.

The realtime model maintains Mirror State

The model isn't only speaking. While the conversation runs, it updates structured commercial understanding through narrow tool calls:

update_mirror_profile()   set_primary_problem()   set_goal()
set_timeline()           set_budget_context()    add_objection()
set_decision_criteria()  set_intent()            select_offer()
select_proof()           update_proposal_section()  set_next_action()

Worked example

Prospect: "We want more implant patients."

industry     = dental
primary_goal = more implant cases

Later: "We tried Facebook ads but most of the leads weren't serious."

previous_attempt  = facebook_ads
main_objection    = lead_quality
decision_priority = qualified_implant_prospects

The proposal, page or recommendation updates during the conversation rather than after it.

Separate conversation from asset rendering

The realtime engine is the listener, conversational brain and signal extractor. It should not spend the session writing a long proposal from scratch.

CHAT / WEBRTC / PHONE
        ↓
REALTIME CONVERSATION ENGINE
        ↓  tool calls
MIRROR STATE
        ↓
ROUTER + APPROVED LIBRARY + ASSET COMPOSER
        ↓
PROPOSAL / PAGE / QUOTE / REPORT / RECOMMENDATION
        ↓
INSTANT FINAL REVEAL

Continuous asset composition

Every asset has a predefined schema, and sections fill as answers arrive:

Proposal
├── Current Situation       / building
├── Desired Outcome         / building
├── Recommended Solution    / building
├── Deliverables           preset
├── Relevant Proof         selected dynamically
├── Timeline                / building
├── Pricing                approved / preset
└── Next Step               / building

Mirror State — the central object

{
  "profile": {},
  "intent": {},
  "recommended_route": "proposal",
  "recommended_offer": "implant_growth",
  "selected_proof": ["case_17"],
  "selected_objections": ["lead_quality"],
  "asset": {},
  "completion": 0.92
}

The frontend subscribes to Mirror State and updates visually as information arrives.

Adaptive questioning

Not a fixed questionnaire. The agent knows which fields the selected outcome still needs:

Problem             Goal      
Timeline    ?        Budget    ?
Objection           Intent    medium

It then asks the smallest number of further questions required to complete the asset and choose the route.

Prebuilt Mirror Blocks

Approved component libraries rather than generating every commercial statement.

Deterministic

Layouts · services · pricing · deliverables · guarantees · proof · allowed offers · CTA structures

Generative

Situation summary · goal wording · reasoning · connective copy · natural reflection of the prospect's language

Realtime retrieval

Proof, offers, bonuses, FAQs and templates are retrieved as soon as the profile has enough — not after the call.

industry = dental · service = implants · problem = lead_quality
  → retrieve matching case studies + objection blocks during conversation

By the time the call ends the system is selecting from already-prepared candidates, not beginning a search.

HTML-first, PDF second

Proposals, reports and quotes render first as responsive web experiences from Mirror State. PDF export is generated afterwards from the same structured asset.

Instant reveal

  1. Lock the final Mirror Profile
  2. Calculate final intent and route
  3. Validate required fields
  4. Select the approved offer and proof already prepared
  5. Render the already-composed asset
Generate during the conversation, not after it.
v3 · the latency budget, and the kill criterion
"Split second" needs a number. Assembly fires per answer, and the visible update must land inside 1.8 seconds at p90:
 turn detection → transcript      ≤ 300 ms
 extraction (streaming, strict JSON) ≤ 900 ms
 moderation + confidence scoring  ≤ 200 ms
 component selection + render     ≤ 400 ms
 ─────────────────────────────────────────
 visible update per answer        ≤ 1.8 s
This is a kill criterion, not a target. Prove it in a two-week spike on one template before anything else is built. If the loop can't hold 1.8s at p90, the magic doesn't exist and the product should not be built as specified — the entire differentiation is a latency phenomenon.

Two non-negotiables fall out of it: a strict JSON output schema (which is also the prompt-injection defence — the model cannot return off-spec content) and closed tag vocabularies so the router never has to interpret an invented category.
12 Technical architecture & build principles

The AI selects, organises and personalises. It never invents a commercial fact.

Conversation chat · voice · phone · file Mirror Listener transcribe + extract Mirror Profile 15 signals Mirror Brain Router Asset Composer approved library PageProposalQuote Report / PlanFollow-up / Book Seller brief→ CRM Mirror Memory APPROVED LIBRARY CONTROLS: PRICING · SERVICES · FEATURES · TESTIMONIALS · GUARANTEES · OFFERS · DISCOUNTS · AVAILABILITY CLAIMS
Fig. 3 — technical architecture

The build principle

Do not allow the AI to invent commercial facts. Approved libraries control:

  • Pricing
  • Services
  • Features
  • Testimonials
  • Case studies
  • Guarantees
  • Offers
  • Discounts
  • Bonuses
  • Availability claims
  • Compliance-sensitive statements

Generative text is used only inside bounded fields.

v3 · make provenance visible, and the fallback ladder explicit
Cite the source on every assembled block. "Package and deliverables lifted from Rate card › Implant Case Acquisition. Nothing here was generated." Provenance turns an invisible architectural virtue into a visible sales asset, and it is the strongest available answer to "does it hallucinate?".

Four rendered quality tiers, chosen by extraction confidence — not by hope:
TierTriggerWhat the visitor sees
T1All layers confidentFull mirror — their words, proof, offer, objections
T2≥2 layers confidentSoft mirror — goal and objections mirrored, generic headline
T3Thin or weak answersSegment-level page
T4Junk · abuse · injection · timeout · sensitive flagThe customer's ordinary page, unchanged
T4 is simultaneously the latency circuit-breaker, the adversarial-input defence and the legal safety valve. One switch drops to a page containing no visitor content at all.
13 Recommended UX

Progressive Mirror — never force an interview before anything happens.

  1. Visitor sees a valid default experience immediately
  2. AI asks one useful question
  3. Relevant content or recommendation changes
  4. AI asks the next highest-value question
  5. The asset becomes progressively more personalised
  6. The visitor can stop at any point and still receive a useful result

Phone

Output generated at the end of the conversation and sent by SMS, email or link.

WebRTC and chat

The asset visibly assembles during the conversation. This is the demo mechanic, and it is the reason the clip travels.

v3 · three UX rules that carry legal weight
  • Label the AI in-widget before the first message — not in a footer, not in terms. This single line satisfies several disclosure regimes at once and, per the personalisation research, overt disclosure converts better than covert.
  • State the purpose in one plain sentence. "I'll ask a few questions so we can show you the most relevant version of this." Near-free, and it defuses the undisclosed-steering theory almost entirely.
  • Notice at collection next to the input field, and a visible way to clear what's remembered.
14 Measurement

Avoid personalisation theatre. Measure commercial lift.

Track

  • Conversion rate
  • Booked appointment rate
  • Proposal acceptance
  • Qualified lead rate
  • Quote acceptance

And

  • Revenue per conversation
  • Revenue per visitor
  • Average order value where relevant
  • Time from conversation to next action
v3 · the holdout is the product, not a nice-to-have
Ship a permanent holdout control — a configurable share of traffic (default 10%) that never sees a mirrored asset — and put the incremental lift on the customer's dashboard as the headline number. Three reasons this matters more than it looks:
  1. It is the refund defence. A buyer who can see their own lift number at day 29 does not refund. A buyer who can't, does.
  2. It is the retention mechanism. The number only exists while they keep paying.
  3. It is the substantiation. It replaces claims you can't support with each customer's own data.
Plan on +10% to +40% on the converting step. That band is where vendors' own hand-picked best cases cluster. Nothing credible supports 3–5×; every such figure in this category traces to a hero block rather than a controlled test. Two of the best-funded companies in adjacent categories have quietly stopped publishing lift claims altogether. Build the copy on the mechanism and the demo — not on a multiplier that has to survive a refund window.
15 Risk & compliance register

Four design choices neutralise most of it.

This section did not exist in either source document. It is the one most likely to change what gets built, so it sits before the money and the plan rather than after them.

The four choices
  • Don't retain verbatim text beyond the session. Store derived codes.
  • Don't infer emotion. Sentiment routes; it never renders.
  • Label the AI before the first message, in-widget.
  • Contract the model vendor out of training and product-improvement use of customer data.
Together these address the large majority of the exposure below, and they cost almost nothing if decided now. Retrofitting them after launch is expensive.

Product and market risk

RiskAssessment and mitigation
Price / buyer mismatchThe load-bearing risk. Building runtime infrastructure comparable to a $479/mo enterprise product and selling it into a channel whose front-end mode is ~$17 and whose buyers have no recurring-payment habit. Mitigation is the agency motion (§16), built into the launch rather than bolted on after.
Personalisation theatreThe number-one refund driver. If output is "insert first name plus one phrase", buyers see through it. The Vault is the answer — and the 580-component authoring load in §08 is not padding, it is the product.
LatencyNot a risk — a kill criterion. Answered in two weeks (§11, §17).
Adversarial inputStrict JSON schema does most of the work; the four-tier ladder does the rest. Add an adversarial suite as a phase gate.
Fast-followingPositioning is copyable within a launch cycle. The Vault is not — 580 hand-written components per pack is a real moat, just not a technical one.

Regulatory exposure, ranked

Wiretapping / session-recording claims

High-volume, statutory-damages, private-right-of-action territory in some US states, with an active demand-letter industry. The specific hook is whether an AI vendor counts as a third party — which turns on whether they may use the data for their own benefit, including model training. Fix: contractual carve-out plus pre-chat consent.

Undisclosed steering

An AI presenting as helpful discovery while optimising for conversion is the textbook fact pattern regulators have begun describing. Fix: one plain purpose sentence, in-widget. Terms-of-service burial does not count.

Business-opportunity rules

Any agency, reseller or white-label tier carrying an income representation likely triggers disclosure-document obligations. Enforcement in this exact market has produced eight-figure judgments and permanent bans. Fix: sell capability and sub-accounts. Zero income claims anywhere.

Special-category inference (EU)

A free-text "biggest problem" box will surface illness, burnout and financial ruin across any real population, and a dataset containing one sensitive item can be treated as sensitive in its entirety. Fix: sensitive flag forces a lower tier and suppresses verbatim.

AI-interaction disclosure

Multiple regimes now require telling people they're talking to an AI at first interaction. Cheap, and the safe harbour is usually absolute. Fix: in-widget label.

Dark patterns

Countdown timers implying scarcity that isn't real are explicitly named in enforcement guidance. Fix: if a template ships a countdown, the deadline must be real — and the engine must never generate one where no deadline exists.

Not legal advice. Items in this section need counsel review before launch, and specifically before any agency or reseller page goes live.

16 Business model

Three motions, not one ladder.

The launch channel is the cheapest customer acquisition in the world for this buyer and the worst revenue model for this product. Use it for what it's good at.

Motion A — the launch acquisition

Purpose: 2,000–3,000 buyers, a case-study base and JV relationships. Not the revenue event.

TierPriceContents
FE$47Engine + 5 done-for-you master templates + 3 campaigns + 1,000 conversations/month
OTO 1$97Mirror Vault — 30 professionally written master templates. Highest take-rate; where the margin lives.
OTO 2$147Pro — unlimited campaigns and conversations, in-browser voice, multilingual
OTO 3$97Mirror Intelligence — aggregate voice-of-customer analytics across all their traffic
OTO 4$197Agency — 20 sub-accounts, commercial rights, white-label
OTO 5$67Live training — template authoring and personalisation mastery
Bundle$397Everything
v3 · OTO 3 changed, and why
The original OTO 3 was Phone Call Mirror as a BYOK-telecom unlock. Two findings kill it as a launch asset: it is not first-to-market (the AI-caller niche in this channel is crowded and actively relaunching at $27–$67), and the regulatory picture worsened materially (§02).

Mirror Intelligence is a better OTO on every axis. Zero incremental legal risk. Genuinely novel — nobody sells "here is what your market actually said, in their own words, at scale". Nearly free to build, because the data is already being extracted. And critically, it accumulates: a buyer three months in has three months of objection data they'd lose by leaving. That is the switching cost this product otherwise lacks, and the highest-leverage retention feature available.

Motion B — agency recurring the actual business

Consultants and agencies paying monthly to run Mirror on client sites.

The pitch is a ratio, not a feature: a consultant paying $197 runs it on ten client sites at $300/month each.

Motion C — direct SaaS, later

$49 / $99 / $249 per month, self-serve. Price-comparable to the mid-market personalisation tools and massively undercutting the enterprise ones. Do not open this until the Vault covers three or more verticals — a thin vault plus self-serve signup produces churn and bad reviews.

The bridge — build it before launch, not after

Metered conversations

1,000/mo resets; overage is a $27/mo top-up. Usage-based, so it only bills people getting value.

Vault packs as content

One new niche pack per month for subscribers. Highest-margin revenue in the stack.

Intelligence data gravity

Accumulated objection data is the switching cost.

Ladder OTO4 buyers

Anyone buying the agency tier is a Motion B prospect. Onboard personally — that cohort is worth 20× an FE buyer.

v3 · a unit error in v1, corrected
v1 projected "$200K–$500K MRR by end of year" from 30% of launch buyers retaining at $47/mo. Check it: $200K MRR at $47 requires ~4,250 retained subscribers, which at 30% retention needs ~14,000 FE buyers — roughly five times what a launch of the stated benchmark produces. The figure conflated MRR with ARR and is out by about an order of magnitude.
ScenarioFE buyersRetain @12moMonth-12 MRRARR run-rate
Conservative2,00010%~$9.4K~$113K
Base2,50015%~$17.6K~$212K
Upside3,50022%~$36K~$434K
30% twelve-month retention is aggressive for a buyer with almost no recurring-payment habit. 15% is the honest base case. Motion B is what gets past $45K MRR — not launch-buyer retention.

And the number worth optimising isn't launch gross. It's cost per agency customer acquired. If launch week nets $80K and delivers 150 agency subscribers at $150/mo, you have bought $27K of ARR plus the cash, for a fourteen-week build. That's the win condition.
17 MVP & build plan

Build the engine broadly. Demonstrate three radically different workflows.

MVP inputs

  • Imported transcript first
  • Chat
  • WebRTC voice

MVP assets

  • Personalised page
  • Proposal
  • Recommendation
  • Seller brief
  • Follow-up

MVP showcase packs

  • Affiliate Mirror — recommendation + bonus stack + pre-sell page
  • Local / Agency Mirror — proposal + seller brief + follow-up
  • High-Ticket / SaaS Mirror — recommendation + proposal + booking CTA

Three packs across three very different commercial models is the proof that the engine is genuinely universal.

v3 · phase plan with kill criteria — 13 to 16 weeks, not 6 to 10
The 80% reuse estimate is probably right for the infrastructure. It is wrong about where the work is: the critical path is copywriting.
PhaseWeeksOutputGo / kill criterion
0 — Latency spike2One template, one campaign, end to end, instrumentedPer-answer visible update ≤1.8s at p90. If it can't be hit, stop.
1 — Engine MVP4Extraction schema, routing, assembly, four-tier fallback, moderation, campaign wizardAdversarial suite renders T4 100% of the time
2 — Vault v14 (parallel from wk 3)5 master templates × ~580 components, written by a real DR copywriterBlind test: 10 marketers can't pick T1 output from hand-written
3 — Closed beta420–30 existing buyers running real traffic≥15 usable case studies with before/after numbers
4 — Launch assets4 (overlaps 3)Demo video, real-world landing mockup, sales page, JV kitA-tier JV commitment secured
Realistic total from a start today: launch Q4 2026 or Q1 2027. Q3 is not available.

Build-vs-buy, decided early: mature conversational-AI form products already ship the entire input layer for around $59/mo. If the Phase 0 spike struggles on the conversation side, prototyping the input against one of those while building the assembly layer is a legitimate de-risk. It also tells you something strategic — that category is one product decision away from being a competitor. Speed matters more than elegance.

Killer demo concept

Split screen. Left: live chat, WebRTC or phone conversation. Right: Mirror Profile and assets appearing progressively.

  1. What the AI understood
  2. The personalised proposal
  3. The personalised page or recommendation
  4. The seller follow-up brief

Then repeat quickly with an affiliate or local-service example.

One conversation. Four different assets. All written for one person.
v3 · the two assets that do most of the selling
1 — The real-world landing page mockup. Not the annotated split-screen; the polished page a real prospect actually lands on, with all four layers blended invisibly into normal copy. Build it twice — once affiliate, once local service. Those two artefacts sell both motions, and neither waits on the build.

2 — The fifteen-second before/after clip. Two pages side by side from two different conversations on the same offer. No narration, no annotations. Budget $3K–$8K for production and treat it as the largest single line item in the launch: it is the forwardable asset, and per the competitive scan, no competitor can currently produce it.

What already exists

18 Open decisions & next 30 days

Seven decisions, and they're yours.

DecisionRecommendation
Which motion leads?Launch on affiliate for acquisition, with the agency tier fully built and priced at launch. Simultaneous, not sequential.
Standalone or a tier inside the existing product?Standalone, with the hybrid beta — beta free with existing buyers in exchange for case studies, then launch publicly with 30–50 in hand. Solves the day-one credibility problem that kills new launches.
Does phone ship?Not in this funnel. Replace with Mirror Intelligence. Revisit as a separate product with its own compliance posture.
Is the Emotional Mirror out?Yes — replaced by the Proof Mirror. Better commercially, better across verticals, removes most of the exposure.
Vault pack #2?Local services. It's the audience the portfolio already owns and the doorway to Motion B pricing.
Memory on or off at launch?On, derived-codes-only, with a visible clear control. Verbatim retention off and consent-gated. Off entirely for health-adjacent packs.
Launch quarter?Q4 2026 or Q1 2027 — decided once Phase 0 returns a real latency number.

Next 30 days, in order

#ActionWhy it's in this position
1Phase 0 latency spikeHighest-information action available. Two weeks. Everything else is contingent on it.
2Build the real-world landing mockupsCornerstone asset for JV pitches and Motion B validation. Doesn't wait on the build.
3Hire the DR copywriter; freeze slot mapsFour weeks of authoring on the critical path. Starting late is the most likely cause of a slipped launch.
4Counsel reviewModel-vendor training terms, in-widget disclosure copy, retention default, agency-page claims. Cheap now, expensive after launch.
5Open the beta list20–30 existing buyers, four weeks of real traffic, case studies as the price of entry. Must be running before the JV window opens.
6JV shortlist and first pitches60–90 days out means now, on the strength of the mockup, before the build finishes.
7Lock the launch quarterOnce #1 returns a number.
The long-term product is not a document generator. It is persistent prospect understanding that continuously produces the next best sales experience.
19 Changelog

What moved between v1, v2 and v3.

Areav1 — Mirror Methodv2 — Adaptive Mirror 2v3 — this document
ScopePersonalised sales pageConversation-to-Outcome Engine, 15 assetsUnchanged from v2, with the engine / vault / campaign split stated as an architectural rule
ChannelsBrowser chat, phone in OTO3Chat, WebRTC, phone, transcriptSame four. Transcript promoted to first; phone out of the launch funnel
Layers4 (Linguistic, Emotional, Goal, Objection)11 layers + reflection strength11 minus Emotional; Proof elevated; Budget hard-bounded to presentation only
Positioning leadHeadline mirroringMulti-asset generationComponent re-composition — headline generation is a funded, crowded category
MemoryNot presentPersistent cross-session, verbatimDerived codes only by default; verbatim consent-gated; off for health-adjacent packs
OTO 3Phone Call Mirror (BYOK)Mirror Intelligence — novel, free to build, creates the switching cost
Business modelOne launch ladderNot addressedThree motions: launch / agency recurring / SaaS
MRR projection$200K–$500K MRR by year end$9K–$36K from launch buyers. v1 conflated MRR with ARR
Timeline6–10 weeks13–16 weeks. Copywriting is the critical path, not code
Latency">3s and the magic dies""Split second"Budgeted: ≤1.8s p90 per answer, as a kill criterion
Lift claimsImplied transformational"Measure real lift"+10% to +40%, plus a permanent holdout control shipped as a product feature
Risk register5 product risksNot addressedFull product, market and regulatory register (§15)
VerticalsAffiliate only17 industries17 retained; 4 reclassified as regulated markets needing their own workstream
PrecedentNot addressedNot addressedTwo venture-backed teams shipped this mechanic and left. Their failure mode is now a design constraint

Companion documents

Partner kit — features, benefits, angles, hooks, full use-case book.
v1 interactive demo — four scenarios, dual mode.
v2 realtime demo — channels, sources, tool calls, routing.

Still unwritten

Real-world landing mockups. Five-email promo sequence per angle. Pack authoring guide. Holdout measurement spec. Counsel memo.

Single biggest unknown

Whether the 1.8-second loop holds at p90 on real infrastructure. Two weeks of work answers it, and it changes everything downstream.

Adaptive Mirror v3 — master product document. Supersedes the original Mirror Method thread and the Adaptive Mirror 2 expansion; both are preserved in full within it. Prices, tiers, timelines and vertical priorities are working assumptions, not commitments. Figures cited for prior launches are historical and are not projections. Nothing in this document is legal advice; §15 requires counsel review before launch.