How this guide was built
We built a fixed nine-part prompt specification and a 20-point review rubric, then checked current official documentation for involve.me, HighLevel, and Perspective on September 9, 2026. The review distinguishes ongoing conversational funnel editing, AI page generation, and connected funnel generation. This is a reproducible specification, not a claim that every product can create every requested layer or that AI output is ready to publish without review.
What is an AI funnel prompt?
An AI funnel prompt is a structured instruction that tells a model what funnel to create, which business decision it supports, and how the result will be tested. A complete prompt covers more than copy. It defines the visitor path, logic, scoring or formulas, personal outcomes, stored data, follow-up actions, and operational boundaries.
The prompt does not replace product strategy. A model should not invent who qualifies, which regions are served, what a service costs, or which claims are approved. Those facts belong in a reviewed source block supplied by the business.
Which nine parts belong in the prompt specification?
Give each part its own output block so a reviewer can find omissions without interpreting a long narrative response.
| Part | What to provide | What the output must show |
|---|---|---|
| 1. Job | One funnel family and one measurable action | Checkout, client workflow, or interactive qualification path |
| 2. Audience | Role, situation, need, and source context | Language and examples that fit the intended visitor |
| 3. Offer | Promise, boundaries, supported cases, and next action | No invented price, guarantee, or eligibility rule |
| 4. Evidence | Facts the funnel must collect | Only questions that change a route or support the handoff |
| 5. Decision | Result bands, formula, score, and hard overrides | Inspectable rules with stable result identifiers |
| 6. Outcomes | Explanation and action for every result | Meaningfully different result pages, not renamed generic copy |
| 7. Record | Contact fields, source, consent, answers, score, and result | A complete handoff contract for later systems |
| 8. Follow-up | First message, delays, branches, and stop conditions | Actions that match the result and stop after booking or purchase |
| 9. Tests | Fixed inputs and expected outputs | High, middle, low, override, error, and mobile cases |
What complete AI funnel prompt can you copy?
Build a [funnel family] funnel for [audience] who need to [decision or job]. The final business action is [checkout, booking, recommendation, or routed handoff]. Use only these approved facts: [facts]. Do not invent prices, eligibility, evidence, guarantees, or integrations.
Collect only the inputs needed to decide [decision]. Ask no more than [number] questions before contact details. Use these hard overrides: [rules]. Use this score or formula: [model]. Create these result identifiers and routes: [result ID, explanation, next step]. Every result must explain why it fits the submitted inputs.
Store [contact fields], [source fields], [answers], [score or calculated value], [result ID], and [consent]. Draft the first follow-up action for each result and include these stop conditions: [booking, purchase, opt-out, manual closure]. Keep the result page, stored record, and follow-up action consistent.
Return the flow map, question list, rules, scoring table, result copy, contact schema, follow-up map, and test cases separately. Flag any requested feature that the platform cannot create or that still needs manual configuration.
How does the template work for a B2B readiness assessment?
Imagine an operations consultancy assessing whether a company is ready for a six-week implementation. The approved facts define supported company sizes, regions, delivery capacity, and the minimum access the project requires. The funnel asks about the active process problem, measurable consequence, deadline, internal owner, data access, and implementation capacity.
The score rewards a defined problem, active owner, near-term event, and available resources. Unsupported region is a hard override, not negative points. A high-fit result offers a project review, a preparation result gives a three-step readiness plan, and an unsupported result explains the boundary. All three routes store the same result identifier shown on the page and used by the next action.
| Test case | Fixed inputs | Expected result |
|---|---|---|
| High fit | Defined problem, owner, access, supported region, near-term event | project_review |
| Preparation | Clear need but no owner or data access | readiness_plan |
| Low urgency | General interest and no active event | educational_route |
| Override | Strong score but unsupported region | unsupported_region |
| Contradiction | Urgent deadline but no implementation capacity | readiness_plan with a capacity explanation |
How do you score the generated funnel out of 20?
Give zero, one, or two points for each of ten checks: correct family, complete structure, specific copy, valid question order, inspectable logic, hard overrides, distinct outcomes, complete contact schema, matched follow-up, and executable tests. Two points means the layer is present and internally consistent. One means present but incomplete. Zero means absent or contradictory.
Do not combine the functional score with a design review. A polished visual can hide missing routing, and a logically sound draft may simply need brand styling. Record functional and visual findings separately.
- 18 to 20: complete enough for detailed human QA, not automatic publication.
- 14 to 17: useful draft with named gaps that require configuration or repair.
- 8 to 13: partial build that should not be treated as a complete funnel.
- 0 to 7: mostly copy or page structure, with core decision layers absent.
How do current AI creation models differ?
Current official material describes different boundaries. involve.me documents an AI Agent that builds interactive funnels with native components, then accepts later chat instructions to adjust design, functionality, copy, logic, and other elements. This is broader than a one-time creation screen. HighLevel documents guided and prompt-based creation for funnel or website pages, with chat-based refinement of copy, sections, layout, and design. Perspective describes one-prompt mobile funnels with qualification logic and connected hosting, analytics, CRM, and email sequences. These are vendor descriptions, not fresh authenticated tests by AI Funnel Index.
A careful buyer should run the same prompt in each shortlisted product and record what appears before manual work. Feature availability elsewhere in the platform does not prove that the AI created it, and a generated asset still needs business, accessibility, data, and route validation.
Sources: involve.me AI Agent, HighLevel Funnel and Website AI, Perspective MCP and AI builder
Where does the prompt still need human review?
Review every eligibility rule, price, claim, calculation, result, consent statement, integration, and stop condition. Test contradictory inputs and failure states. Confirm that a person can understand and edit what was created. A model may produce plausible logic that does not reflect the business policy, or copy that overstates what a result can prove.
AI output should also respect the product's actual scope. A page generator may not create scoring or connected follow-up. A connected platform may still require account settings, sender verification, payment configuration, domain setup, consent text, or integration credentials.
The decision in one paragraph
Treat the prompt as an auditable build contract. Supply the business truth, require nine clearly separated output layers, and test the same result identifier across the page, contact record, and next action. A strong prompt makes missing capabilities and manual work visible before the funnel reaches production.