Skip to content
SEO2026

Templates · AI website prompts

AI Website Prompts: Detailed Templates for Websites, SaaS and Apps

Asking a builder to make a 'modern site' often leads to repeated redesigns. A good prompt gives a real project scope, functional pages, design constraints, data rules and acceptance checks.

Vendor plans change. This guide does not guarantee rankings, indexing or business results.

Free implementation brief generator

Don't start from an empty chat. Choose the type of site or app you want, customise a detailed specification and copy it.

Generate my prompt
01

A prompt for a service website

Specify the exact offer and target customer. Request public routes for home, services, about, contact and privacy. State what each CTA should do. A form should deliver an actual enquiry and handle validation failures. No fake logos or testimonials.

Include mobile-first layout, semantic HTML, a unique H1 and a coherent canonical/sitemap structure. If business details are missing, use clearly labelled placeholders instead of invented addresses.

EXAMPLE / CHECK
Build a consultation website for an interior design studio.
MVP: home, 3 service pages, licensed portfolio,
working contact delivery and privacy notice.
No invented reviews. Test 360px and raw HTML.
Show code changes and don't deploy without approval.
02

A prompt for a SaaS or AI app

Describe one full journey: sign in, create an object, edit, save and reopen it. Explain the data ownership and permissions. Don't claim an AI feature works until the real server-side model integration and budget controls are configured.

Define two-user isolation, refresh persistence and failure states as explicit acceptance tests. Avoid prematurely requesting team billing and a dozen optional screens.

EXAMPLE / CHECK
Build a task-planning SaaS.
Users sign up, create a project, add tasks,
save and reopen them in a new session.
Enforce owner-scoped access on the backend.
Test two accounts, reload and network failure.
03

A prompt for a store and a CRM

For a store, specify products, variations, cart, server-authoritative totals and sandbox order processing. For a CRM, specify roles, record list, filters, edit actions and server permissions. These systems involve greater operational risk than a static landing page.

Require an honest payment or save status, not a fake success toast. Test permissions through an API call rather than only checking which buttons appear in the interface.

EXAMPLE / CHECK
Build a support CRM for three agents.
Routes: login, tickets, ticket detail, admin.
Agent accesses only assigned tickets; admin sees all.
Persist status changes and deny foreign ticket IDs
at the API boundary, not just in React.
04

A precise prompt for fixing a technical SEO bug

Once a site exists, avoid 'rewrite all SEO' requests. Supply the affected route, observed HTTP output, expected result and the list of files that may change. Ask for a focused regression test and a reviewable pull request.

For a sitemap error, require well-formed XML with one entry per real canonical URL. Protect unrelated payment, authentication and editorial logic from modification.

EXAMPLE / CHECK
Fix /sitemap.xml only: HTTP 200, valid XML,
one entry per canonical URL and reciprocal hreflang.
Do not touch auth, checkout, book content or UI.
Add an XML test and provide a focused diff.

Practical library / No registration

7 implementation prompts with testable outcomes

No vague “make a modern website”. Each brief includes concrete routes, real data flows, security, SEO and a pass/fail test. Preview every prompt and copy it independently.

01 / Implementation brief

Conversion-ready landing page

Outcome: Lead form with verified delivery

Show the full implementation prompt

Copy locally; nothing is submitted to a server.

02 / Implementation brief

Fix Lovable SSR and indexing

Outcome: Crawlers receive useful HTML pages

Show the full implementation prompt

Copy locally; nothing is submitted to a server.

03 / Implementation brief

SaaS with real accounts and RLS

Outcome: Two users cannot access each other's data

Show the full implementation prompt

Copy locally; nothing is submitted to a server.

04 / Implementation brief

Store: cart, orders, no fake checkout

Outcome: Order with server-verified total

Show the full implementation prompt

Copy locally; nothing is submitted to a server.

05 / Implementation brief

CRM: permissions and audit trail

Outcome: Permissions enforced by backend, not hidden buttons

Show the full implementation prompt

Copy locally; nothing is submitted to a server.

06 / Implementation brief

Move from Lovable to GitHub without SEO loss

Outcome: Preserve routes, data, auth and redirects

Show the full implementation prompt

Copy locally; nothing is submitted to a server.

07 / Implementation brief

Programmatic SEO without doorway spam

Outcome: Useful data-backed pages without thin duplicates

Show the full implementation prompt

Copy locally; nothing is submitted to a server.

Pass/fail checks are part of every brief. A generated screen does not prove functionality, security or indexing; verify the result.

Comparison and acceptance criteria
TemplateUseful outcomeAcceptance
Service siteReal enquiryDelivered successfully
SaaSSaved objectReload and second account
StoreOrder recordServer total and sandbox
CRMRole-specific accessForeign ID denied
SEO bugCorrect HTTP and markupcurl and CI

A practical first-build scenario

A reusable eight-part AI builder prompt

  1. 01

    Describe the product, user and MVP boundary.

  2. 02

    List the pages and the main user journey.

  3. 03

    Specify data, design, SEO and privacy.

  4. 04

    End with verifiable tests and a reviewed code diff.

Definition of done: The prompt is good when another person can verify the resulting application without trusting a marketing claim.

Field workbook · outcomes, not theory

Turn a vague AI website prompt into a real, testable change request

The best prompt is not the longest one: it specifies one failing URL, a concrete expected outcome, forbidden modifications, and proof of success. Copy this example and replace its domain, route and acceptance checks.

For whom and when: Owners of an existing Lovable/React site who want a precise fix without unexpected redesigns or credit waste.

Copyable SSR bug-fix prompt

Replace placeholders. Keep secrets out of the prompt and verify outcomes against production, not a chatbot's summary.

TXT
ROLE: Senior React / TanStack developer.
ISSUE: /services/consulting returns 200 but no useful H1 in initial HTML.
GOAL: Server returns title, H1, full main content and self-canonical.
SCOPE: Only relevant route/loader/head; no auth, payments, UI redesign or DB changes.
STEP 1: Reproduce with HTTP GET without JS.
STEP 2: Identify root cause and enumerate files before changes.
STEP 3: Implement the minimum code patch and regression tests.
PASS: Original URL -> 200 + H1 + content + canonical.
PASS: Unknown URL -> actual 404.
PASS: Existing auth and lead form still work.
DELIVER: Full diff, real CI logs and live URL evidence.
Never deploy before owner approval.

Do this in order

  1. 01

    Identify one reproducible defect

    Provide failing URL, raw HTML, current output and expected behaviour rather than 'improve SEO'.

    Evidence: A second developer can reproduce without guessing.

  2. 02

    Constrain the scope

    List relevant modules and explicitly exclude billing, access controls and unrelated design components.

    Evidence: Diff only touches permitted paths.

  3. 03

    Define pass/fail criteria

    A real form delivery, blocked private API, H1 in initial HTML, or 404 status is testable. Vague 'modern/beautiful' is not.

    Evidence: Each requirement has an observable result.

  4. 04

    Demand independent evidence

    Require build, typecheck, browser test and final-domain HTTP results before declaring completion.

    Evidence: Links to actual logs and smoke reports.

Troubleshooting the failed result

Builder rewrites unrelated components

Likely cause: The prompt lacks protected scope.

Action: Split into a small PR with blocked file groups.

Agent claims success without tests

Likely cause: No evidence requirement.

Action: Reject completion until CI and production results exist.

Same prompt fails on another framework

Likely cause: Assumed architecture differs.

Action: State version, router, backend and exact file boundaries.

Acceptance checklist

0/6

Check only outcomes you've verified yourself. Your progress stays in the current page session and is reset on reload.

Verify every criterion before launch.

A written brief does not ensure correctness; tests and code review remain necessary.

Questions before you begin

Does one prompt work in all AI builders?

The core specification is reusable, but platform integrations and deployment instructions differ.

How do I prevent accidental deletion?

Restrict file scope, require diffs and test critical user journeys.

Should prompts be thousands of words long?

No. Complete, testable requirements are more useful than repeated filler.

Next: validate your result

After building an app, test its real Google search eligibility and measurement. These are independent acceptance guides, not more AI prompts.

Related reading

Sources and technical documentation