All guides

Create with prompt

Create a chatbot with your chat AI

Turn a quick business description into a complete conversation flow — a visitor answers a few good questions and becomes a booked appointment or a qualified lead — without designing it question by question. ChatMakeee writes the prompt, your chat AI (ChatGPT, Claude, or any model) designs the flow, and you import it with one paste. No AI credits are ever spent on the ChatMakeee side: the prompt runs in your own chat AI.

You are designing the visitor conversation for a ChatMakeee chatbot
for the business below.

Business name: Northwind Dental
Website: https://northwind.example.com — if you can browse, fetch this page
  first and extract the services, pricing, hours, and tone from it. …
Business description: A family dental clinic offering check-ups, …
Who they talk to: New and existing patients in the area — write questions
  your patients will instantly understand.

What this chatbot must accomplish:
- Capture leads — end the flow by asking for contact details …
- Book appointments — include one book_meeting question near the end …

Approach this as a booking and lead concierge: keep the flow short, warm …
Write every question and option in a friendly, welcoming voice …

Rules for the conversation:
- Ask 4 to 8 questions max. Keep each question short (under ~80 characters) …
- Only use these question types: text, number, email, phone, date,
  single_select, multi_select, book_meeting, file_upload.
- Branching works by nesting child questions inside an option's "children".

Output format — return ONLY a JSON array of question objects. …

[ …a complete working example flow in the simplified format… ]

Now design the conversation for the business above. Return only the JSON array.

Where to find it

Open New chatbot from your dashboard. The first screen offers three ways to start — Start from scratch, Create with prompt, and Template gallery. Pick Create with prompt and a three-step wizard takes over:

  1. Business & role — describe the business and choose how the AI should sound.
  2. Prompt — copy the tailored prompt into your chat AI.
  3. Import — paste the AI output, watch it validate, and open it in the builder.

You can create a public shareable link page this way too: use New shareable page and choose Create with prompt. The same flow becomes a page you can share anywhere instead of a widget on a website.

How it works — outcome first

Every step exists to get a visitor from “I arrived on your page” to “I’m booked” or “I left my details and I’m qualified”. The mechanism is simple on purpose: the wizard gathers what a good intake conversation needs, the AI drafts it, and ChatMakeee validates that the draft can actually run.

  • No AI meter, no tokens, no per-message fees. ChatMakeee does not call a model for this — it assembles a prompt and validates JSON. Flat pricing only.
  • Only real capabilities. The prompt restricts the AI to the nine ChatMakeee question types, so what it returns can be imported and run — nothing is invented.
  • Nothing silently discarded. Everything the importer changes or rejects is reported as a blocker or a warning.

Step 1 — Describe the business and choose the AI’s role

Everything here feeds straight into the prompt. The more honest the input, the sharper the questions the AI writes.

  • Business name — used as the working name for the chatbot.
  • Website URL (optional) — if your AI supports browsing (ChatGPT, Claude, and Perplexity all do), include it. The prompt tells the AI to fetch the page first and pull services, pricing, hours, and tone from it, so the questions match the real business. No browsing? The AI falls back to your description.
  • What the business does (required) — a sentence or two. Required unless you provided a website URL.
  • Who it talks to — the audience, e.g. “new and existing patients in the area”. Questions are written in words that audience instantly understands.

Then choose the assistant:

  • Assistant role — how the AI approaches the conversation:
    • Booking & lead concierge — short, warm, decisive; steers toward a booking or contact details.
    • Qualification specialist — asks the few facts that decide which path the visitor belongs on.
    • Support author — resolves common questions with clear option-based answers.
    • Sales guide — recommends the right service or product, then collects details or a booking.
  • Voice — the wording tone for every question and option: Friendly & warm, Professional, Direct & concise, or Playful.

Finally, pick what the visitor should be able to do (tap all that apply). Every option maps to a real capability:

  • Capture leads — end by asking for contact details so each visitor becomes a saved lead.
  • Qualify visitors — option questions branch into the right follow-ups.
  • Answer common questions — visitors pick a question and get a short answer.
  • Recommend options — guide visitors to the service or product that fits.
  • Book appointments — a booking step near the end books automatically.
  • Collect files — a file-upload step attaches documents to the lead.

Step 2 — What is inside the prompt

The wizard assembles one copy-paste prompt from your answers. It contains:

  • Your business section — name, website (with the browsing instruction), description, and audience.
  • The goal lines for each outcome you chose, phrased as things the conversation must accomplish.
  • The persona line from your chosen role, and the voice line from your chosen tone.
  • Conversation rules — a visitor-outcome-first flow, 4–8 questions, questions under about 80 characters, options under about eight words, required email/phone for lead capture, and branching by nesting children inside an option.
  • The locked vocabulary — only text, number, email, phone, date, single_select, multi_select, book_meeting, and file_upload; no IDs, variables, conditions, or position fields.
  • A strict output contract — return only a bare JSON array, no fences or commentary, and reply with exactly [] if it cannot build a valid flow.
  • A complete working example flow and links to the Flow format and this guide.

Because the shape is fixed and an example is included, most models return something importable on the first try.

Step 2 — Run it in your chat AI

You will always see the prompt before anything runs. The card shows a summary of your choices (business, role, voice, outcomes) and a Copy prompt button.

  1. Open chatgpt.com or claude.ai (links are on the step).
  2. Paste the prompt. If your AI offers search or browsing, turn it on so it can read the website URL you provided.
  3. Copy only the JSON block it returns, then continue.

Not sure what good output looks like? Expand What a good answer looks like to see a full example flow, or press Use this example instead to load that sample and skip straight to the import step.

Step 3 — Paste and watch it validate

Paste the AI output into the box. You do not press validate: the flow is checked as you paste (a short debounce after typing stops). A live status line tells you the moment it is importable — for example “Looks good — 8 questions, no blockers” — or how many blockers and warnings remain.

The importer tolerates real-world AI output: it finds the JSON even with surrounding prose and strips markdown fences automatically. Give your chatbot (or page) a name — it defaults to the business name — then press Import & open in builder.

Blockers vs warnings

Every issue is listed with its path in the flow and, for common failures, a plain-language fix.

  • Blockers stop the import until fixed: pasted text that is not JSON, an empty flow, an unsupported question type, an empty question text, a select question with no usable options, or nesting deeper than six levels.
  • Warnings describe what was already handled, so nothing changes silently: unknown fields are ignored, a missing type becomes text, IDs and variables are generated for you, and an unusually large flow is flagged.

The flow is validated against the same rules the builder uses, including every nested branch, so you never import a broken conversation.

What gets generated for you on import

  • IDs and variables — created automatically and kept unique across the whole flow, including nested branches.
  • Booking settings — each book_meeting question receives your default availability; adjust it anytime.
  • File-upload settings — each file_upload question gets the default accepted types and size limit, editable per question.

The bundled example flow

The wizard ships a complete, realistic example you can reuse. It shows a dental-clinic intake: a greeting and name, a new-vs-existing branch with nested follow-ups, an email, a phone number, an optional file upload, a booking step, and a final multi-select. Press Use this example instead or Paste an example flow to load it and try the import end to end.

After import — edit it like any chatbot

Once imported, the flow opens in the builder as a normal chatbot (or shareable page). Change wording, reorder questions, add or remove options, adjust booking availability and file settings, and preview it the way a visitor would — then save and publish. See the Flow format guide for the exact JSON the import accepts and the full list of auto-fixes.

Cost and privacy

There is no AI model, no token meter, and no per-message fee inside ChatMakeee. The prompt is generated locally in your browser and runs in your own chat AI account, so your business details go only to the AI you choose. ChatMakeee only ever sees the JSON flow you paste back, which is validated immediately.

Troubleshooting

  • “The pasted text is not valid JSON” — copy just the JSON block the AI returned. Prose around it is fine; markdown fences are stripped automatically.
  • The AI wrapped the JSON in an explanation — that is okay; paste the whole reply and the importer extracts the block for you.
  • A blocker keeps reappearing — read its Fix hint. Most come from an unsupported type or a select with no options; ask the AI to follow the prompt’s type list exactly.
  • Missing email or phone — to capture leads, the flow must include at least one email or phone question. Add it to the prompt’s outcomes and re-run the AI.
  • Branching looks flat — follow-up questions must live inside an option’s children, not beside it. See the Flow format guide.