Contact form security: spam, bots and data leaks
Contact form security in practice: how to stop spam and bots, block injection and abuse, and keep personal data from leaking, without hurting real leads.
AI chatbot
AI chatbot for your website, trained on your content
A custom AI chatbot for your website. It answers visitors from your own content, captures leads in the conversation and hands off to a person when it cannot help. AI model usage and hosting are billed to your accounts by the providers.
2 knowledge sources included
Who it's for
For companies whose site gets the same questions every day, and who want answers and leads outside office hours.
- Sites with clear content to answer from: product pages, pricing, policies, help articles
- Teams that miss leads at night and on weekends
- Support teams answering the same questions by email
- Companies selling in several languages
Probably not a fit if
- You have little written content for the chatbot to answer from yet
- You need a chatbot that takes payments or makes binding decisions on its own
What's included
- Answer rules: what the chatbot answers, refuses and hands off
- Chat widget designed to match your site, desktop and mobile
- Answers drawn only from your knowledge sources, with links
- Lead capture inside the conversation, sent to your inbox or CRM
- Hand-off to a person by email, with the chat transcript
- Test set of real questions, run before launch and after changes
- Conversation log, so you see what visitors ask
Not included
| Item | If you need it | Price |
|---|---|---|
| AI model usage and hosting fees | Billed by the providers to your account, estimated in the proposal | Quoted |
| Writing new help content | Add copywriting | $1,800 |
| Chat events in your analytics tool | Add analytics setup | $800 |
| Ongoing tuning after launch | Add a care plan | $1,500 |
| Voice, phone or messenger channels | Quoted after discovery | Quoted |
Custom AI chatbot development vs an off-the-shelf widget
Subscription chat widgets are quick to switch on. A custom chatbot takes longer, but you control what it says and where your data goes.
If an off-the-shelf widget does the job for you, we say so on the first call.
| What matters | Custom chatbot | Off-the-shelf widget |
|---|---|---|
| Answers from | Only the sources you choose, with links to them | Whatever the tool indexes, under its own rules |
| Design | Designed to match your site | The vendor's widget in your colors |
| Leads and hand-off | Into your inbox, CRM or help desk | Into the vendor's inbox, often on a paid tier |
| Data and accounts | Model account and logs in your name | Stored by the vendor |
| Cost | From $12,000 once, plus usage fees | A monthly subscription, often per seat or per conversation |
How we build it
A good chatbot is mostly content work and testing. The code is the smaller part.
UX/UI design
We design the conversation and the widget: greeting, suggested questions, lead form and hand-off, on desktop and mobile.
Your stack, your choice
We recommend a stack and you pick: the bot's server on Next.js route handlers, Hono or Python, with PostgreSQL where it needs storage, hosted on Vercel or AWS. The widget can sit on any site, not only one we built.
Headless CMS as a source
If your content lives in a headless CMS, the chatbot can read from it as a knowledge source. Setting up a new CMS is not part of this product.
Multilingual
The chatbot answers in each language you add in the calculator, and we test every language with its own set of questions.
Answers and actions
The bot answers from your own content through retrieval (RAG with embeddings) and links its sources. With tool use it can also act: book a call, look up an order or create a lead in your CRM. On request we add an MCP server, so your team's own AI tools can query the same content.
Model choice and guardrails
You pick the model: Claude, Gemini or OpenAI, or Claude through AWS Bedrock if you run on AWS. Guardrails keep it on topic: answer rules, prompt-injection guarding and a hand-off to a person.
- AI
- Anthropic Claude API
- Claude on AWS Bedrock
- Google Gemini API
- OpenAI API
- RAG with embeddings
- LLM agents with tool use
- MCP servers
- prompt-injection guarding
- Back end and APIs
- Next.js route handlers
- Hono
- Python
- TypeScript
- Data
- PostgreSQL
- Drizzle ORM
- Redis
- Email and notifications
- Telegram bots
- Viber
- Microsoft Teams
- Resend
- Integrations
- Salesforce
- n8n
- Google Calendar
- Cloudflare Turnstile
- Cloud and infrastructure
- Vercel
- AWS Lambda
- Quality and observability
- OpenTelemetry
- Sentry
- Vitest
- Security and supply chain
- rate limiting
- secret management
- dependency scanning
- SBOM
Timeline, revisions and price
Extra knowledge sources, integrations and each extra language add time. Testing with real questions is part of every schedule.
Discovery and content audit · 1 week
› Your sign-off: Question list and source list
Conversation and widget design · 1 week
› Your sign-off: Answer rules and widget design
Knowledge base and answers · 2 weeks
› Your sign-off: Answers on staging, with sources
Leads, hand-off and integrations · 1 week
› Your sign-off: Test leads and hand-offs received
Testing with real questions · 1 week
› Your sign-off: Test set results
Launch and tuning · 1 week
› Your sign-off: Live chatbot and handover
Revisions and sign-off
- Two revision rounds per stage.
- You approve the answer rules: what the chatbot answers, what it refuses and when it hands off.
- Before launch we run a test set of real questions and share every answer with you.
- Changes after sign-off go through a priced change note.
Running costs
Two costs continue after launch. Both are billed to your own accounts by the providers, not by us, and neither is part of the $12,000 price.
Estimate · billed to your accounts
| Cost | Billed by | What changes it |
|---|---|---|
| AI model usage (API fees)The AI provider, to your account | Charged per use: more conversations and longer answers cost more | |
| Hosting and databaseYour hosting and database providers | Often a free or low-cost tier at small volume, growing with traffic |
- Both costs depend on how many people chat and how long the conversations are, so we do not print one number here. The proposal estimates a monthly range from your traffic.
- Where the provider allows it, we set a monthly spending limit on your account, so a busy month cannot surprise you.
- Data-processing terms are between you and the AI provider. We help you pick a provider whose terms and data location fit your needs.
Screens
Questions about AI chatbots
Does the chatbot make things up?
Language models can give wrong answers, so we limit what it may say. It answers only from your sources, links to the page it used, says when it does not know and offers a person instead.
What counts as a knowledge source or an integration?
One place the chatbot reads from or sends to: your website, a help center, a folder of PDFs, a CRM or a help desk. The base price includes two. Email hand-off is always included.
Which AI model does the chatbot use?
We choose one with you in discovery, from providers such as OpenAI, Anthropic or Google, based on answer quality, cost and where your data is processed. The account is in your name.
Is visitor data safe with an AI chatbot?
We send the model only what it needs to answer, keep conversation logs in your accounts and help you review the provider's data-processing terms. We do not use your data to train models.
Further reading
Designing for AI search: getting cited
AI search optimization for company websites: how AI answers pick sources, why server HTML matters, and how to write pages that get cited and clicked.