Many craftsmen, practices, clubs and solo businesses have no website or an outdated one, because design, copywriting, photography and legal pages each need a specialist. AI website builders turn a five-minute interview into a hosted, mobile-ready site with structure, text, images, forms and SEO basics, and let the owner change it afterwards by chat. The owner stays responsible for facts, legal pages and accessibility.
Problem & value
- Today: owners use an agency (expensive, slow), a template builder (blank-page problem: what to write, which sections), or a family member. Sites stay thin, stale and legally incomplete.
- AI value: removes the blank page. The builder proposes sitemap, section order and first-draft copy from a few answers, picks a style and imagery, wires a contact or booking form and fills title tags and descriptions. Edits happen in natural language ("add a prices section", "make it friendlier").
- Evidence: all large builders now ship this as a headline feature (Wix, Jimdo, IONOS, Squarespace, Hostinger) [1][2][3][4]. Independent controlled studies of time or cost savings are scarce; vendor claims such as "site in minutes" describe the first draft, not a finished, fact-checked site.
- Not the same as: CODE-007 (custom application logic, generated code) or GEN-004 (static graphics).
Variants
- Local-business site: craftsmen, restaurants, medical and therapy practices, tax advisers; map, opening hours, services, reviews, local SEO.
- Campaign landing page: single page with one call to action, form and tracking; many variants for A/B tests.
- Private & association use: wedding, club or school pages; password protection matters more than SEO.
- Multilingual site: one source language plus machine-translated versions with human review (LANG-009).
How it works
- brief / interview (+ optional import of existing site, Google Business profile, social page)
- sitemap + section planning (home, services, about, contact, FAQ)
- template/theme selection (component library: hero, features, gallery, form, map)
- copy generation per section (tone, language, local keywords)
- image sourcing (stock search, text-to-image, user uploads) + alt text
- forms, booking and shop widgets, analytics, SEO metadata + schema.org
- legal scaffolding (Impressum, privacy notice, cookie consent)
- owner review and edits (chat or visual editor)
- publish on hosting + domain
- Layout is composed from a vendor's template and component system, not free-form code, which keeps output responsive, accessible by default to a degree, and editable by non-experts.
- LLM handles planning, copy and conversational edits (typically via tool calls that modify a page JSON/component tree). Image models generate or adapt hero and section pictures.
- Human-in-the-loop: the owner must verify every factual statement (prices, certifications, medical or legal claims), supply the legal data and approve publication. Legal text is a template, not legal advice.
Data preparation
The "data" here is mostly company facts and brand material, and it is usually scattered: services and price lists live in Word or Excel files and the ERP, opening hours and address in the Google Business profile, photos on phones and in mail attachments, logos in old print PDFs, and legal identifiers (register number, VAT ID, chamber, professional regulator) in the accounting or contract folders. Typical gaps are outdated prices, low-resolution or unlicensed photos, no alt texts, missing consent for photos of staff or customers, and no single person who owns the facts.
Concrete steps: 1. Build one "fact sheet" (a shared document) with services, prices, hours, address, contact, legal identifiers and tone of voice; have the process owner sign it off. 2. Collect real photos and the logo in original quality, check image rights and written consent of people shown, and name files meaningfully. 3. List the third-party tools the site will load (maps, fonts, video, analytics, booking) for the privacy notice. 4. If an old site exists, export its text and sitemap for import; delete outdated pages first. 5. For many price lists or product data, clean the spreadsheet first (see PREP-011); for terminology and brand voice see PREP-015 and PREP-016.
Effort share: low, roughly 10-20 percent of the total, but decisive for quality, since the AI only writes well from correct facts. Tools needed are an office suite and a shared folder; no data pipeline.
Implementation paths
Which path for whom
- SMB: A. Mature, EU-hostable builders (IONOS, Jimdo, Wix) cover a standard content site at small monthly prices without technical staff.
- Private: A. Club, family or event pages need a hosted builder with a free or cheap tier, not their own infrastructure.
- Developer: B. Agencies and freelancers deliver more value with an LLM generating structured page JSON for a component system or headless CMS they control.
Path A — Buy (recommended for SMBs and private users)
- Platform builders: Wix (AI Website Builder, chat-based editing), Squarespace (Blueprint AI), Webflow (AI-assisted site building for designers), Hostinger (AI website builder, Horizons for app-like sites), Shopify (AI store and theme generation for shops).
- EU-based: IONOS MyWebsite (AI builder, Germany), Jimdo (AI-assisted builder, Germany), STRATO, Webnode (Czechia), Hostinger (Lithuania). Check EU data centre, DPA, and whether the Impressum and cookie tooling are included.
- Choose A when: content site with fewer than ~20 pages, no custom logic, owner wants no tech involvement. Watch lock-in: most builders cannot export a portable site.
Path B — Build on cloud APIs
- Pipeline: LLM (Claude, GPT, Gemini or Mistral models via EU regions) generates structured page JSON against a fixed component schema; an image API supplies visuals; a static generator (Astro, Hugo, Next.js) or headless CMS (Storyblok, Contentful, Strapi Cloud) renders it.
- Developer-facing tools: v0 (Vercel), Lovable (Sweden), Bolt, Framer AI, and Builder.io produce code and sit closer to CODE-007; useful for agencies who deliver a code-based site.
- Compliance add-ons: consent tools (e.g. Usercentrics, Cookiebot) and cookie-less analytics (Plausible, Matomo).
Path C — Self-host (open source)
- Models: open-weight LLMs such as Qwen3-Coder or Qwen3, Mistral and Llama-family models for structured page JSON and copy; FLUX.1 [schnell/dev] or Stable Diffusion models for images (check licences).
- Stack: Astro/Hugo/Eleventy as renderer, Ghost, WordPress or Strapi/Directus as CMS, Open WebUI or a small custom chat front end for edits; vLLM or Ollama for serving.
- Hardware: HW2 handles a 30B-class quantised model plus an image model for an agency's volume; HW1 is enough for text-only generation with small models.
Path D — Custom training
Rarely worthwhile; prompt templates with brand examples and a locked component set suffice.
State of the art (2026)
- Conversational site building is standard: builders interview the owner, generate a full multi-page site, and accept edits in chat. The boundary with app builders (CODE-007) is blurring, for example when a vendor merges its classic builder with an AI app-builder mode [1].
- Structured generation on component systems gives more reliable output than raw HTML generation: fewer layout bugs, consistent mobile behaviour.
- Agentic extras: builders add blog post generation, product descriptions, image generation, and automated SEO suggestions. Visibility is also shifting to AI answer engines, so schema.org markup and clear factual pages matter next to classic SEO.
- Limits and failure modes: generic or repetitive copy; invented facts (opening hours, awards, certifications, testimonials); stock-like or inaccurate AI images of staff, premises or products; weak local SEO unless address data is correct; inconsistent translations; accessibility gaps in generated sites; and legal-page templates that do not match the real data processing (analytics, embedded maps, fonts, video).
Shared resources & combinations
- Same stack (llm-general + image-gen): GEN-001 marketing imagery, GEN-004 flyers and banners, GEN-006 logo and brand assets feed the same brand kit; LANG-005 SEO content and LANG-009 translation extend it.
- Downstream: CONV-001 website chatbot, CONV-005 lead qualification; forms and bookings connect to CRM through workflow automation.
Effort, cost & KPIs
- Path A: hours to a first draft, a few days to a reviewed site; subscription pricing per month plus domain. Compare current prices on vendor pages, as reviews disagree.
- Path C: 1-3 months incl. design system; GPU server and operations.
- KPIs: time to publish, share of AI text kept after review, organic search impressions and clicks, form conversion rate, Core Web Vitals, accessibility audit score, number of factual corrections found in review.
People & change
Involve management at the start (budget, brand, who is responsible for the site) and the department lead for marketing or sales as process owner, who supplies and approves the facts. End users are the one or two people who edit the site afterwards; let them try the builder in the pilot. The DPO (or an external one) reviews forms, bookings, maps, fonts, analytics and the AVV with the builder (Art. 28 GDPR); a DPIA is normally not needed for a standard site. Legal/compliance checks the Impressum and claim rules in regulated sectors. If customers can book or write via the site, test the flow with a few real customers.
Works council: normally not needed, since the tool does not monitor employee behaviour or performance (§87(1) no. 6 BetrVG); it becomes relevant only if staff profiles or photos are published or editor activity is logged for evaluation. Where one exists, inform it in good time anyway (§90(1) no. 3 BetrVG) and for public bodies involve the Personalrat.
AI literacy: support staff who use the builder (Art. 4 AI Act) with a short briefing on invented facts, image rights and labelling of AI images. Typical concerns are "will it replace the agency or copywriter" and "will the site sound generic"; answer by positioning AI as a first draft with a human sign-off before publication. See GUIDE-006, GUIDE-007 and GUIDE-005.
Risks, regulation & compliance
- EU AI Act: minimal risk for the generator. Art. 4 AI literacy applies to the operator. Art. 50 transparency duties concern deepfakes and AI-generated text published to inform the public on matters of public interest, and apply mainly to providers and deployers of such content; deepfake-style images of real people need labelling; independent of that, avoid fake staff or customer photos as misleading [5].
- Impressum / provider identification: German sites need a complete Impressum under section 5 DDG; AI cannot invent data such as register numbers, chamber or professional-regulator details, and the VAT ID [6].
- GDPR and ePrivacy: contact forms, bookings, embedded maps, fonts, video and analytics process personal data; the privacy notice must name each tool and its legal basis (Art. 13 GDPR) and a signed DPA is needed with the builder. Consent is required for non-essential cookies and similar storage (section 25 TDDDG in Germany) [7][8]. A template banner that loads trackers before consent is non-compliant.
- Accessibility: from 28 June 2025 the Accessibility Act (BFSG in Germany) covers online shops and services to consumers; micro-enterprises providing services are exempt. Many other sites are good practice per WCAG 2.1/2.2 [9].
- Content liability: invented claims can be misleading advertising (UWG); regulated sectors (health advertising law, lawyers, tax advisers) have additional claim rules. Copyright: check image licences and model terms.
- Mitigations: fact checklist before publishing, cookie-less analytics, consent tool that blocks scripts, review of legal pages by a lawyer or a certified legal-text service, alt texts, backups and exportable content.
Getting started
- Collect the facts (services, prices, hours, address, legal identifiers, real photos) before opening a builder.
- Try two builders from Path A with the same brief, one EU-hosted; compare copy quality and editability.
- Replace generic text and images with real ones; verify every claim; add the Impressum data.
- Configure consent, privacy notice and processors; test the site without cookies accepted.
- Submit sitemap to search engines, check Google Business profile, then set a quarterly review.
Sources
- Wix — AI Website Builder — https://www.wix.com/ai-website-builder
- Jimdo — AI website builder — https://www.jimdo.com/
- IONOS — AI website builder (MyWebsite) — https://www.ionos.com/websites/website-builder
- Hostinger — AI website builder — https://www.hostinger.com/ai-website-builder
- Regulation (EU) 2024/1689 (AI Act), Art. 4 and 50 — https://eur-lex.europa.eu/eli/reg/2024/1689/oj
- Digitale-Dienste-Gesetz (DDG), section 5 — https://www.gesetze-im-internet.de/ddg/__5.html
- Regulation (EU) 2016/679 (GDPR) — https://eur-lex.europa.eu/eli/reg/2016/679/oj
- TDDDG, section 25 — https://www.gesetze-im-internet.de/ttdsg/__25.html
- Directive (EU) 2019/882 (European Accessibility Act) — https://eur-lex.europa.eu/eli/dir/2019/882/oj