How to build a website in 2026

Step 1: Decide whether you are building a website or improving one

A new business without a website needs infrastructure: domain, hosting, content system, templates, analytics, security, and perhaps commerce. An established business already has those things. Its problem may be presentation, consistency, conversion, performance, or age, not absence.

This distinction can save months. A redesign can keep the domain, pages, products, platform, analytics, and search history while replacing the customer-facing experience. A rebuild recreates the system and then migrates accumulated value into it.

Google does not say that all migrations destroy rankings. It does warn that significant moves can cause temporary fluctuations and require careful redirects, canonical tags, internal-link updates, sitemaps, and monitoring. If the current foundation works, do not create a migration just to obtain a new color palette.

Step 2: Write the smallest brief that can prevent the wrong website

Before opening a builder or prompting a model, answer six questions:

1. Who must act? Name the primary customer and action.
2. What must the site do? Inform, collect leads, schedule, sell, support, or operate an application.
3. What content exists? Copy, products, images, video, reviews, policies, and proof.
4. What must connect? Payments, inventory, CRM, email, analytics, booking, tax, shipping, or accounts.
5. Who maintains it? Founder, marketer, merchant, designer, or developer.
6. What cannot break? URLs, rankings, checkout, accessibility, performance, or native editing.

A good brief describes constraints and evidence. “Make it modern” is a preference. “Reduce mobile clutter on our top product template without changing checkout or published URLs” is work a person or agent can verify.

Option 1: A template is the fastest route when your needs are conventional

Templates solve hundreds of decisions before you arrive: grids, breakpoints, type hierarchy, menus, forms, galleries, and content patterns. That is a feature. Most local services, portfolios, simple publications, event sites, and early-stage businesses do not need original interface architecture.

Choose a template by structure, not demo photography. Replace its sample content with your real longest headline, weakest image, most complicated service, and mobile navigation before committing. A template that survives real content is more valuable than one that looks beautiful only in its preview.

Option 2: Squarespace and Wix make hosting invisible but keep design decisions visible

Squarespace’s annual plans begin at $16 per month and include managed hosting, templates, drag-and-drop editing, SEO tools, and support. Wix offers a free plan plus paid tiers that vary by region, with hosting, an AI builder, thousands of templates, and business tools.

These platforms are appropriate when one team wants one vendor and predictable operations. The limitation is the other side of convenience: the platform defines the ceiling for unusual behavior. The risk is editor freedom. A merchant can change anything without understanding the system that made the original template coherent.

Set rules before editing: one primary button, one type scale, two section widths, a fixed spacing rhythm, consistent image ratios, and named reusable blocks. A builder becomes safer when every page does not become a new design project.

Option 3: Framer is the fastest route to a polished, expressive marketing site

Framer is especially good for startup launches, portfolios, campaigns, editorial landing pages, and brand-forward marketing. Its annual Basic plan starts at $10 per month and Pro at $30, before extra editor, localization, or usage costs.

It offers high visual quality with a shorter learning curve than traditional front-end development. It is less natural as the complete operating system for a complex store. Use it when storytelling and velocity dominate, and confirm where forms, content, localization, analytics, and future ownership will live.

Option 4: Webflow gives visual builders access to front-end concepts

Webflow provides responsive layout, classes, components, interactions, and structured CMS content through a visual environment. Annual site plans now begin around $15 per month after its May 2026 plan changes.

The power comes with vocabulary. To use Webflow well, someone still needs to understand boxes, flow, flex, grid, positioning, class reuse, responsive behavior, and semantic structure. It removes the need to type every CSS declaration; it does not remove the underlying model of the web.

Choose it for a design-led marketing team willing to maintain a system. Avoid it when the goal is for the merchant never to think about layout mechanics.

Option 5: Shopify is the safest starting point when products, orders, and checkout are central

Shopify begins at $29 per month on annual billing, with higher tiers for growing operational needs. It provides catalog, inventory, checkout, payment, order, theme, and app infrastructure. That reduces the number of critical systems a small merchant must assemble.

Start with a free or proven theme. Configure products, navigation, policies, shipping, payments, tax, analytics, and domain. Add apps only when a measured need justifies another dependency. If the store already operates, duplicate the theme, redesign the unpublished copy, preview it, and publish only after verification.

Option 6: ChatGPT and Claude can build a custom site for less money than any human alternative

In direct software cost, general AI assistants are hard to beat. ChatGPT and Claude both offer free plans and $20 consumer plans; power-user tiers reach $200 per month. They can generate semantic HTML, CSS, JavaScript, React components, metadata, structured data, tests, and deployment files.

They are also better than many drag-and-drop builders when you need a specific responsive behavior, unusual art direction, clean custom code, or reusable logic. You can describe the outcome instead of manually placing each box.

But chat-based creation requires active refinement. The model does not automatically own your business context or quality bar. Expect to review generated claims, links, mobile layouts, accessibility, security, analytics, forms, and dependencies. The lower invoice is partly exchanged for more founder attention.

AI code still needs hosting, a domain, deployment, and an owner

A website on your computer is not a website customers can visit. A simple AI-generated static site can be deployed to Cloudflare Pages, GitHub Pages, Netlify, or Vercel. Cloudflare Pages includes a free tier with 500 builds per month. Vercel Hobby is free for personal and non-commercial projects; Vercel Pro is $20 monthly for professional work.

A practical AI-code launch looks like this:

1. Put source code in a Git repository.
2. Connect the repository to a hosting provider.
3. Set a production build command and output folder.
4. Connect a custom domain and verify HTTPS.
5. Configure forms, email, analytics, and environment secrets.
6. Test desktop, mobile, keyboard navigation, performance, metadata, social previews, and broken links.
7. Document how to update, deploy, roll back, and restore the site.

If those steps are unfamiliar, the site is not “free.” It is borrowing future technical work from someone.

Option 7: A freelancer buys focused judgement without agency overhead

A strong freelancer can translate a clear brief into brand, UX, copy, and implementation, or specialize in one part while coordinating with others. Expect roughly $1,500 for small template-led projects and several thousand to $15,000+ for custom sites, ecommerce, or migrations.

Ask to see live work, not only static portfolio frames. Confirm responsive behavior, performance, accessibility, SEO migration, revisions, analytics, source ownership, native editing, documentation, and support. The handoff is part of the product.

Option 8: An agency is justified when the hard problem is multidisciplinary coordination

Agencies become valuable when research, strategy, identity, copy, information architecture, UX, engineering, SEO, and stakeholder management must move together. Clutch’s 2026 guide puts the broad market at about $2,000 to $100,000, with most reviewed design projects under $10,000 and larger custom work lifting the average.

Do not hire an agency merely because you are afraid of tools. Hire one because the business problem needs its mix of disciplines, process, and accountability.

Step 3: Build content and structure before polishing animation

Define the path from arrival to action. A basic business website usually needs a homepage, offer or product pages, proof, about, contact, policies, and useful support content. Ecommerce adds collections, product details, cart, search, account, shipping, returns, and post-purchase communication.

Write headings that carry the argument. Use subheadings people can scan. Show evidence close to claims. Put price, shipping, returns, compatibility, and risk-reversal information near the decision rather than burying it in a FAQ.

Step 4: Make the real website work before making the demo impressive

Test with actual products, long names, sold-out states, discounts, weak photography, slow connections, keyboard navigation, reduced motion, and small screens. Baymard’s 2026 research found that 52% of desktop, 62% of mobile, and 64% of app ecommerce product pages had “mediocre or worse” UX. Attractive design alone does not resolve missing information or difficult interaction.

Use motion to explain state, not to delay access. Compress images. Avoid unnecessary third-party scripts. Respect the user’s reduced-motion preference. Make focus states visible. Check tap targets, contrast, form labels, errors, zoom, and screen-reader structure.

Step 5: Launch with evidence and a rollback plan

Before launch, crawl links, verify analytics, test forms and checkout, inspect metadata, capture important pages at desktop and mobile, and benchmark performance. For a migration, map every valuable old URL to the closest equivalent new URL and keep redirects long enough for users and search engines to adapt.

For an existing Shopify site, use an unpublished theme as the release candidate. That creates a real preview without turning customers into testers. The merchant should publish only after the new version is visually and functionally acceptable.

Step 6: Treat launch as the first measured version, not the finished website

Watch search coverage, traffic, conversion, form completion, product engagement, checkout behavior, performance, and customer questions. Fix regressions first. Then prioritize improvements by expected customer impact and evidence.

A website becomes outdated because the market, brand, content, products, devices, and customer expectations move. The sustainable answer is not a fresh demolition every few years. It is a system that can observe, improve, test, and remain coherent.