Tecof • September 9, 2026
How to Build a Website With AI From Scratch
In Brief
Building a website with AI starts when you describe your business in a few paragraphs and get back an editable draft site. For most small businesses a publishable first version is one afternoon of work. The hard part is not generating the draft. It is everything after: rewriting the copy in your own voice, measuring mobile speed, adding KVKK and distance selling texts, wiring up payments and shipping. You do not need code. You need the discipline to make decisions.
Picture a two-person bakery in Ankara that makes custom celebration cakes. Three hundred Instagram followers, a camera roll full of unsorted product shots, orders arriving as WhatsApp messages. No website. The agency quote came back with a six-week timeline and half the fee due up front. The same brief typed into an AI site builder produces a first draft on screen in five minutes.
Speed is not the only gain here. Seeing a draft in five minutes also means seeing what is wrong with it in five minutes. If the menu order is strange, if the hero image does not show the product, if the pricing block is confusing, you catch it now instead of at a revision meeting six weeks out. The expensive line item in the old agency process was never the design work. It was the feedback loop.
The order below is the one that works when you are starting from nothing: brief first, then structure, then content, then technical measurement and legal pages last. People who scramble that order end up with a page that looks fine and sells nothing.
What actually happens in the first forty-five minutes
Most AI site builders move you through three screens. Each one asks something different from you, and knowing where to spend your time is half the job.
The brief screen: five minutes here saves forty later
The opening box asks what you want to make. The three or four sentences you type there feed every decision that follows: what the business is, what it sells, who it talks to, and what the site should ask the visitor to do. An appointment, a cart, a phone call. Rushing this box means spending the next forty minutes undoing its consequences.
The structure screen: page list and navigation
After reading the brief, the model proposes a page list. For a small service business the typical set is home, services, about, testimonials, contact. An online store adds category, product detail, cart and checkout. Your only job here is deletion. An eight-item menu almost always performs worse than a four-item menu, because the visitor cannot tell where to click.
The correction pass: put your own product in
Product names, prices and photos in the first version are placeholders. Do not judge anything until your real data is in there. Every layout looks good filled with placeholder text. Load your twelve actual products, your actual prices and the photos you shot on your phone, and the weak spots surface immediately. Long product names push headings onto a second line, vertical photos break the grid, a three-digit price next to a five-digit price throws the alignment off.
Those three screens add up to forty-five minutes if you arrive prepared, and to two days if you do not, because you end up gathering product copy and photos mid-process. Before you start, drop your logo, fifteen product photos, your price list and two customer reviews into one folder. Everything after that moves quickly.
Prompting: the gap between a weak brief and a working one
The quality of your instruction sets the ceiling on the quality of the site. Weak briefs share one habit: piling up adjectives. Modern, clean, professional, user friendly. Those words tell the model nothing, because they rule no design out. A working brief supplies constraints instead. Constraints narrow the model's choices and produce something that belongs to you.
| Brief element | Weak version | Working version |
|---|---|---|
| Business | I want a bakery site | Two-person studio in Ankara making celebration cakes, orders taken at least three days ahead |
| Audience | Everyone | Parents shopping for a child's birthday cake who want to see prices before contacting anyone |
| Job of the site | Promote the business | Two clicks from the homepage to an order form that asks for date and serving count |
| Visual language | Modern and elegant | Product-led photography, off-white background, deep burgundy accent, no thin typefaces |
| Constraints | Not stated | No inventory tracking, no on-site payment, deposits handled over WhatsApp |
Give instructions in pieces
Do not compress fifteen requests into one sentence. Models drop the middle items of long lists. Get the structure standing first, then correct one thing at a time: "remove the second section on the homepage", "put the pricing cards in three columns", "add a delivery date field to the contact form". Check the result with your eyes after each instruction. This is considerably faster than trying to write one perfect prompt.
Learn to say no
Sometimes the model proposes a layout that does not fit your trade. A full-bleed hero image suits a property listing site and buries the phone number on an appliance repair site. Delete the section without hesitating. The model is a suggestion engine and you are the approver. The only party with a real opinion about what the site should do is the one talking to customers every day.
Start from a template or generate from scratch
Both routes are legitimate, and the choice mostly depends on how much content you already have. With copy and photography ready, a template is faster. With nothing in hand, generation leaves fewer empty boxes staring at you.
| Criterion | Ready-made template | AI generation |
|---|---|---|
| Time to first version | Structure exists, filling it is on you | Structure and draft content arrive together |
| Distinctiveness | Other sites run the same template | Layout and palette shift with the brief |
| Predictability | You see what you are getting up front | First output can surprise you and needs edits |
| Best fit | Business with copy and images ready | Business starting cold that also needs content |
| Ongoing changes | You stay inside the template's limits | New sections added by instruction |
The hybrid route wins in practice
Most businesses end up combining the two. They take a template close to their sector as scaffolding, then rework the copy and section order with AI. The template hands you a page hierarchy that already works, and the model fills that hierarchy with your specifics.
Platform choice matters at exactly this point. However good the builder is, if the platform behind it is weak on inventory, orders and invoicing, you will be migrating in six months. Migration costs more than the original build, because your URL structure, your search rankings and your customer data all go on the line. Reading about the costly mistakes people make choosing a platform up front removes most of that risk.
Copy and imagery: turning machine output into your own voice
The first pass of AI copy is grammatically clean and commercially empty. "We put customer satisfaction first with our quality-driven service approach" appears on tens of thousands of sites. It claims nothing, so it sells nothing.
Harden the copy
- Swap adjectives for numbers. Instead of "fast delivery", write "orders within Ankara delivered by noon the next day".
- Pair every claim with proof. Instead of "experienced team", write "over 1,400 celebration orders since 2019".
- Add one sentence only you could write. Two lines about how the studio started break the plastic tone of model output instantly.
- Put measurements in product descriptions. Serving size, weight, allergen notes, lead time. These speed up the buying decision and happen to match long-tail search queries.
Where generated imagery stops working
Generative image tools earn their place on backgrounds, textures and conceptual visuals. They do not work for the physical product you sell. A customer wants a photo of the cake that will arrive at their door, not the cake the model imagined. Shoot the product on your phone, near a window, against a plain white backdrop. Use the model for the work around that photo: cleaning up the background, normalising dimensions, generating category cover art.
File size deserves attention too. A photo straight off a phone runs 4 to 6 MB, and dropping it onto a page untouched leaves mobile visitors waiting. Check whether your platform compresses to modern formats automatically. If it does not, even a twelve-item catalogue can double its mobile load time. After copy and images comes measurement, because until you measure, you only know your site is fast on your own machine.
Speed, mobile and a measurable technical baseline
Most e-commerce traffic in Turkey arrives on mobile, and a large share of it browses on cellular connections that are far from ideal. A page that loads in a second in your desktop browser can take five for someone on the metro. Speed is a subject for measurement, not for assumptions.
Google's published Core Web Vitals thresholds settle what counts as good. AI-built sites usually start clean on these numbers because there is no pile of plugins underneath. They degrade later, as large photos, embedded video, third-party chat widgets and ad pixels get added one by one.
| Metric | Threshold considered good | Most common cause of failure |
|---|---|---|
| LCP, largest contentful paint | Under 2.5 seconds | Uncompressed hero image |
| INP, interaction to next paint | Under 200 ms | Third-party chat and popup scripts |
| CLS, cumulative layout shift | Under 0.1 | Images and ad slots without declared dimensions |
Mobile checklist before you go live
- Open the site on your own phone with wifi switched off. That single test reveals more than most desktop checks.
- Look at what sits above the fold on the homepage. Is the phone number or order button there.
- Tap the form fields with your thumb. Small tap targets produce mis-taps and abandoned forms.
- Run one real order through checkout yourself, with a real card rather than a test card.
Speed connects straight to revenue, and that connection shows up in your ad account. When you pay per click for a slow page, part of that spend goes to people who leave before the page renders. Compressing the hero image is the cheapest cost-per-click reduction available to you.
Getting found in search engines and AI assistants
A finished site does not attract traffic on its own. You now need visibility in two separate channels: classic search results and assistants that answer directly. Each wants something different from you.
The classic search side
The fundamentals have not changed. One H1 per page, an intact heading hierarchy, meta descriptions that state the page's promise, alt text on images. Submit your sitemap to Google Search Console and check indexing a week later. AI can fill these fields for you, but read what it filled in before you accept it. Models sometimes write the identical meta description across five pages.
Visibility inside AI search
Assistants reading your page do not see the design, they see the structure. Clear question headings, short direct answers, tables and definitions are what pay off on this side. An FAQ section gets quoted far more often than a decorative hero. Write price, scope and delivery terms as text on your product and service pages rather than burying them inside an image.
If you want both channels planned together, it helps to treat what AEO and GEO actually mean as separate disciplines, because the same content has to produce two different outputs. Classic search rewards depth. An assistant rewards clarity.
Expecting meaningful organic traffic in a new site's first three months is not realistic. During that stretch most visitors come from social and paid. You feed the organic side with writing: two posts a month that answer a question a real customer asked, and six months later part of your ad budget frees up.
A realistic budget, with a worked example in lira
The figures move with your provider and the scope of the work, but the list of line items does not. The table below shows what a one-person service business should plan for in year one. Treat the entries as planning categories rather than a quote.
| Line item | Agency route | AI-assisted platform |
|---|---|---|
| Design and build | Large one-off fee, revisions billed | Inside the subscription, revisions done by you |
| Domain name | Annual, billed separately | Annual, billed separately |
| Hosting and SSL | Monthly, rises with traffic | Usually inside the subscription |
| Plugin and theme licences | Several renewing annual items | None |
| Maintenance and updates | Monthly retainer or hourly rate | Handled by the platform |
| Copy and image production | Copywriter and photographer fees | Model drafts it, you correct it |
| Payment processing | Per transaction, provider dependent | Per transaction, provider dependent |
The invisible line item is your own time
No budget table ever lists the owner's hours. On an AI-built site the build hours drop and the maintenance hours approach zero, but the content hours stay. Shooting product photos, updating prices, adding a customer question to the FAQ are all still your work. Set aside four hours a month and put those four hours in your calendar. Everyone who skips the calendar entry has a stale site three months later.
Plan year two while you are at it. Some routes that look cheap in the first year jump sharply on hosting and plugin renewals once traffic grows. Ask where the pricing thresholds sit before you commit. Seeing them laid out in advance on something like Tecof's plans and pricing page beats discovering them on an invoice.
Legal and operational steps you cannot skip in Turkey
Most guides written abroad leave this section out, because they do not know what Turkish law asks for. Several pieces have to be in place before the site goes live, and some of them carry administrative fines.
KVKK and the privacy notice
The moment you add a contact form, a newsletter box or an order form, you are processing personal data. You need a privacy notice, an explicit consent checkbox and a privacy policy page. The consent box must not be pre-ticked, and marketing permission has to be asked separately from the processing required to fulfil an order. Publications from Turkey's Personal Data Protection Authority explain which text belongs where.
İYS registration
If you plan to send commercial electronic messages, meaning email or SMS campaigns, registering with İYS and uploading your permission records is mandatory. Do not send anything before the consents you collected through your form are loaded into the system. A permission list takes months to build and one careless send to waste.
Distance selling and returns
If you sell, the distance selling agreement, the pre-information form, return and exchange conditions and delivery timelines all belong on the site. Do not publish AI-drafted legal text as it stands. Have it checked against current legislation, because models will invent dates and article numbers in regulatory copy.
Payments and shipping
- Local payment providers handle instalments and 3D Secure more smoothly than foreign ones. In Turkey the number of instalments offered has a direct effect on conversion.
- Connect shipping during setup. Automatic label generation with carriers such as Yurtiçi, Aras or MNG beats maintaining a tracking spreadsheet by hand.
- If you also sell on marketplaces like Trendyol or Hepsiburada, set up stock synchronisation on day one. Selling the same unit twice across channels damages your store rating.
- Plan what your order and return flow looks like in the warehouse. A manual process that copes with ten orders a day jams at a hundred.
The thirty days after launch
Publishing does not end the work, it starts the measurement. Month one is when behaviour talks and design stops talking. You do not have much data yet, but you do have small signals pointing in a direction.
Week one
Verify your analytics install and confirm data is actually arriving by visiting the site yourself. Submit the sitemap in Search Console. Ask five people you know to open the site on their phones and screen-record it. Watch where they hesitate rather than asking them what they thought. Hesitation is the most honest feedback a design ever gets.
Weeks two and three
Real traffic starts arriving. Three numbers matter: which page they landed on, how many seconds they stayed, which page they left from. Exits concentrated on the homepage mean your message is not clear. Exits on product detail mean a question about price, shipping cost or lead time is going unanswered. Those two problems need different fixes.
Week four
Run a small first advertising test. The goal is not sales, it is learning which message gets clicked. Build two sets with two headlines and change exactly one variable. Same image, different headline. Structuring that test properly once makes every campaign afterwards cheaper.
By day thirty you should hold two lists: questions your site answers, and questions it does not. Every item on the second list is next month's work. What you can do tomorrow morning is simpler than that: open your phone, go to your own site as a customer would, try to place an order, and write down every point where you get stuck. If that note has more than three items on it, fix those first.
Frequently Asked Questions
Do I need to know how to code to build a website with AI?
No. Describing what you want in plain language is enough. Knowing HTML or CSS makes some edits easier but is not required. The skill that actually matters is describing your business and your customer in clear sentences. On the technical side, your only job is reviewing what the model produced.
How many hours does a publishable site really take?
Four to six hours is realistic if your content is ready. That covers writing the brief, correction passes, adding legal pages and a mobile check. Gathering content mid-process stretches it to two days. Preparing photos and a price list in advance is the single biggest accelerator.
Will Google penalise AI-written copy?
Search engines evaluate whether content helps the reader, not how it was produced. The problem is never AI itself, it is publishing generic text that carries no information. Copy that includes your prices, your delivery terms and your own experience does its job regardless of who typed it.
Can I rebuild my existing site with AI?
Yes, but protect the URL structure. If addresses change you have to set 301 redirects from the old pages to the new ones, or you lose your rankings. Before starting, write down your current page list and the ten addresses that receive the most traffic.
Does an online store get built the same way?
The core flow is identical, with inventory, variants, shipping and payment layered on top. As product counts grow, category structure and SKU discipline start to decide how usable the store is. We covered the store-specific steps separately in building an AI-powered e-commerce store.
Can I use AI-generated images commercially?
That depends on the licence terms of the tool you used. Most allow commercial use, some restrict it on free plans. Do not use generated imagery for the physical product you sell, because it misleads the buyer and pushes your return rate up.
Can I change the design myself after launch?
On modern platforms, yes. Adding sections, changing colours and editing text happen in a visual editor or through written instructions. On Tecof those edits happen inside Studio, and changes can be previewed before they go live. You never open a code file.
How many pages should I start with?
Four to six covers most small businesses. Home, a services or products page, about, contact, plus your legal pages. People who launch with twenty pages typically never update fifteen of them. Add pages as traffic and real questions arrive.
Can I use my own domain name?
Yes. If you already own one, you point its DNS records at the new platform, and the switch usually completes within a few hours. If you do not own one, you can buy it during setup. Pick the extension closest to your brand name and avoid hyphens.
Should I build a mobile app as well?
Not at the start. A mobile-friendly site does most of what an app would do early on, without app store review cycles. Apps start making sense for businesses with repeat, habitual ordering. We worked through where that threshold sits in mobile app versus mobile site.