AI in the workflow: what it does and doesn't do for your site

When we say Devebo is AI‑native, we don't mean a program builds your site on its own. We mean we research, prototype and iterate with AI, so more of your budget goes on design decisions and less on routine work.

Routine work is where AI earns its keep:

• Copy drafts. From rough notes or a product list, you quickly get a first page structure and a first version of the descriptions. A starting point, not finished copy.
• Translation as a first pass. On a bilingual site, machine translation saves hours of typing. The right wording still comes from someone who writes that language well.
• Code scaffolding. Repetitive components, page templates, redirect lists from the old site to the new one. They get written faster and checked before they go live.
• Image variations. Backgrounds, crops for different screens, mock-ups so you can see an idea before you decide. They don't replace real photos of your products.
• Audits. AI can go through every page and flag missing titles, images without descriptions and broken links. It hands over a list; a person sets the priorities.
• Structured data. From details you already have, such as your address, opening hours and products, it quickly drafts the code that describes them to search engines.

What these share: AI produces the first version. Someone has to check it before your customers see it.

AI sounds just as confident when it's wrong as when it's right. That's why its mistakes don't show at first glance, and why customers often spot them before you do.

• Generic copy. "Quality, reliability, years of experience." Phrases that fit every business and so set none apart. Visitors leave without learning what you do differently.
• Factual errors. Opening hours, phone numbers, sizes, pack weights. When the tool doesn't know something, it often fills the gap with something that merely sounds right.
• Wrong Greek. Informal and formal "you" mixed on the same page, sentences that read like a translation from English, terms nobody in your market uses. Greek readers sense it at once, even if they can't say why. Stilted English on the English pages does the same damage.
• Claims you never made. "Award-winning", "number one in Thrace", guarantees and certifications that don't exist. In sectors such as health or food, that can become a legal problem, not just an embarrassment.
• Accessibility. Image descriptions like "person smiling", form fields without labels, low-contrast text. The page works, just not for everyone.
• Brand voice. A neighbourhood bakery and an accountancy firm end up sounding the same. The character you built over years disappears in a single paragraph.

A language model predicts the most likely continuation of a text. It doesn't know your business, hasn't spoken to your customers and won't pick up the phone when someone asks about something it got wrong.

Four things can't be handed to a tool:

• Strategy. Who you want to reach, what they're looking for, and the one goal of each page: a call, a booking or an order. That comes from a conversation with you, not from a prompt.
• Facts. Only you know what's true: what you send by courier, what's collection only, which days you're closed. Every detail on the site should have passed through your hands.
• Tone. If customers know you from phone calls and Viber messages, the site should sound like you, not like a brochure.
• Final review. Someone reads every page in full, on a phone, before it goes live, and takes responsibility for it.

That's why at Devebo you speak with the person who designs and builds your project, not with a tool. The prototypes you open on your phone before anything is built exist for exactly this: so your knowledge can correct what no tool could have known.

Some searching now happens as a question: someone asks ChatGPT or Perplexity, or reads the AI Overview at the top of Google. These tools compose an answer from pages they can read, and often show a link to the source alongside it. To be that source, three things matter:

• Clear facts. Business name, address, phone, opening hours, service area: written as text, not inside an image, and identical on your site, your Google Business Profile and directories. When sources disagree, the engine can't tell which one to trust.
• Structured data. Schema.org code that states plainly "this is the business, these are the hours, this is the product". AI helps draft it, as long as someone confirms every value is correct.
• Answer first. Headings that mirror your customers' questions, and a first sentence that answers directly. "Do you deliver across Greece?" "Yes, by courier." The details follow.

The foundation doesn't change: a page that loads quickly and is clearly structured is easier to read for people and machines alike. That's why speed is the first step of SEO here too.

Nobody can guarantee that an AI answer will cite you. You can, however, be the clearest source on everything about your business. If you'd like to see where your site stands today, a technical and content audit is a good place to start.

Does Google penalise content written with AI?

Not for being written with AI. Google has said publicly that it judges content on quality and usefulness, not on the tool used. It does, however, treat mass-produced content made only to climb the rankings as spam.

Can I translate my site into English with AI alone?

You can start that way, but don't stop there. Machine translation gives a decent first pass; your sector's terminology and a natural tone need someone who writes the language well.

Do I need anything special to appear in ChatGPT or Perplexity?

No tricks. You need accurate, consistent details, structured data, fast pages, copy that answers clearly, and a robots.txt file that doesn't block these tools' crawlers. It's largely the same good SEO that helps you on Google.