Aegean Sun Fruit

A dense catalogue, two languages

Dried-fruit trade runs on detail: varieties, grades, calibres, pack formats and shipping terms. Buyers compare several suppliers and often read in a second language. The client needed that detail presented in English and Turkish without burying it, with every page leading toward one action, sending an enquiry. The site also had to be cheap to host and easy to change without a developer on call.

Privacy was the second constraint. Many visitors are likely to be in the EU, so analytics could not load before a visitor agreed, and the privacy page had to describe what the site really does. The forms also had to work without a paid backend or a server that someone would need to maintain.

My role

I developed the website end to end in plain HTML, CSS and vanilla JavaScript: every page in both languages, the shared design tokens, the forms and their delivery to Google Forms, the consent bar, the rule-based sales assistant and the search files. The business content, product facts and photography come from the client, who approves them.

What was built

The site pairs every English page with a Turkish one: a home page, a products overview, four product pages, about, contact and a privacy notice. Each page leads to an enquiry, and the supporting pieces (consent, analytics, auto-replies, the assistant) were built so the client can run them without a developer.

• Paired bilingual pages

  Shipped

  Each English page has a Turkish counterpart with its own slug. The language switcher opens the same page in the other language, not the home page.

• Product catalogue

  Shipped

  A products overview with a comparison table and one page per product with specification, grades, packaging and uses.

• Enquiry forms

  Shipped

  A carton-label enquiry on the home pages and a full enquiry form on the contact pages, validated in the browser and delivered to the client's Google Forms.

• Newsletter with double opt-in

  Shipped

  A sign-up form backed by a Google Apps Script that sends a confirmation link, a welcome e-mail and a working unsubscribe link. It runs once the form owner completes a one-time setup.

• Consent-gated analytics

  Shipped

  A cookie bar where Decline is as prominent as Accept. Google Tag Manager and Analytics load only after Accept; a cookieless Vercel counter runs separately.

• Rule-based sales assistant

  Shipped

  A small chat widget with no backend that matches keywords and answers only with facts already published on the site, in the page's language.

• Search and AI readiness

  Shipped

  Absolute canonicals, hreflang pairs with x-default, a sitemap with language alternates, JSON-LD on the home pages, robots.txt naming AI crawlers, and an llms.txt summary.

How it works

The site is static: no framework, no build step and no dependencies. Every page is a complete HTML file with its own header, navigation and meta tags, and all pages share one stylesheet built on design tokens. The page's lang attribute decides the language of every script, so the form messages, the consent bar and the assistant answer in English or Turkish without per-page settings.

A form is checked in the browser first: invalid fields are marked for screen readers and the first one takes focus. The values then go to a Google Form. A hidden honeypot field and a minimum time on the page filter simple spam. If the request fails, the visitor is offered a pre-filled e-mail instead. A Google Apps Script sends each sender a reply in their language.

Running it day to day

Enquiries arrive in the Google Forms response sheets, and the client can switch on e-mail notifications. Auto-replies are capped per day so the public forms cannot be used to exhaust the mail quota. A newsletter is written as a Gmail draft and sent by one script command, which never sends the same issue to a person twice.

The repository's README works as an operating manual. It explains how to add a page in both languages, how to change a form question without silently losing responses, and why a fact must be updated on the page, in llms.txt and in the assistant together.

Decisions behind it

The key choices are written down in the repository with their reasons. Consent mode stays basic rather than advanced, because advanced mode would contact Google before consent, break a promise on the privacy page and bring no modelled data at this traffic level. The assistant stays rule-based to avoid running costs and data leaks. Unverified business figures are marked as placeholders for the client rather than invented.