Launch checklist for a portfolio or marketing site
A grouped launch checklist for portfolio and marketing sites: DNS, redirects, CSP headers, hreflang, sitemap, OG images, JSON-LD, GA4 consent and forms.
A small site looks finished long before it is ready to launch. The pages render, the animations feel good, and then a week later you find the www version indexed separately, the Indonesian pages missing from search, and a contact form that bots love.
This is the checklist I ran before putting this site live. Every item comes from something the repo actually does, with the reason it matters. The site is static Astro on Cloudflare Workers static assets (I explain that choice in Astro vs Next.js for marketing sites), but almost all of it applies to any stack.
Domain, DNS and HTTPS
- Pick one canonical host and stick to it. This site uses the apex,
https://ikhsansdq.id. Every canonical tag, sitemap URL and JSON-LDurlis built from one constant, so they cannot drift apart. - Redirect
wwwto the apex with a 301. Without it, search engines can treat two hosts as two sites. On Cloudflare this is a Redirect Rule, and thewwwname needs a proxied DNS record (a dummyAAAA 100::works) or the rule never fires. - Keep the path and query in that redirect. A
wwwlink to a blog post should land on the same post, not the home page. Test it withcurl -sIand read thelocationheader. - Turn on Always Use HTTPS, and add HSTS later. Wait until HTTPS has run cleanly for a week or so, then start with a moderate
max-ageand no preload. HSTS mistakes are hard to undo because browsers remember them.
Security headers
- Ship a baseline set for every path. This site sends
X-Content-Type-Options: nosniff,Referrer-Policy: strict-origin-when-cross-origin,X-Frame-Options: DENYand aPermissions-Policythat switches off camera, microphone and geolocation. They cost nothing and close common gaps. - Write a Content Security Policy that lists real origins. Start from
default-src 'self'and add only what you load: here that is Google Tag Manager, Google Analytics and the Cloudflare Insights beacon. Addobject-src 'none',base-uri 'self',form-action 'self'andframe-ancestors 'none'. - Be honest about
'unsafe-inline'. This site still allows inline scripts because of the theme switch and the GTM loader. Hashes, or Astro’s CSP API (stable since Astro 6), are the stricter path if you can move to them. - Know where your headers apply. A
_headersfile on Workers static assets covers files served from assets, not responses your Worker code generates. The contact API sets its ownnosniffandcache-controlfor that reason. - Cache hashed assets forever, everything else briefly.
/_astro/*getsmax-age=31536000, immutablebecause file names change with content. OG images get one day.
Canonical URLs and hreflang
- One trailing-slash policy. Every page here canonicalises to a path ending in
/, and file paths like/rss.xmlstay as they are. The canonical strips query strings and hashes. - Add hreflang for every language, plus
x-default. Each page links to its English and Indonesian versions and names English as the default. Google’s localized versions guide explains why the tags must point both ways. - Leave canonical and hreflang off noindex pages. A 404 or a hidden page should not claim a canonical URL.
Sitemap, robots.txt and llms.txt
- Generate the sitemap from your content. This one is built at build time from the static pages and the content collections, one entry per locale with hreflang alternates. Adding a post updates it without touching config.
- Use real
lastmoddates. The sitemap takesupdatedDatewhen a post has one and falls back topubDate. Fake “today” dates teach crawlers to ignore the field. - Keep noindex pages out of the sitemap, but crawlable. The case studies here are noindex (they outranked the clients’ own sites for their brand names). They carry
X-Robots-Tag: noindexand are not blocked in robots.txt, because a crawler that cannot fetch the page cannot see the noindex. - Decide on AI crawlers on purpose. This robots.txt allows everyone, names GPTBot, ClaudeBot, PerplexityBot and Google-Extended explicitly, and disallows only
/api/. Naming them means a later change to the default cannot block them by accident. - Add an
llms.txtif you want answer engines to have a clean summary. It is a community proposal, not a standard, so treat it as cheap extra context rather than an SEO lever. - Noindex your preview deployments. A preview URL in search results splits your signals.
Open Graph images
- Give every page a 1200 by 630 image. This site renders one PNG per post at build time with satori and resvg, from the entry’s
ogTitleandogTag. No one has to design a card by hand for each post. - Keep OG titles short and send the image size. The schema caps
ogTitleat 70 characters so previews do not cut it off, andog:image:width,og:image:heightandog:image:altsave platforms from guessing.
Structured data
- Describe the person and the site once. The home page emits
PersonandWebSiteJSON-LD with stable@idvalues, and every post points back to them as author and publisher. - Add
BlogPostingandBreadcrumbListto posts. IncludedatePublished,dateModifiedandinLanguage. They help search engines understand the content even when no rich result shows. - Escape
<when you inline JSON-LD. A title containing</script>should not be able to break out of the tag. The serializer here replaces<with the escape\u003c.
Performance and Core Web Vitals
- Set an islands budget. Only a few components hydrate here (navigation, the blob, the bento tiles), each with a deliberate
client:directive. Everything else is HTML. - Self-host fonts. It removes a third-party connection and lets the CSP keep
font-src 'self'. - Check field data, not just Lighthouse. The Core Web Vitals “good” thresholds are LCP at or under 2.5 s, INP at or under 200 ms and CLS at or under 0.1, measured at the 75th percentile. Test on a mid-range phone.
Accessibility
- Add a skip link to the main content. Here it sits at the top of every page and appears on the first Tab press.
- Respect
prefers-reduced-motion. The splash screen never runs with reduced motion, and the shared animation helpers, plus smooth scrolling, switch off and simply show the content. - Never let JavaScript hide content for good. Reveal animations hide content only while the script is expected to run. If it has not started within two seconds, everything shows, and a
noscriptstyle covers the no-JS case. - Check contrast and keyboard focus in both themes. Dark mode is where contrast bugs hide.
Analytics and consent
- Load analytics in production only. The GTM snippet is rendered only in production builds, so local development never pollutes reports.
- Set Consent Mode defaults before the tag loads. This site has no ads, so all ad signals are denied. Analytics storage is granted unless the browser sends Global Privacy Control. If you run ads or target EU and UK visitors, you need a consent banner. See Google’s consent mode docs.
- Tag pages by template. A
page_typevalue (home, blog entry, resources index and so on) that is the same in every language makes GA4 reports readable. - Say what you collect. The privacy page lists GA4 and Cloudflare Web Analytics.
404 page
- Serve a real 404 status with a helpful page. The Workers config sets
not_found_handling: "404-page". The page is noindex, links back home, and is never listed in the sitemap.
Forms
- Validate on the server, not just in the browser. The contact endpoint checks required fields, minimum and maximum lengths, email format and allowed values for every choice field.
- Accept same-origin JSON only. It rejects other origins and other content types.
- Add a honeypot and a rate limit. A hidden field that bots fill in, plus a per-IP limiter. The in-code limiter is best-effort, so add a WAF rate-limiting rule as the real limit.
- Keep secrets out of the repo, verify the sending domain, then test. The email API key is a Worker secret, SPF and DKIM are set before the first message, and one test from production checks the inbox, the spam folder and the analytics event.
When everything above is ticked, tell people about the site. If you are handing designs to developers, the Figma to React handoff checklist covers the step before this one.