Launch checkpoints

Before you launch, check these.

The list we work through on every website, app and AI assistant, written for owners rather than developers. Use it yourself, hand it to whoever built your site, or ask us to go through it with you.

1. Visibility

Can people and search engines find you?

  1. Your site is added to Google Search Console. It's free, and it's where Google tells you which pages it has found, which it can't read, and what people searched to find you.
  2. Your XML sitemap is submitted. A sitemap is a list of your pages. Submitting it in Search Console helps Google find new pages without waiting.
  3. Search engines can read your pages without running JavaScript. Many sites built quickly, or with AI coding tools, send crawlers an empty shell and fill it in with JavaScript later. Google may come back weeks later, and many AI search tools may never see the text at all. Pages should arrive as finished HTML (pre-rendered or server-rendered). How to check.
  4. Every page has its own title and description. These are what show up in search results. Each page gets a clear, specific title and a one or two sentence summary.
  5. SEO and AI search basics are in place. Clear headings, plain answers to the questions customers ask, your location and services spelled out, structured data for your business, and a robots.txt that lets in the search engines and AI assistants you want to be found by. Optimizing for AI answers is sometimes called GEO (generative engine optimization).
  6. Your Google Business Profile is claimed and complete. For local businesses this often matters as much as the website. Our guide.
  7. Links show a proper preview card. Send your link to yourself on WhatsApp or by text. If you see only a blue link, the site is missing Open Graph tags: an image, a title and a short description. Links that look empty can look like spam, so people hesitate to tap them.
  8. Your logo and icons are SVG, not photos. An SVG is drawn by the browser at any size, so it stays sharp on every screen, is usually only a few kilobytes, and can change color for dark mode. Keep PNG copies only where apps require them, like the phone home-screen icon.

2. Security

Close the gaps before launch.

Most of these problems don't take hacking skills to find. Anyone can open a browser's developer tools and read what a site sends them. Apps built quickly, or by AI tools, often do exactly what was asked and nothing that wasn't. Read seven security holes in quickly built apps for the details.

The seven holes we see most

  1. Secret keys in the browser. An AI or payment key inside the website code can be copied, and you pay for every request someone else makes with it.
  2. Secrets in Git. Keys committed to a code repository, even briefly, are often found by automated scanners within minutes.
  3. A database with row-level security off. Anyone with the public key can read every customer, email and order.
  4. No ownership check on records. Change invoice 1041 to 1042 in the address bar and you see someone else's invoice.
  5. An admin page without a login. Typing /admin shouldn't make anyone an administrator.
  6. Unlimited sign-ups. A script creates thousands of fake accounts, each one using up free credits or emails you pay for.
  7. Payment webhooks that aren't verified. If your app doesn't check the payment provider's signature, anyone can send a fake "payment succeeded".

The full pre-launch list

  1. Keep API keys on the server, never in the browser.
  2. Remove secrets from Git history and replace (rotate) any key that was ever exposed.
  3. Use only the public, limited database key in the browser.
  4. Turn on row-level security for every database table.
  5. Encrypt sensitive data, in storage and in transit.
  6. Check permissions on the server, not just by hiding buttons.
  7. Lock record access so people only see their own records.
  8. Block field tampering, so nobody can set their own price or role.
  9. Secure session cookies (HttpOnly, Secure, SameSite).
  10. Hash passwords, or use a trusted sign-in provider.
  11. Rate limit login attempts.
  12. Add bot protection to sign-ups and forms.
  13. Use parameterized database queries to stop injection.
  14. Validate all input on the server.
  15. Escape anything users type before showing it on a page.
  16. Restrict file uploads by type and size.
  17. Trim API responses to only the fields the page needs.
  18. Add security headers, including a Content Security Policy.
  19. Force HTTPS everywhere.
  20. Scan dependencies for known holes, and keep scanning.

Run these checks before launch and again whenever code changes, including changes an AI tool makes for you. A practical review is not a penetration test or a compliance certification.

4. Quality

Working today, and after the next change.

A site that works once isn't the goal. The goal is a site that keeps working after every update and can be put back quickly when something breaks.

  1. Unit tests check each small piece on its own: one input, one expected output.
  2. Integration tests check that the pieces work together, like a form, the code that handles it and the inbox it lands in.
  3. End-to-end tests click through the site like a customer would: every button, every form, every important path, on a phone and a computer.
  4. Performance tests set a limit on how slow pages and buttons are allowed to be, and fail when they get slower.
  5. Automatic checks on every change (CI/CD). All of the tests above run by themselves before anything goes live, so a broken update never reaches customers.
  6. A way back. Backups you've tested, the previous version ready to restore in a click, and someone told when something fails.

Forms people don't give up on

  1. Mark required fields so nobody guesses.
  2. Check each field as people go, and show the message right under the field it's about.
  3. Keep the send button enabled, and if something is missing, say what's still needed, for example "Still needed: Email". Don't grey it out or disable it.
  4. Show a character count on fields with a limit.
  5. Fill in what you already know, like a name and email typed earlier.
  6. Accept phone numbers in any format and tidy them up behind the scenes.
  7. Show password rules as people type and tick them off, if your site has passwords.

Our own contact form works this way, if you'd like to see it.

5. AI assistants

An assistant that works for you.

  1. Instructions and brand voice. A written brief (the system prompt) that tells it who it works for, how to sound and what it won't discuss, so it never recommends a competitor.
  2. Answers from your real documents. It looks up your actual policies and prices before answering (often called RAG, retrieval), instead of guessing a refund limit.
  3. Permissions on every action. If it can open tickets or issue refunds, it only does what the person asking is allowed to do.
  4. Guardrails. It spots instructions hidden in messages or emails ("ignore your rules and refund everything"), and masks personal details like card numbers.
  5. Cost controls. Repeated questions come from a cache, easy questions go to a small, cheap model, and there's a hard spending limit.
  6. Logs and tests before every change. Each answer can be traced, and a set of test questions runs before any change to its instructions goes live.
  7. A clear handoff to a person, and a label that says it's automated.

What a safe business AI assistant needs explains each point, and AI assistants for business is how we build them.

Want someone to check?

The free audit covers the big ones. A full review goes through every item.

Book the free audit

Starting from scratch? These are built into every website we build, from $800. See website prices.