Stop faking testimonials, start collecting real ones, and run four checks before your app goes live.
Last week we covered cookies and accessibility. This week: the testimonials AI quietly made up for you, how to build a system that collects real ones, and four more things that protect your visitors and your users' data.
Ask AI to build a landing page and it will happily fill the testimonial section with glowing reviews from people who don't exist. Placeholder names, five stars, a stock photo. A lot of people launch without ever deleting them.
Search my entire web app, every page, component, and data file, for testimonials, reviews, star ratings, customer counts, and customer logos. Remove anything that didn't come from a real customer, and show me what you removed.
Then add this as a permanent rule to my project's rules file (like AGENTS.md or CLAUDE.md):
Never create or keep fake or placeholder testimonials, reviews, star ratings, customer counts, or customer logos. Only use reviews I provide from real customers. If a section needs reviews and I haven't given any, leave it empty and tell me.
I was paying for Senja to collect and display testimonials. So I built my own inside my website, using Claude, Supabase, and Vercel. Now I own the testimonials and the system that collects them.
A simple form: star rating, written or video story, optional photo, and permission to publish. No account needed.
Nothing goes public until you approve it. Choose where each story shows up.
A public page showing only approved stories. Emails are never shown.
Each workshop gets its own link, so its stories land on that workshop's page.
Build a testimonial system into my existing website. Match my current branding. Build it locally first.
1. A public "Share your story" page with a 1 to 5 star rating, a written or video testimonial, name, email, role or business, an optional photo, and a required checkbox giving me permission to publish it. Show a thank-you screen after submitting.
2. Store submissions in Supabase with Row Level Security turned on, so names, emails, and permissions can never be read from the browser. Add a limit on how many times someone can submit, to stop spam.
3. A private dashboard, protected by a login only I can use, where I approve, hide, or feature each story and choose where it appears.
4. A public testimonials page that only shows approved stories and never shows anyone's email.
5. Let me create a separate link for each workshop or offer, so stories are tagged automatically.
Explain each step in plain language before you build it, and walk me through testing the whole journey at the end.
You don't need something to sell yet. Every time you help someone, ask for their story. You'll want them later.
I used Senja as my example, but it doesn't have to be Senja. It could be any testimonial tool you're paying for, or one you like that does a good job. Study what it does well, then build your own version.
Most AI-built sites get their fonts from Google with one line of code, hidden in the page:
That line tells every visitor's browser: "Go to Google and get this font."
When someone opens the page, their browser quietly makes two trips to Google before they've clicked anything:
| Request | What it is |
|---|---|
| fonts.googleapis.com/css2?family=DM+Sans… | Asks Google for the font instructions |
| fonts.gstatic.com/s/dmsans/…woff2 | Downloads the font file from Google |
Your visitor never agreed to share any of that with Google. In 2022, a German court ruled a website broke Europe's privacy law by doing exactly this, and ordered it to pay the visitor €100. That started a wave of demand letters across Europe. It's the same issue as the pixel: a third party gets your visitor's data without asking.
Download the font and serve it from your own site. Same look, no trip to Google, and it often loads faster.
Find every place my site loads fonts from Google Fonts (fonts.googleapis.com or fonts.gstatic.com). Download those fonts, save them in my project, and load them from my own site instead. Keep the same fonts and weights so nothing looks different. Then show me how to confirm in the Network tab that nothing loads from Google anymore.
Supabase has Row Level Security (RLS), but it's a switch, and it isn't on for every table. Tables made in Supabase's Table Editor usually have it on. Tables your AI creates with code often don't.
The stranger's request returns every user's information.
The same request returns an empty list, and adding fake data is blocked.
Using the Supabase connector, check every table in my database. Confirm Row Level Security is turned on for each one, and that users can only see and change their own rows. Run Supabase's security advisor too. Then test it: using only my public key, try to read and add rows the way a stranger would, and show me the results. Explain anything that's wrong in plain language before you change it.
Rate limiting is a cap on how many times someone can try something in a set amount of time, like 5 login attempts per minute.
Supabase limits some things automatically, like login emails. Your own forms and AI features need their own limit. My testimonial form has one.
You choose the rules. The common options:
| Option | How it works |
|---|---|
| Short window | "5 tries per minute." After that, wait 60 seconds. Most people never notice. |
| Gradual slowdown | Each failed try adds a longer wait: 1 second, then 5, then 30. People barely feel it. Bots slow to a crawl. |
| Prove you're human | After a few failed tries, show a "Verify you're human" checkbox instead of blocking. |
| Reset by email | Offer "Send me a login link" so a real person is never stuck. |
Limits reset on their own once the time window passes. Nothing to do manually.
Add rate limiting to my web app. Cover login, sign-up, password reset, every form, and any feature that calls an AI or other paid service. Use short time windows (like 5 tries per minute) so real people are never locked out for long. After repeated failed logins, show a "verify you're human" check and offer a login link by email. Explain the limits you chose in plain language, then show me how to test that they work.
Last week: your cookie banner asks before tracking anyone. This week: keep proof of the answer.
In your database (Supabase), in its own table. One row per choice:
| What's saved | Example |
|---|---|
| A random ID for that browser | a7f3-29c1 |
| What they chose | Accepted |
| When | Oct 16, 2026, 2:14 PM |
| Which banner version they saw | v2 |
That same random ID is also saved in the visitor's browser cookie. If anyone claims they were tracked without permission, you can match the ID in their browser to your record showing that browser clicked Accept on a specific date, without ever storing who they are.
Your record has to prove a choice was made, not who made it. An IP address counts as personal data, so storing it means more to protect, more to disclose in your privacy policy, and more to delete if someone asks. If your database ever leaked, a list of random IDs exposes nobody.
Add a consent record to my cookie banner. When a visitor clicks Accept or Decline, create a random ID, save it in their browser cookie, and save a record in a new Supabase table with that ID, their choice, the date and time, and the banner version. Don't store their name or IP address. If they're logged in, link the record to their account ID. Turn on Row Level Security so no one can read these records from the browser. Then show me how to check that a record was saved.
I'm building a tool that checks your site for all of these at once: tracking that fires too early, Google Fonts, missing accessibility basics, and more. As I build it and fix my own platform, I'll keep you in the loop.
Missed a week? See all past live sessions →