Note: This essay is cross-posted from pagecoder.ai/blog/why-your-ai-chatbot-is-a-tracker/. The version there is canonical. PageCoder.ai uses no third-party analytics, no cookie banner, no tracking pixels — same stance applies to this Mirror cross-post (read-only, not minted).

Last week we sat in a café in Wien with a friend — a freelance dev who's been shipping WordPress sites for a decade. He'd just installed an AI chat plugin for a client, a small cosmetics brand. It looked nice. Brand colors, custom name.

Then he opened the network tab.

"Three. Four. Five." He scrolled. "Eight. Wait — eleven? Where is this one going?" The cappuccino arrived; nobody touched it.

If you run a WordPress site with any AI chat plugin from the last two years, this is what's happening every time someone opens that bubble in the corner.

The architecture of the leak

A visitor types a question. Three things happen before the bot replies.

First, the question goes to a third-party AI provider — one of the big three in the US. The vendor's terms say the prompt isn't used for training "by default", but the request still hits their infrastructure, gets logged, sits in a datacenter for some retention window the plugin developer never read.

Second, a copy goes to the chatbot vendor's own backend. Browser fingerprint, IP, geolocation, the full conversation, the URL the visitor was reading. This data is the actual product the chatbot vendor sells. The widget is the loss-leader.

Third, the request fans out to CDNs and analytics endpoints embedded in the widget — the rest of the eleven. None of them owe anyone a privacy policy. By embedding the widget, the site implicitly authorized everyone in its dependency tree.

A single question on a small WordPress site running a popular AI chatbot can leak to six to eleven different companies before the page renders the reply. None of them appear in the site's cookie banner. EU regulators are catching up. Default architecture isn't.

Why we started over

We'd been installing those plugins for clients for years. We'd ticked the privacy-policy boxes. We'd stopped noticing. Then a client asked, casually, "where exactly does that go when someone types it in?" We didn't have a clean answer.

So we started over. Not because we were better at privacy than the people writing those plugins — we weren't, and aren't — but because the gap between cookie banner and network tab had become embarrassing.

What we built instead

RAG Chat for WordPress. Three architectural decisions:

Vectors live on your server. The numerical representations of your content — what the chatbot searches against — sit in your WordPress database. Not our cloud. Not a third-party vector DB. Uninstall the plugin and the data is gone, because it was always yours. We can't sell it, lose it, or even see it.

The math is stateless. Our backend runs the similarity search and forgets the request happened. No logs. No "queries collection" tab in our admin, because the collection doesn't exist. The only honest way to promise we won't lose your visitors' data is to never have it.

Zero third-party calls from the widget. No analytics pixel. No social tracking. No CDN fonts. No public package registries. Open the network tab on a site running our plugin and you'll count to two.

This is the boring, harder, less-impressive version of an AI chat plugin. It's also the only one that lets a small brand keep its privacy policy honest.

The Loop

Standard chatbots forget the conversation. The next visitor asks the same question, the bot answers again, the site never learns.

We thought that was the actual product.

Our plugin watches what visitors ask. When several variations of the same question come in, it clusters them, drafts a clean FAQ entry, and shows the admin two buttons: publish or discard. AI proposes. Human curates. The output is a real indexed page at /faq/your-question/, with FAQPage schema, discoverable by search engines.

We call it FAQ Builder. Nothing publishes without admin approval. The bot turns a never-ending support backlog into a weekly editorial habit.

Most chatbots are a feature. Ours is a feedback loop.

Why we made a game

Reading about privacy is exhausting. So we made a 3-minute interactive game. It walks you through what your fingerprint contains, who's watching, and what happens in the hundred milliseconds between a click and a page load (an auction; you're sold). By the fifth level you've effectively designed our product without us mentioning we exist.

The game has no tracking pixels. This article has none. The site has none.

The thing nobody writes blog posts about

A handful of small WordPress sites are running our infrastructure today. You wouldn't know which — they don't say so. The bubble is in the client's brand colors. The bot has a name the client picked.

Most chatbot vendors put their logo on the widget because every visit is a brand impression. We don't. Trust doesn't compound when there are two names on the door. The brand belongs to the business. The data belongs to the business. The infrastructure belongs to us — somebody has to build the thing — but the visitor's relationship is with the business, not us.

What to do about it

If you're shipping a WordPress site for a client this quarter and considering an AI chatbot, ask three questions:

1. Where do my visitors' questions go? If the answer involves any company name other than yours, the answer is "to that company too".

2. Where do you store my content? "Our cloud" = your content is part of their dataset. "Your database" — ask to see the table.

3. What happens if I uninstall? "Your data stays with us forever" vs "the data is gone, because it was always yours".

We pass all three. We're one of the few that do.

We sent our friend the link to the game. He played it, swapped out his client's plugin, and sent back a screenshot of the network tab. The count was two.

Try the chatbot →

Or: pagecoder.ai · Privacy Tour · 3-min game

No tracking pixels were used in this manifesto. None will be, ever.

Originally published at pagecoder.ai/blog/why-your-ai-chatbot-is-a-tracker on 7 May 2026. PageCoder.ai is privacy-first marketing infrastructure for honest businesses. The chatbot product is called RAG Chat.