Your site

Everything Baseline knows about your own site, in one place.

Every other package works from the outside, on a domain you type in. This one starts by putting Baseline on your site. After that, one set of screens says which AI crawlers came, which visits the answer engines sent back, and how much of the site still carries correct markup.

Open your site dashboard Connect your site

A plugin upload, a worker deploy or one module call. All three are free.

Connect, watch, maintain

The package is one loop in three parts, and they only work in that order. Nothing can be measured until something on your site is reporting; nothing is worth correcting until you can see what is wrong. Connect once and the other two turn on for every domain on the study.

Connect

Pick the channel that matches how your site is served. Each one has its own walkthrough, with a check at the end that asks our side whether anything actually arrived.

The channels

WordPress - a standard plugin: upload it, paste a key, five steps with a screenshot each
Cloudflare - one worker at the edge, ahead of every cache, in front of whatever you run
Custom stack - a dependency-free module for Node, Deno, Bun or the edge, called once per request
Google - read-only Search Console and GA4, drawn on the same days as the traffic we measure

The custom module reports crawlers but cannot inject schema, because it never touches your HTML. Connecting Google installs nothing at all: it reads your own numbers back and delivers nothing to your site.

Connect your site

Watch

Four screens on the same study, from four angles. Two read what your connection reported; two read the scans you have already run.

The screens

Site dashboard - what is connected, which crawlers came, how much of the site carries schema
AI Traffic - visits with the answer engine that sent them, named rather than filed under referral
Analytics - your Baseline usage: reports, scans, backlinks and per-engine query performance
Sources - the domains the engines cited across your visibility runs, ranked by citation share

Open your site dashboard

Maintain

A connected site can be corrected as well as measured. Both of these write back through the channel you already installed, and both are off until you switch them on.

What keeps running

Rolling Schema - schema on every page of a connected domain, regenerated as the pages change
Rolling Metadata - swaps a fact your page now contradicts, and writes a description only where you have none

Rolling Schema belongs to the Schema tools package, beside the audit that finds the gaps and the generator that writes the first version. It is listed here because it needs the same connection everything else on this page needs.

See Rolling Schema

Which one do you need?

I have no idea whether AI crawlers read my site at all: Connect a channel and the Site dashboard answers within a day. This is the one question you cannot ask from outside: a crawler fetching your page leaves a mark on your server and nowhere else. A tag in the browser only ever sees people, so a server-side channel is what makes bots visible. If you want the complete count rather than most of it, the Cloudflare worker runs ahead of your cache and sees every request that reaches the domain.
People tell me ChatGPT sent them, but my analytics never shows it: That is what AI Traffic is for. To an ordinary analytics tool an answer engine is just another domain, so those visits land under referral or direct and the engine's name is lost. Our tracker ships the raw referrer and any utm parameters, and the classification happens on our side at ingest, so the visit arrives already attributed to ChatGPT, Perplexity, Gemini or Google AI Mode. Connect Google and Search Console clicks sit on the same days for comparison.
My pages keep changing and the markup behind them goes stale: Rolling Schema regenerates the markup for a connected domain as its pages change, rather than leaving you with whatever was true the day you pasted it in. Rolling Metadata does the smaller, sharper version of the same job on your titles and descriptions: it may swap a price, a time or an address the page now contradicts, and nothing else. Your site stays the source of truth, so the moment you edit the line yourself, our correction retires and never comes back.

How the three fit together

Connect. Install one channel on the domain your study points at. The walkthrough ends by asking our side whether anything arrived, which is a different question from whether the code is on the page: a full-page cache can answer every request before a correct install ever runs.
Watch. Crawler visits and AI referrals start accumulating against that domain from the first request. The Site dashboard is the one screen that puts them together with your schema coverage, so you can see a crawler arriving and whether it found anything to read.
Maintain. Once the site reports, the two rolling features can write back through the same channel: schema for pages that have none, and a correction where your own metadata says something the page no longer does. Both stay off until you turn them on, per domain.

What the tag collects, and what it does not

The traffic tracker is a thin script with a deliberately short list. It sends the path, whether the hit was a pageview or a conversion, an anonymous first-party visitor id, a session id, and, on the first hit of a visit only, the referrer and any utm parameters that brought it. That is the whole payload. Classification never happens in the browser: the script ships raw signals and our server decides which answer engine they represent, so a change in how we recognise an engine does not need you to paste anything again. The visitor id is a random string kept in localStorage, with a cookie only as a fallback where localStorage is blocked. Internal navigation is dropped, because a link from one of your pages to another is not a traffic source.

It never asks for a name, an email or anything you would have to collect a consent for, and it sets no cross-site identifier: the id is meaningless on any other domain. No IP address is stored. The one place an address is read at all is the Cloudflare worker's bot reporting, where it is checked against the ranges the crawler operators publish and thrown away in the same breath - never written to a row, never logged, never used as a cache key, with one true-or-false answer surviving. What is kept verbatim is the request's user agent, which is how a bot that slipped past the filter can be recognised later instead of being lost. The script fails open throughout: it never throws into your page and never blocks rendering.

Common questions

No. Every scan and every free tool works from your domain alone, with no plugin, no tag and no DNS change. Your site is the one package that needs a connection, and the reason is physical rather than commercial: an AI crawler fetching your page leaves a record on your server and nowhere else, so nothing outside your site can report it.

The page path, whether the hit was a pageview or a conversion, an anonymous first-party visitor id, a session id, and the referrer and utm parameters of the visit's first hit. No name, no email, no cross-site identifier and no stored IP address. The id lives in localStorage, with a cookie only where localStorage is blocked, and it means nothing on any other site.

Coverage differs by channel, and it is a property of where the code runs rather than a bug. A plugin or a module runs inside your application, so any request a full-page cache answers never reaches it. A worker at Cloudflare's edge sits ahead of the cache and sees every request that reaches your domain. We say which of the two you have rather than quoting a partial count as a total.

Yes. The three install channels deliver the same schema and report the same crawler traffic, and they run together happily - a worker in front and a plugin behind it is a normal setup. Connecting Google is separate again and never conflicts with any of them, because nothing of ours runs on your site as a result of it.

No, and it is built so that it cannot. Two switches, both off by default: correction may only swap a fact the page itself now contradicts - a time, a price, an address, a phone number, a date, a count - and never tone, length or wording. Generation only writes a description for a page that has none, and may never touch a title. Every proposal has to carry one sentence quoted from the page, and an answer we cannot read is treated as no change. The moment your own metadata differs from the text a row was made against, that row retires and is never applied again.

AI Traffic is people arriving at your site, with the engine that sent them named. Sources is the other end of the same conversation: the domains an engine actually cited while answering your tracked questions, taken from your visibility runs, and most of them will not be yours. One measures traffic you received, the other measures who got quoted in the answer you were hoping to be in.

Connect it once, and the rest of the package turns on

Pick whichever channel matches how your site is served, and none of them needs a plan. After that the crawlers, the AI referrals, the schema coverage and the metadata drift all report against the same site.