9 November 2026

Your website data, read and watched by AI

Most small business websites have Google Analytics on them. Very few small business owners look at it. It’s not laziness. The dashboards are confusing, the numbers don’t tell you what to do, and there’s always something more urgent.

This is where AI changes things more than anywhere else in this series. Once your analytics and Search Console are connected to your AI, you stop reading dashboards and start asking questions: “Which posts brought people in this month, and what did they search for?” And once that works, you can hand it a schedule, so every Monday it reads the numbers for you and tells you what changed.

This guide covers getting the data in place, connecting it to your AI, and setting up scheduled tasks to keep an eye on it.

At a glance

Set up three data sources: Google Analytics (what people do on your site), Search Console (how Google finds you) and Cloudflare (what’s hitting your site, including bots and AI crawlers).
Connect them to your AI so you can ask questions in plain English. Read-only to start with.
Use scheduled tasks to have the AI check the data regularly and report back.
Pick the right kind of schedule. Some run in the cloud with your laptop closed, some need your computer on.
Keep it read-only and draft-only. The AI reports and suggests. You decide.

Step 1: get the data flowing

Google Analytics 4

Analytics tells you who visits, where they came from, what they read and whether they did what you hoped (signed up, got in touch, bought).

On WordPress, Google’s Site Kit plugin connects Analytics and Search Console in a few clicks. Or ask Claude Code to add the Google tag to your theme.
On a static site, ask Claude Code to add the Google tag to every page.
Set up key events for the things that matter: a form sent, a signup, a click on your phone number. Without these, you’ll only ever know how many people turned up, not whether it worked.
Cookie consent. In the UK, analytics cookies have generally needed consent, and the rules have been changing. Check the ICO’s current guidance before you go live, and make sure your cookie banner actually does what it says.

Google Search Console

Search Console shows how your site appears in Google: which searches you show up for, how often, where you rank and how many people click. Clicks from AI Overviews and AI Mode are counted in there too, under the normal “Web” results.

Add your site as a domain property and verify it with a DNS record. If your DNS is on Cloudflare, that takes a couple of minutes.
Submit your sitemap.
Be patient. The data runs a couple of days behind, and it takes a few weeks to show anything useful on a new site. It keeps 16 months of history, so start early.

Cloudflare

Cloudflare already sees every request to your site. Its analytics show traffic, threats blocked, and (through AI Crawl Control) which AI services are reading you. It’s the only one of the three that sees bots and crawlers, because Google Analytics ignores them.

Step 2: connect your data to your AI

This is the step that turns data into answers. There are a few ways, from easiest to most hands-on.

Google Analytics: Google publishes an official Google Analytics MCP server. MCP is the standard way to plug a tool into an AI like Claude. Google’s version is read-only by default (it can’t change anything unless you deliberately switch that on), and it can run reports, real-time reports and look up your properties. Setting it up needs a Google Cloud project with the Analytics Admin and Data APIs switched on, and a sign-in with read-only access. It’s a fiddly half hour the first time. Ask Claude Code to walk you through it step by step, and it’ll do most of the typing.

Search Console: there’s no official Google MCP server for Search Console at the time of writing, but the Search Console API is well documented. You can either use one of the community-built MCP servers (read the code or have Claude review it first, because it’ll have access to your data), or ask Claude Code to write you a small script that pulls the last 28 days of searches into a file. The API can query your search data, list and submit sitemaps, and inspect individual pages.

Cloudflare: Cloudflare’s skills plugin for Claude Code (from part 2) includes its own MCP server, which can look at your account. Start read-only here too.

The no-setup option: export a CSV from Search Console or Analytics and drop it into a chat. It’s not automatic, but it’s a perfectly good way to start, and it’s how we did our first proper read of Search Console data.

Step 3: ask better questions

Once it’s connected, the trick is asking questions that lead to an action. Some we use:

“Which pages got the most search impressions in the last 28 days, and which of those have a click-through rate under 1%? Suggest a better title for each.”
“Which searches are we showing up for on page two? Which of those could we answer properly with one new post?”
“Compare this month with last month. What went up, what went down, and what’s the most likely reason?”
“Which posts bring people who then sign up, and which bring people who leave straight away?”
“Which AI crawlers visited this month, and did any of them hit pages we didn’t expect?”

And always: “What’s the one thing you’d change this week, based on this?”

Step 4: let the AI watch it for you, with scheduled tasks

This is where it stops being a tool you remember to use and becomes something that works for you. A scheduled task is a prompt that runs on a timer: daily, weekly, monthly, whatever you choose.

There are three kinds in the Claude world, and they’re good at different things.

Cloud scheduled tasks run on Anthropic’s servers, so “they keep working when your laptop is closed”. They can use your connected tools (like Analytics), but they can’t see files on your computer. The most often they can run is once an hour. They’re on Pro, Max, Team and Enterprise plans.
Desktop scheduled tasks run on your own computer through the Claude app. They can reach your local files and tools, but they only run while the app is open and your computer is awake. If it was asleep at the scheduled time, you get one catch-up run when it wakes.
/loop in Claude Code repeats a prompt while a session is open. It’s for watching something for an hour or two, not for weekly reports. We cover it in part 1.

Anthropic’s own advice sums it up: use cloud tasks for work that should run without your machine, desktop tasks when you need local files, and /loop for quick checks during a session.

A weekly website check you can copy

Here’s the kind of prompt we’d give a weekly scheduled task, set for Monday morning:

Every Monday, read the last 7 days of Google Analytics and Search Console for our website and compare them with the 7 days before.
1. Report visits, signups and top 5 pages, with the change from last week.
2. List any search query where we’re in positions 5 to 20 with more than 50 impressions. Suggest one improvement for each.
3. Flag anything unusual: a page that dropped sharply, a spike from a new source, errors in Search Console.
4. Check the site homepage and contact form load and work.
5. Write it as a short report in our knowledge base, and save an email draft of the top three actions for me to read. Don’t send anything. Don’t change anything on the site.

Notice the last line. Our standing rule for every scheduled task is that it reads, reports and drafts, but doesn’t publish, send or change anything without one of us saying yes. We already run scheduled tasks this way. One updates our free skill pack on the 1st of every month: it does the work and commits it locally, then asks before it pushes anything public.

Getting scheduled tasks right

Start with one. A weekly report. Add more when you trust it.
Write the prompt as if for a stranger. Each run starts fresh with no memory of your chat, so say everything it needs: which site, which data, what to compare, where to save.
Make it save its findings somewhere. A file, a knowledge base, an email draft. A report you never see is no use.
Avoid the busy hours. Lots of schedules are set for exactly 9am, so pick an odd minute like 8:47.
Read-only access wherever the tool allows it.
Review it after a month. Is it telling you things you act on? If not, change the questions.

Mistakes to avoid

Putting analytics on but never setting up key events.
Waiting until launch day to add Search Console. Start collecting data early.
Giving an AI write access to your analytics or website “just in case”.
A scheduled task that emails or publishes without anyone checking.
Desktop tasks for things that must run every week, on a laptop that’s usually asleep.
Reports full of numbers with no “so what”.

The short version

Set up Google Analytics with key events, Search Console with your sitemap, and Cloudflare in front. Connect them to your AI, read-only, so you can ask questions in plain English. Then set a weekly scheduled task to read the numbers, tell you what changed and draft the actions, without touching anything. That’s the last piece of a website that’s built, not boosted.

This is part 5 of our Vibe coding a website guide. Back to the start: Vibe coding a website.

Last checked: 9 November 2026. These tools are changing quickly, and we update this guide when they do.

Sources

More on: Claude Scheduled tasks Vibe coding Website build Your data

See everything we have made