Skip to content
SleekSite AI Docs
On this page

Analytics with AI & MCP

How the AI Assistant and AI clients like Claude or Cursor read your analytics, find what to fix, change your site, and check whether it worked.

Site Admins and Editors (AI Assistant); members with MCP access

The numbers on the Analytics page are also available to AI. The AI Assistant inside SleekSite, and any AI client you connect over MCP (Claude, Cursor, VS Code and others), can read them. Because the same AI can also edit your pages, forms and content, it can go from "what's wrong" to "here's the fix, ready to preview" in one conversation.

Ask, change, measure#

Improving a site with the AI Assistant and analytics Ask, change, publish, then measure the difference 1 Ask “Which pages lose the most visitors?” 2 Read site_analytics: sources, landing and form pages 3 Change Rewrites a headline, moves a button, shortens a form 4 Preview & publish You check it in Preview, then publish 5 Compare Two weeks before against two weeks after (from/to) 6 Keep or undo Keep what worked, or restore the old version AI Assistant or any MCP client
  1. Ask a question in plain words
  2. The AI reads your analytics: the same report as the Analytics page, narrowed to what it needs
  3. It changes your site: copy, layout, a button, a form
  4. You check it in Preview, then publish
  5. A week or two later, ask it to compare the days before the change with the days after
  6. You keep what worked, or put the old version back

Analytics is read-only to AI: it can't delete or change your numbers. Edits to your site follow the usual rules: nothing reaches visitors until you publish.

What you can ask#

How the site is doing

  • "How did the site do this month compared with last month?"
  • "Where do our visitors come from, and which source brings the most bookings?"
  • "Are we getting visitors from ChatGPT or Perplexity?"

What to fix

  • "Which landing pages have the highest bounce rate? Suggest a better headline and call to action for the worst one."
  • "Our pricing page gets lots of views but few enquiries. What would you change?"
  • "Do phones convert worse than desktop? If so, fix the mobile header so the call button is always visible."

Content and SEO

  • "Which blog posts bring visitors who go on to enquire? Write two more posts on similar topics."
  • "Which pages get no visits at all? Should we merge or remove any?"
  • "Most of our visitors are in Texas. Update the home page and SEO to mention the areas we serve."

Forms and flows

  • "How many people submit the Quote request form, compared with how many view the page it's on?"
  • "Shorten the Quote request form to the fields we really need, and add a phone link next to it."
  • "Compare Quote request submissions in the two weeks before 3 November and the two weeks after."

TipAsk for the reason as well as the change: "…and tell me what number you expect to move, so I can check in two weeks." You get a change you can measure, not just a guess.

How AI uses the data#

The AI uses the numbers the way a good marketer would:

It looks at To find
Landing pages with a high bounce rate Pages whose headline or first screen doesn't match what visitors expected
Conversion pages against their pageviews Pages people reach but don't act on, and forms that are seen but not sent
Traffic sources by conversion rate Channels that bring buyers, and ones that bring browsers
Devices by conversion rate Layouts, buttons or forms that are hard to use on phones
Top pages and exit pages Popular pages without a next step, and where visits end
Countries and cities Where to focus local wording, opening hours and service areas
Campaigns Which emails, ads and posts are worth repeating
Recent conversions Real paths that ended in an enquiry or order

Then it changes what it can reach: page copy and headings, calls to action and their placement, form fields and labels, links between pages, SEO titles and descriptions, and new pages or posts. It shows you the change in Preview before anything goes live.

For developers: the site_analytics tool#

The AI Assistant and MCP clients read analytics through one read-only tool, site_analytics. It's listed on every site's MCP server, and any member of the site who can connect over MCP can use it.

Argument What it does
range 7d, 30d (the default), 90d (weekly points) or 12m (monthly points). Starter keeps 30 days
from, to Exact UTC days (YYYY-MM-DD) instead of a range, to compare before and after a change
conversion Count only one kind: form, booking, queue, order, call, email or directions
form Count only one form's submissions, by its name
limit Rows per list, from 1 to 100 (about 10 by default)

conversion and form narrow every conversion figure and rate in the report, while visits and pageviews still count every visit. So a rate always means out of all visits.

The tool returns JSON plus notes: plain-language explanations of the figures, and anything the site's plan leaves out. An example, shortened:

{
  "plan": "full",
  "from": "2026-10-20",
  "to": "2026-11-18",
  "totals": { "visitors": 3482, "sessions": 3720, "conversions": 126, "conversion_rate": 3.4, "bounce_rate": 48 },
  "previous": { "visitors": 3110, "conversions": 104, "conversion_rate": 3 },
  "sources": [{ "name": "Google search", "sessions": 1640, "conversions": 61, "conversion_rate": 3.5 }],
  "landing_pages": [{ "path": "/order-cakes", "sessions": 410, "bounce_rate": 71, "conversion_rate": 1.2 }],
  "conversion_pages": [{ "path": "/contact", "conversions": 48, "pageviews": 620 }],
  "forms": [{ "name": "Cake order", "count": 39, "conversion_rate": 1 }],
  "notes": ["Covers 2026-10-20 to 2026-11-18 (UTC days); \"previous\" is the same length just before, for comparison. ..."]
}
Field Contents
totals, previous Visitors, pageviews, sessions, conversions, order value, conversion rate, bounce rate and average visit length, for the period and the one before
series One point per day, week or month, with its own conversion rate
top_pages, conversion_types, forms On every plan
sources, landing_pages, exit_pages, engagement, devices, countries, cities, browsers, campaigns Plus and Business
conversion_pages Where conversions happened, with each page's pageviews, so you can work out a per-page rate (Plus and Business)
recent_conversions The latest conversions, each with its form, page, source, landing page, device and value (Plus and Business)

Rates are percentages. If a client asks for more history than the plan keeps, or the site has analytics turned off, notes says so.

A good prompt for an MCP client: "Use site_analytics for the last 90 days. Find the landing page with the most visits and the lowest conversion rate, read that page with get_resources, and propose changes to its headline, first section and call to action. Save them, build, and give me the preview link."

Was this page helpful?

Last updated November 18, 2026