Lumineth
EN
Get in touch

AI SEO: AI agents analyse your site every week, you decide

Every week, an agent built on Claude Sonnet, Anthropic’s model, reads your site’s Search Console searches and your pages. It suggests new titles and descriptions for Google, and automatic checks verify every figure it quotes. You accept or decline: the Hub applies what you accept.

Lumineth Hub · Under the hood

  • Claude Sonnet (Anthropic)
  • Anthropic API
  • Google Search Console API
  • The Hub’s safeguards
  • Audit log
Lumineth Hub Proposals screen: two pages to improve for two Google searches, with sourced figures

Real screenshot of the Hub, demonstration data.

Concrete moves for Google, every week, without spending your evenings on it

Every week

Proposals ready to go

Titles and descriptions rewritten page by page, ready to apply.

Your real data

Real searches

The agent starts from what people type into Google before they reach your site.

One click

You decide

Accept or decline. The Hub applies what you accept, and nothing else.

Checked figures

Nothing invented

A quoted figure that is not in your data blocks the proposal.

How it works

A week of the SEO agent, from your data to your decision

  1. The Hub gathers

    Search Console searches, site pages, visit statistics.

    Search Console API · weekly task
  2. The agent analyses

    Claude reads these signals and drafts its proposals in a fixed format.

    Claude Sonnet · Anthropic API · structured output
  3. The safeguards check

    Figures present in your data, evidence, length, banned phrasing.

    Automatic checks
  4. You decide

    The proposal arrives in “Proposals” with its evidence: accept or decline.

    Proposals screen
  5. The Hub applies it

    What is accepted gets applied. Each decision is logged in the audit trail.

    Audit log

The agent only runs on sites where it is switched on. A proposal blocked by a safeguard can never be accepted.

Technologies

Under the hood, piece by piece

A well-known AI model, called directly by the Hub, with checks written in code around it.

  • AI

    Claude Sonnet (Anthropic)

    Reads your signals and drafts titles and descriptions, with its reasons.

  • Provider

    Anthropic API

    The Hub calls the model directly, with no intermediary tool.

  • Data

    Google Search Console API

    Reads the searches that bring up your pages, read-only.

  • Format

    Structured output

    The reply comes back field by field, checkable by code.

  • Checks

    The Hub’s safeguards

    Check figures, evidence, length and AI-style phrasing before showing you anything.

  • Traceability

    Audit log

    Keeps who accepted or declined what, and when.

  • Measurement

    Cookie-free audience measurement

    Adds page views to the analysis, without a consent banner.

This module’s AI: Claude Sonnet, Anthropic’s model, called through the Anthropic API. It proposes, it does not publish: every proposal passes automatic checks, then a person accepts or declines it. No personal data is sent to the model without your agreement, and never for a medical site. In the built-in help, Claude Haiku answers questions from the guides.

Shared foundation · PHP 8.2 · Twig 3 · MySQL, one database per site · admin React 19 + TypeScript · nginx HTTP/2 · monitored servers · encrypted nightly backups, copied off the server

A person at the end of every proposal

Activation

Starts with your agreement

The agent is off by default. It only runs once switched on for your site.

Figures

Anchored in your data

Every figure quoted must appear in the data read. Otherwise, the proposal is blocked.

Evidence

Every idea backed up

The proposal rests on searches and pages that actually exist, and shows them to you.

Style

No AI-sounding phrasing

A filter removes the phrasing typical of machine-written text, and the em dash.

Data

Your personal data protected

Nothing personal goes to the model without your agreement. Never for a medical site.

Log

Every decision logged

Accepted or declined: the decision is recorded in the audit log, with its author.

Frequently asked questions

How does an AI SEO agent work on my site?

Once a week, the Hub gathers your Search Console searches, your pages and your statistics. Claude Sonnet suggests new titles and descriptions, checks verify every figure, then you accept or decline them in “Proposals”.

Can the agent change my site without me?

No. It proposes, a person decides. The Hub only applies what has been accepted, and the decision is logged.

Which AI model is used?

Claude Sonnet, from Anthropic, called directly by the Hub through the Anthropic API.

What does the agent actually suggest?

Page titles and descriptions for Google, based on real searches. A monthly report of your site’s figures can also be switched on.

Is my data sent to the AI?

SEO signals, yes: searches, pages, statistics. No personal data without your agreement, and never for a medical site.

Is access to Google Search Console required?

Yes. The Hub reads Search Console through Google’s API, read-only. I connect your property to the Hub.

Let’s talk about your SEO

Tell me which site you want to move up in Google. I will get back to you within 24 working hours.