Connect a site

A site starts as a domain and a crawl. Every connection after that turns a description of the site into a measurement.

This page lists each connection and what it reads. It says what you hand over, and what happens when it is missing.

The crawl

You need nothing but the domain. It runs as soon as you add the site.

The crawl reads your sitemap, walks the navigation, and probes the paths a site usually has. Per page it extracts the title, the meta description, the headings and the body text. It also takes image alt text and the links in and out.

It also detects what the site is built on. That detection seeds the brand profile as suggestions, and it decides which shipping path is available to you later.

Free currently crawls up to fifty pages per refresh. Pro has defined crawl allowances. Pages beyond the cap or behind an error are counted as skipped.

If the crawl finds no sitemap, it falls back to navigation and probes. You get fewer pages and the first report says so in its missing-data list.

Search Console

This is the connection that matters most. It is a measurement source, and the loop cannot ship an experiment without one.

You need a Google account with access to the verified property. Open settings, then integrations, and click connect on the Search Console card. Approve read-only access in Google’s own consent screen.

The guided flow on the integrations page handles connecting, because the production Google OAuth client is not created yet.

Sitearc then reads clicks, impressions, position and query data for the property, plus URL inspection results. Those values carry the measured tier, the highest one on the ladder.

Nothing is asked of you beyond the sign-in. There is no key to paste, and the grant is revocable from your Google account or from the card.

When it is missing, three things happen. The first report drops that source into its missing-data list. Values that would have been measured fall to a lower tier. ship_experiment refuses the change, because no measurement source is connected to read the result.

Google Analytics

Same flow, same card stack, a separate grant. Disconnecting one does not disconnect the other.

Analytics adds sessions, landing page performance, traffic acquisition, key events, ecommerce performance, site search and audience breakdown. It also reports its own measurement health. A broken property shows up as broken rather than as a low number.

When it is missing, Search Console still measures clicks and position. You lose the conversion side of an experiment, so a proposal predicting revenue has nothing to prove itself against.

GitHub

Today the GitHub connection is read-only, and it runs on a single access token. It answers two questions: which preview deployment belongs to a pull request, and which commit a merge produced.

Create a fine-grained personal access token scoped to the repositories whose sites you run the loop against. Grant exactly four permissions, all read-only: metadata, pull requests, contents and deployments.

Do not grant write access. Nothing in Sitearc pushes a branch, commits a file, opens a pull request or posts a comment. The agent you work with opens the pull request using its own credentials, not ours.

When it is missing, resolve_preview_url cannot find the preview deployment. You can still merge by hand, and the audit step becomes yours to run.

The GitHub App is being built. It will hold a per-installation token, fetched per job. Permissions stay read-only, plus pull request write. It will replace the single token.

WordPress

The Sitearc Connect plugin is being built and is not available yet. See the WordPress plugin doc for the endpoints and the credential model.

Until it ships, a WordPress site is still fully readable. The crawl, Search Console, Analytics, ranks, backlinks, competitors and the audit all work against any site on any stack. What is missing is a write path that does not go through a repository.

Local presence

Nothing to connect. Business Profile data comes through the research stack, keyed by the business name and location you give it.

You get the profile, its reviews, its posts, its questions, its categories, and a rank grid across a geography. Those are point-in-time captures, so they carry the observed tier rather than the measured one.

Rank tracking

Nothing to connect, and it is not free. You create a tracker, add keywords, and run it on demand.

estimate_rank_tracker_cost tells you what a run will spend before it spends it. Current free capacity is twenty-five tracked keywords per site. Pro has defined allowances.

The brand profile

Being built. One versioned document per site. It holds colours and type by role, positioning, audience, tone, vocabulary and words to avoid.

It will be seeded by the connect crawl as suggestions. Agents will propose patches with a mandatory summary, and a person applies them.

What a missing source actually costs

Sitearc never substitutes one source for another. A keyword tool answers a different question at a different tier. None of them can measure a change you made to your own site.

When a source is missing, the platform names it and names the concrete fix. That is the whole behaviour. It does not fill the hole with an estimate and hope you do not check.