Documentation changelog

What changed in these pages, newest first. This is a changelog for the documentation, not for the product: it records what was written, corrected or retired here.

For whether a generated table is current, read its stamp rather than this page. Every generated reference carries the product commit it was read from, which is a better answer than a date on a changelog entry.

19 August 2026

These docs were written. /docs went from one page to a full set: getting started, concepts, every measurement and action section, the MCP server, the agency track, account and this reference.

The REST API claim was removed. The previous index page advertised "API reference: REST endpoints, authentication and rate limits". RankX AI has no public REST API, and that page was promising one. The MCP server is the programmatic interface, and it is now documented in full rather than mentioned.

Five reference surfaces became generated rather than written. The MCP tool tables, plan limits, the audit issue codes, the AI Readiness checks and credit costs are read from the product itself and stamped with the commit they came from. Hand-written versions of exactly these tables had already drifted twice inside the product's own documentation.

Search was wired. The search dialog had mounted on /docs since the day it shipped and had never worked.

Machine-readable twins were added. Every page here is now available as clean Markdown by appending .md to its URL or by sending Accept: text/markdown, and /docs/llms.txt indexes the tree. Generated tables render into the Markdown too, so a tool reference fetched as Markdown carries its tools.

Two pages were held rather than written. Dedicated setup pages for Google Search Console and Google Analytics wait on a change to RankX AI's Google consent screen. Until then a connection has to be reauthorised on a short cycle, and documenting the setup without saying so would be a promise the product cannot keep. The connections work, Traffic is documented in full, and the four availability states are correct regardless.

Three corrections were made while writing, all from the product's code:

• Scheduled checks run at most every seven days while an account is on trial, on every plan and every check family. A first-week page promising nightly runs would have been wrong for every reader it was written for. See check cadence.
• RankX AI does not report sentiment. It was designed and removed before it shipped, and naming its absence stops a reader hunting for a screen that does not exist. See metrics defined.
• One plan's audit cadence floor was documented wrongly in the plan these pages were written from, and was corrected against the migration that sets it.

Screenshots

None yet. Ninety-eight captures are specified with their intended page and their alt text already drafted, and they are recorded in the repository rather than left as broken references, because a missing image fails this site's build. They are added as they are taken.

How to tell whether a table is current

Every generated reference page carries a line naming the product commit it was read from and the date of that commit. That stamp is the answer, and it is more precise than a changelog entry, because it moves when the data moves rather than when someone remembers to write an entry.

The check that keeps it honest runs in the build: if a vendored file is edited by hand, or a tool is added to the product without a line being written for it here, the build fails rather than shipping a stale table.

Where to go next

• Start here, for the whole tree.
• Credit costs and plan limits, the two tables most worth checking the stamp on.