How does an AI read your site?
CiteraHub measures whether your site is ready to be read by search engines and AI answer engines, applies the changes you approve, and measures the result again.
No account and no email address for the score. It reads a single page, so it can measure some of the fourteen dimensions; the result says how many it could measure.
What a report looks like
A dimension that could not be measured is not a zero and not a blank: it says why. That distinction is what the whole product rests on.
- Answer formatting92
- Structured data78
- Crawler access96
- Citability64
- Page speedNot measured: Core Web Vitals are not collected in this deployment.
- AccessibilityNot measured: the crawl does not read this dimension.
It never calls an unmeasured thing zero
Each of the fourteen dimensions is either measured or reported as “could not be measured”, with the reason. Counting an unmeasured dimension as zero shows the reader a problem that does not exist.
It never writes what your page did not say
A rewritten passage may not contain a number, date, price, award or superlative that was absent from the original. A suggestion that adds one is refused by name.
It writes nothing until you approve it
Every change is shown as a plan first. An approved change is opened as a pull request; nothing is ever pushed to the default branch.
Four steps, and the first three are worthless without the fourth
- 01
Measure
Fourteen dimensions. Each is either measured or says why it could not be.
- 02
Propose
Every finding carries the rule behind it and the primary document that rule was read from.
- 03
Approve
A change is shown as a plan first. Nothing you did not approve is ever written.
- 04
Verify
Measured again after it lands. What says it improved is the measurement, not us.
Three surfaces, one answer
The panel, the CLI and the MCP server call the same service. Wiring your own agent does not get you a worse audit than the one on the screen.
Panel
In the browser. Audits, reports, approvals and the history of every change.
CLI
In your own pipeline. Same audit, same report, with exit codes.
MCP
For Claude or your own agent. It returns what the plan covers and states what it held back.
Plans
No account is a plan too: you see the three most severe findings, and the answer says how many were not shown. A silently shortened list would contradict the reason this product exists.
Starter
One site, audited weekly
- 1 site, 8 audits a month
- Full report and evidence
- Change proposals
- 90 days of history
Professional
Brand tracking and automation
- 5 sites, 60 audits a month
- 12 AI measurements a month
- Connectors and MCP
- Applying changes
Enterprise
Teams with many sites
- 25 sites, 400 audits a month
- 120 measurements a month
- Sector benchmark
- 3 years of history
Open reference
- What the score measuresFourteen dimensions, the rules that score each, and the primary document each rule was read from.
- AI crawlersWhich one trains a model, which one makes you citable, and which arrives because somebody asked.
- SourcesThe documents the audits rest on, when each was last read, and which ones nobody here has checked.