Can an AI assistant quote you at all, and how easily

Before it matters how well a page is written, three things decide whether an answer engine may use it at all. A snippet directive such as nosnippet can forbid any preview, so there is nothing to quote. robots.txt can refuse the search crawlers, like Googlebot and Bingbot, or the answer crawlers, like OAI-SearchBot, ChatGPT-User, PerplexityBot and Claude-SearchBot. And the HTML your server sends can arrive nearly empty, with the page appearing only after a browser runs its scripts.

One closed gate makes the verdict Blocked, whatever else the page does well. Your site’s verdict is the worst verdict of any of its pages, never an average. One blocked template is exactly the thing worth saying, and averaging it with forty healthy pages would hide it. A blocked Googlebot goes to the top of your week. A closed PerplexityBot is a choice many sites make on purpose, so the plan treats it as one.

The gates are part of every reading, on every plan, including the free one. The public scan on the Citability page checks the snippet and robots gates on any address, without an account. For the third, it tells you how many words a fetcher without JavaScript reads, and says plainly that the rest was not measured.

Past the gates, the score is a weighted sum of sections. Each section asks a question your page either answers or does not.

The weights add up to 110, and that does not matter. Only sections that could be measured count. A check that could not run is left out of the number, not counted as failed, so a site without Search Console is not marked down for four checks it was never asked. The same reading always scores the same: nothing in the score reads the clock, chance or the network.

• Gates (40): snippet directives, crawler access, whether the served HTML carries the page.
• Rendering (25): how much of the page exists before scripts run.
• Crawler access (15): the edge, the firewall, per-agent refusals.
• Structure (12): headings that say what follows, answers close to the question.
• Substance (8): figures, dates, attribution, something worth quoting.
• Structured data (5): markup that agrees with the visible text.
• Media and links (5): alt text and outbound sources.

On a failed check you can ask for a draft of the fix. A model writes it, and code decides what survives. Every number in a draft must be one your page prints or carries in its address. Every edit must point to a page from this reading. An edit that changes nothing is dropped.

Text that can be written from the page’s own words gets a before and an after. Everything else gets steps. The first real draft on a live site showed why. Asked to fix five dates in structured data, the model invented no date and proposed blanking all five. A date your pages do not print sits in a CMS we have never seen, so it belongs in the steps.

What passed stays, and what did not is listed. A draft with nothing left says so, instead of coming back empty.

Changed something? Press Read it again on that check. It re-runs that single check on the live pages, so you can confirm a fix without waiting for the next reading.

It is the same machinery as marking a card fixed, asked the other way round: check first, before anything is claimed. It shares your workspace’s daily verification budget, because each re-check is a small crawl of your server. A re-check that finds nothing closes the card on what was read.

On the Agency plan, Citability across your sites is one portfolio, worst first. Blocked pages decide the order, and only then the score, because blocked pages are what decides the week.

One site’s score is a number you read once. The scores of fifty sites are a queue you work through. That is why the portfolio sits on Agency.

Citability reads every page of every reading. There is no toggle and no upload. The module, from the Business plan, unlocks the score, the section breakdown, the trend, the check list with its drafts and re-checks, and fan-out.

Fan-out asks what an assistant would search for next after your page, and checks whether the page covers those follow-up questions. It is the one part with a cost outside the reading, so it has a budget per reading: 20 follow-up queries on Business, 40 on Agency. There is no counter to watch and no line on the invoice.

Without the module, the findings are still counted. The reading says how many it found and in how many kinds of check, and keeps the names to itself. Turning the module off does not change your Monday briefing, because its cards were never in it.

Answer engines publish very little about how they read. So Citability stays inside what can be measured on your own pages, and says so where it cannot.

• No answer engine publishes whether its fetcher runs JavaScript. The rendering checks measure what your server returns, nothing more.
• The edge probe sends one request per user agent, to your home page, once per reading. It can catch a firewall rule aimed at a named bot. One request is a hint, not proof, unless your server logs agree.
• The llms.txt card says only whether the file exists. It does not tell you to create one: an analysis of a large sample of domains found no link to citations.
• It does not monitor what AI engines say about your brand, serve separate Markdown versions for bots, or advise chopping content into micro-paragraphs.
• Structured data is checked for agreeing with the visible text. It is not sold as a citation lever, because there is no evidence that it is one.