What does the check look at?

Five things an answer engine needs before it can cite a site: whether its crawlers are allowed in robots.txt, whether there is an llms.txt map, whether schema names the entity, whether the sampled pages open with a quotable answer and carry tables, and whether pages expose dates and name sources. It reads the homepage, robots.txt, llms.txt, the sitemap and ten pages.

Does it measure traffic from ChatGPT or Perplexity?

No. Traffic lives in your analytics. This check measures whether the site is readable and quotable in the first place, which is the part you control. The OperStack plugin for Claude Code has an ai-visibility command that reads the GA4 side.

We keep a line for each check: the address checked, the score, and which link brought you here. We do not keep your IP address, and we do not sell or share any of it. Your email is kept only if you leave it to get the fix, and every email we send has one-click unsubscribe.

Why did my site score low on content when the pages are long?

Length is not the signal. The check looks for an answer-first paragraph of 20 to 90 words with a figure right after the H1, at least three H2 sections and a table. Long pages that bury the answer score low because engines quote the opening, not the middle.

What is the difference between this and the audit?

This is the free, automatic version of the first two pages of the audit, and it takes about a minute. The audit reads a hundred pages instead of ten, scores six areas instead of five, puts five rivals beside you, and hands over finished files: the llms.txt map, the Organization markup and fixed titles.

AI Visibility Check: Can AI Assistants Cite Your Site?

Can AI assistants read and cite your site?

Enter a site. In about a minute you get a score out of 100 across the five things an answer engine needs before it quotes a page, and the three fixes that move it most. Then ChatGPT answers three questions a buyer would ask: you see whether it names you and who it names instead.

Other tools check one page. We check ten.

Public signals only. We log the address checked and the score, never your IP.

Our own site scores 93 of 100 on this same check, measured 2026-10-27 across 10 pages. Put oper-stack.com in the box above and see for yourself: a tool that grades your site should be willing to say what it gives its maker. The number is measured on every build, never typed, so it drops the day we let something slip.

Prefer to do this inside Claude or ChatGPT?

If you already work with Claude, ChatGPT or Cursor, they can run this check for you, in the window you are already in. You write in plain words: look at my site, compare me with these three, how many AI visits did I get this month.

https://oper-stack.com/api/mcp/

Paste it into your app's settings once. Nothing to install, no OperStack account, no payment. The slash at the end matters: without it some apps do not follow the redirect. In ChatGPT this needs a paid plan with developer mode on.

Step-by-step setup for each app →

Where these rules come from

+41%More often a page is picked up when it carries a quote with the speaker namedAggarwal et al., GEO, KDD 2024Position-Adjusted Word Count

+32%What a concrete figure adds, instead of a general claimAggarwal et al., GEO, KDD 2024same metric

44.2%Share of ChatGPT citations taken from the first third of a pageKevin Indig1.2M answers, 18,012 citations

GEO can boost visibility by up to 40% in generative engine responses.

Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, DeshpandeGEO: Generative Engine Optimization, KDD 2024

What the check measures

Access, 25 points. robots.txt read per user agent for fourteen AI crawlers and fetchers. A blocked search fetcher costs more than a blocked training crawler, because the fetchers are the bots that cite pages live. For the full robot-by-robot list, firewall test included, use the free AI crawler checker.

Agent map, 15 points. llms.txt exists, is text, and links to the site's own pages rather than a template demo. The llms.txt validator follows each link in yours; with no file yet, generate a draft llms.txt from your own pages.

Entity, 20 points. Organization or business schema, FAQPage, a page-level type (Article, Service, Product, Course and the like), a canonical tag and Open Graph on the homepage. To read every type on one page, broken blocks included, use the schema markup checker.

Quotable content, 25 points. On the ten sampled pages: at least 300 words, an answer-first paragraph with a figure right after the H1, three or more H2 sections, a table. If a page looks empty to the check, see what it holds without JavaScript.

Dates and sources, 15 points. Publication dates in meta or schema, source phrases in the text, a sitemap with lastmod.

Who ChatGPT names, not part of the score. After the score we ask ChatGPT, with web search on, three questions a buyer of a business like yours would ask, and show whether it names you and who it names instead. If a checked page lacks a short answer at the top, you also get a ready-to-paste opening paragraph for it, written only from what your own pages say. Three of these a day per connection.

What it does not measure

Traffic or citations already happening. This check asks whether a machine can read and quote you, not whether anybody arrived. To count the visits assistants already send you, put the free counter on your own site, or give them a channel in Google Analytics with a GA4 regex for AI traffic.

Rankings, backlinks or keyword demand.

Pages that need a login, or sites that block the checker itself.

Run the full gates on your own corpus

The check reads a sample. The sixteen gates read every page in your repository, in one command, for free.

npx @operstack/gates · What the gates check

Frequently asked questions

What does the check look at?: Five things an answer engine needs before it can cite a site: whether its crawlers are allowed in robots.txt, whether there is an llms.txt map, whether schema names the entity, whether the sampled pages open with a quotable answer and carry tables, and whether pages expose dates and name sources. It reads the homepage, robots.txt, llms.txt, the sitemap and ten pages.
Does it measure traffic from ChatGPT or Perplexity?: No. Traffic lives in your analytics. This check measures whether the site is readable and quotable in the first place, which is the part you control. The OperStack plugin for Claude Code has an ai-visibility command that reads the GA4 side.
Is anything stored?: We keep a line for each check: the address checked, the score, and which link brought you here. We do not keep your IP address, and we do not sell or share any of it. Your email is kept only if you leave it to get the fix, and every email we send has one-click unsubscribe.
Why did my site score low on content when the pages are long?: Length is not the signal. The check looks for an answer-first paragraph of 20 to 90 words with a figure right after the H1, at least three H2 sections and a table. Long pages that bury the answer score low because engines quote the opening, not the middle.
What is the difference between this and the audit?: This is the free, automatic version of the first two pages of the audit, and it takes about a minute. The audit reads a hundred pages instead of ten, scores six areas instead of five, puts five rivals beside you, and hands over finished files: the llms.txt map, the Organization markup and fixed titles.