Your customers are asking AI.
Is your business in the answer?

Googlebot gets a 200. ClaudeBot gets a 403. Same page, same second.

We asked nytimes.com for its homepage as each AI crawler in turn. The 3 that feed search were served. All 5 that feed AI assistants were refused at the edge before they saw a byte. None of those refusals is in their robots.txt, so a checker that reads that file will tell you all 5 are welcome.

nytimes.com · fetched as each crawler

measured 4 August 2026

Googlebot200Bingbot200Applebot200GPTBot403OAI-SearchBot403ClaudeBot403Claude-SearchBot403PerplexityBot403

our control200a person in Chrome200

200 means the page was handed over. 403 means the door was shut, by the network sitting in front of the site, before the page itself was ever consulted. That is why no edit to robots.txt undoes one.

A neutral client that identifies itself as ours was served too, and so was an ordinary browser. That is what tells us the rule keys on the crawler's name rather than on us being unfamiliar, and it is the difference between a finding and a guess. See every request, the method, and what a spoofed fetch cannot prove.

re-run this yourself takes about five seconds, no signup

THE PROBLEM

AI cannot recommend what it cannot understand.

People ask AI what to buy, who to hire and who to trust, and a strong business can be left out of the answer simply because an assistant could not work out what it does. The goal here is not to manufacture authority. It is to make the authority you already have easy for a machine to recognize.

Three questions decide whether you make it into the answer, and each one is settled by a different part of your site.

LEGIBLE › VISIBLE › TRUSTED › RECOMMENDED

See what AI can reach, fix what it cannot read, then measure whether assistants start citing you.

The whole path, start to finish. Remy walks the same three steps below, slowly and with the stakes attached.

01 · SEE

What AI actually reads

Paste your address. We ask the AI crawlers for your page the way they would, and tell you what each one got back. Free, no signup, about 5 seconds.

8 crawlers fetched · 5 retrieval gates · 8 headline checks · one score out of 100

02 · FIX

One file that controls access, one that helps them read you

You get the actual files, not a report telling you to get them. For most sites it is two small files and one clearer paragraph, and none of it touches your design.

robots.txt · llms.txt · JSON-LD · answer layer · paste it, or let the free MCP server write it into your repo

03 · CITED

Whether it actually worked

We ask the assistants the questions your customers ask, every week, and tell you whether they say your name or somebody else's.

ChatGPT · Claude · Gemini · Perplexity · tracked over time. The score is the input. Being cited is the outcome.

What you get, in plain terms

The story above is one business. This is the list. Every number here is computed by code you can read, and a check fails the build if this page and the code stop agreeing.

• Crawlers that prove who they are

  Anyone can put GPTBot in a user agent. We check the address against the ranges the operator publishes, and against forward-confirmed reverse DNS where they document one. A request is verified, outside the published range, or unverifiable, and nothing is called an impostor without evidence that it is one.

  18 crawlers tracked · 10 with a published range feed · 3 with a reverse DNS convention

• How much of your AI traffic is real

  Read your server log and we report the share of requests claiming to be an AI crawler that genuinely came from one, and how that compares with everyone else we measure. The log is parsed on your own machine and never leaves it. What travels is two integers.

  your rate · the median across the corpus · nothing about which pages were hit

• A score that says what it could not measure

  Forty-nine weighted checks across six dimensions. Anything we could not measure is named and left out of the total rather than counted against you, so the number is over what was actually looked at.

  49 checks · 100 points · 6 dimensions · scored over what we could measure

• Off-site reputation, measured by asking

  Reviews, directories, forums and press live on other people's platforms, so no page audit can see them. We ask the assistants instead and read the pages they cite. A confident answer with no sources counts as unknown, because a model will happily describe a business that does not exist.

  7 questions · 17 of the rubric's points · verdicts read from citations, not prose

• The fixes, written for you

  robots.txt, llms.txt, JSON-LD and a rewritten opening, as files you can commit. Facts we cannot verify ship as bracketed placeholders: it will not invent your price, your review count or your founding year.

  4 generated files · brackets where a real number belongs

• Yours to run, with or without us

  The audit engine, the crawler verifier and the log reader are MIT licensed on npm, and the MCP server puts them inside Claude Code or Cursor. Eight of the tools need no account at all.

  15 MCP tools · 8 that run entirely on your machine · MIT licensed

What the fix looks like, start to finish.

A composite business, invented to show the steps. Illustrative, not a customer result.

See the whole walkthrough: the three findings in full, the files that fixed them, and what changed over twelve weeks.

Point your coding agent at your site. It writes the fixes into your repo.

OpenAEO ships an MCP server. Add it to Claude Code, Cursor, or any MCP client, and your assistant can audit the live site, generate robots.txt, llms.txt and JSON-LD, and write them into your project as files you review like any other diff. It is open source, MIT licensed, and needs no API key and no account.

The generated files carry bracketed placeholders where a real number belongs, and the server instructs the agent never to invent one. If it does not know your refund window or your review count, it asks you or leaves the bracket. A confident fabrication is the thing this product exists to catch.

What answer engine optimization actually is

Answer engine optimization (AEO), also called generative engine optimization (GEO), is the practice of making a website structured and specific enough that an AI assistant can read it, understand it, trust it, and name it in an answer, the way search engine optimization once made a page rankable in a list of links. OpenAEO audits a site against five retrieval gates and eight headline checks in seconds, then returns the exact files and copy changes that can move it above the ship floor. The dashboard tracks those results over time, so you can see which way they are moving. A failed retrieval gate caps the score. If a page depends on JavaScript, blocks AI crawlers at the CDN, disallows bots in robots.txt, or contains unresolved placeholder text, an assistant may not be able to fetch, parse, or cite it at all. The audit is free for one site, requires no signup, and runs in seconds.

What the score is made of, and what it cannot tell you.

One OpenAEO audit covers both AEO and GEO. No agreed standard exists for scoring this yet, so these are our checks and our weighting, published in full and open source so you can disagree with them specifically rather than take the number on faith. The retrieval gates come first because they cap everything else. A site can look crawlable in robots.txt and still return a 403 response to every AI bot at the CDN. OpenAEO fetches your pages as each bot, from the edge, and checks what actually comes back.

• GATE 01
  Renders without JavaScript
  The important text exists in the HTML a crawler receives, not only after a browser bundle runs.
• GATE 02
  Bots reach the edge
  Live per-bot fetches reveal whether CDNs and firewalls are blocking AI crawlers.
• GATE 03
  robots.txt allows
  Current AI agents are checked against an updated registry rather than a stale copy-and-paste list.
• GATE 04
  Indexed where it matters
  Several assistants rely on search indexes, including Bing. If the site is absent, it may remain invisible.
• GATE 05
  No unresolved placeholder text
  Verification markers and unfinished live copy cap the score. Drafts do not receive a passing grade.

Start free, on one site.

The instant audit runs on any URL you paste, on every plan, as often as you like. The number in each plan is how many sites it tracks: saved history, drift alerts, and runs you can compare.