Why We Need CORD-AEO

People ask me why we need CORD-AEO. The answer is easier to show than to explain.

Open ChatGPT. Or Claude. Or Perplexity. Ask any of them a specific question about your business. If your business is small or new, half the answer will be wrong. Wrong prices. Wrong hours. Wrong feature descriptions. Wrong names of the people who work there. If your business is large enough to have deep coverage across the internet, less of the answer will be wrong, but you have to read more carefully to spot the errors that remain.

That answer is what your customers are being told about your business. That answer is where this story starts.

Why AI Engines Get Your Business Wrong

AI engines do not read your website the way Google reads your website. They do not rank pages against queries and return the best match. They read pages, absorb the language, and generate an answer in their own words. Whatever survives that translation is what the customer hears.

Everything on your site is source material for the paraphrase. Marketing copy, feature lists, team page, pricing page, support docs, press mentions, a five-year-old blog post that still ranks for something. All of it goes into the blender. What comes out is not what you would have said. It is what the engine's summary of what it read said.

When the engine is right, this works. When the engine is wrong, you find out the way you find out about a bad review. Someone tells you, or nobody does and you never know.

What We Already Have

The tools that exist today were built for the search era, not the answer era. They still matter. They also still fall short of the job.

SEO. Optimizes your pages to rank in Google's results. A ranked page is a page a human might click. AI engines do not click. They read.

Schema.org markup. Adds structured metadata to your pages so crawlers know what they are looking at. This is a Product page. That is a Review. This is an Article by that Author. Useful for crawlers. Still not the same as telling an AI engine what your canonical answer to a common question is.

Sitemaps. Tell crawlers what pages exist on your site. Useful for discovery. Say nothing about which pages carry the answers you would want quoted.

robots.txt and llms.txt. Control which crawlers can see which pages. Necessary for some things. Do not shape what any crawler that does read your pages ends up saying about you.

Each of these does its job. None of them does the job that matters when an AI engine is answering a question about you.

What None of That Solves

Every tool above operates on the page. AI engines operate on the answer. The gap between "here is a page about our pricing" and "the answer to a customer's question about our pricing" is where AI engines make up the difference themselves. Sometimes they do it well. Sometimes they invent.

An AI engine asked "does this business offer weekend service" has to decide the answer from whatever it read. It reads your homepage, your services page, your FAQ, and a Yelp review from 2022 that says you used to be closed Saturdays. It has no way to know which of these should carry more weight. It has no way to know that the Yelp review is stale. It has no way to know that your official answer to that question is on your Contact page and reads "Yes, we are open Saturdays 9-3."

The engine averages. The average is wrong.

What an Answer Engine Actually Needs

An answer engine needs three things your pages do not carry natively.

A canonical answer. Not a marketing paragraph that includes the answer. The answer itself, in the exact form you want it quoted. Short. Direct. Attributable.

Required claims. The facts inside the answer that must survive any paraphrase. If the engine reformulates your answer and drops one of these, the answer is no longer yours. It is a paraphrase that lost the parts that mattered.

Freshness metadata. When the answer was written. When it was last verified. When the engine should consider it possibly stale and refresh. Without this, a 2019 blog post carries the same weight as a page you updated yesterday.

None of the tools listed above carry these three properties. They were not designed to. The internet those tools were built for was one where humans read pages and made their own decisions. The internet AI engines read is one where the engine reads for the human and decides on their behalf.

What CORD-AEO Adds

CORD-AEO is a small addition to your site. A folder called /.well-known/cord-aeo/ containing an index and a set of JSON envelopes. Each envelope is one question and one canonical answer to it, plus the metadata that makes the answer verifiable and attributable.

Here is what an envelope looks like. This is not simplified. This is the actual shape.

{
  "envelope_id": "env-service-hours-01",
  "profile_version": "0.1",
  "canonical_question": "What are your service department hours?",
  "canonical_short": "Monday through Friday 7am to 6pm. Saturday 9am to 3pm. Closed Sunday.",
  "required_claims": [
    "Saturday hours are 9am to 3pm",
    "closed on Sunday"
  ],
  "published": "2026-06-01",
  "last_verified": "2026-09-15",
  "valid_until": "2027-03-15",
  "authority": {
    "publisher": "Example Motors",
    "publisher_url": "https://examplemotors.com"
  }
}

Nothing in there a business owner cannot read. The question the customer might ask. The answer you want the engine to give. The two facts inside that answer that must not get lost. When the answer was last checked. Who is responsible for it.

Publish the envelope. The engine finds it. The /.well-known/ path is discoverable by design. The engine reads the canonical answer. When a customer asks about your Saturday hours, the engine's paraphrase carries your required claims through. If the engine drops one, the Envelope Fidelity Score reflects it, and you have evidence to bring back to whoever built the engine.

Why This Has to Be a Standard, Not a Tool

Every AI engine could invent its own version of this. Some of them are trying. The result would be that a business has to publish structured answers in Google's format, Anthropic's format, OpenAI's format, and Perplexity's format, all separately, and update all of them every time an answer changes.

That is not sustainable. Nobody does it. What ends up happening is nobody publishes structured answers anywhere, and every engine keeps guessing.

CORD-AEO is Apache 2.0. Open specification. No vendor owns it. Any engine can read it. Any business can publish it. One folder on your site works across every engine that supports the standard. The engines that do not support it yet get better source material to work from than they did before, because the content on your site is more structured than it was.

Standards work when the alternative is worse than adopting the standard. For AI engines quoting businesses, the alternative is what we have now.

Where to Start

The first envelope is the one that matters most for your business. Not all of them. One.

Pick the question your customers ask the most. Write the exact answer you want quoted, in one or two sentences. List the two or three facts inside that answer that must not get lost in paraphrase. Publish it to /.well-known/cord-aeo/env-your-question.json. Add it to the index at /.well-known/cord-aeo/index.json.

Test it. Open ChatGPT, Claude, and Perplexity. Ask each one your canonical question. Compare the response against your canonical answer. If the required claims survived, the envelope is doing its job. If they did not, either the engine has not read your envelope yet or the engine is still overweighting older source material. Both are diagnostic. Both give you something to act on.

The reference envelopes and the validator are at cordspec.org/aeo. Copy one. Point it at your data. Read the score. That is the whole loop.

The businesses that publish this now will be the ones AI engines quote correctly six months from now. The businesses that do not will keep finding out about wrong answers the way they find out about bad reviews.

Structure before tools. Strategy before speed. Systems before scale.

AI engines are quoting your business today. The only question is whether they are quoting what you would have said or what they made up.

The AI Operator Framework is the operator playbook for shipping AI work that compounds. Read the framework here or take the course. Read the CORD-AEO specification at cordspec.org/aeo.