Skip to content
aisays

Answer Engine Optimisation

The AEO audit

Twenty-four weighted checks across seven signal categories, each one returning a status with the evidence behind it. This is the part of AI visibility that is fully under your control.

What Answer Engine Optimisation means

An answer engine does not want your page. It wants one paragraph out of your page, phrased in a way it can reuse, attached to something it can cite. AEO is the practice of making sure that paragraph exists, is easy to locate, and is attributable to a business whose identity is unambiguous.

In practice that splits into three jobs. The first is access: the page has to be retrievable and permitted. The second is declaration: the facts about your business — legal name, location, services, the questions you answer — have to be stated in a machine-readable form rather than implied by layout and imagery. The third is shape: the prose has to put the answer first and carry headings that mark where each answer begins.

None of this is speculative. Every one of the twenty-four checks below measures something that is observably read by search and AI crawlers today. That is deliberate: AEO is the part of AI visibility where the mechanism is known, which is why it is the first thing we audit and the first thing worth fixing.

Why AEO is only 10% of the score

Because being usable is not the same as being used. A flawless site can still be absent from answers if nothing independent corroborates it and competitors dominate the sources a system retrieves. AEO readiness removes the obstacles on your side of the fence; the remaining 90% of the AI Says Score measures what actually happens in the answers themselves.

The four statuses

  • PassFull credit

    The signal is present and within the range that makes it useful. Nothing to do.

  • WarningHalf credit

    Present but weak, or absent in a way that is defensible for your kind of business. Worth clearing, rarely urgent.

  • FailNo credit

    Absent, or actively working against you — a noindex directive, a blanket robots.txt block, no structured data at all.

  • Not testedExcluded

    The evidence could not be retrieved. Removed from the calculation entirely and reflected in your confidence level instead.

How readiness is scored

Every testable check contributes its weight multiplied by its status value — 1 for Pass, 0.5 for Warning, 0 for Fail. That total is divided by the sum of the weights of the testable checks and expressed out of 100. Untested checks are removed from both sides of the division, so a site we could only partially read is scored on what we could actually verify rather than penalised for what we could not.

AI Says Scoring Methodology v1.0

Signal categories

The seven categories we audit

Each check belongs to exactly one category, and each category answers a different question about how legible your site is to a machine.

  1. 01Crawlability

    5 checks · 31% of readiness weight

    Can a machine reach the page and is it allowed to use it?

    Reachability and response time, canonical declaration, robots.txt directives, XML sitemap presence, and whether the page carries a noindex instruction in a meta tag or an X-Robots-Tag header. This category carries the most weight for a blunt reason: every other signal is worthless if retrieval fails or indexing is forbidden.

  2. 02Metadata

    3 checks · 13% of readiness weight

    Is there a clean, correctly-sized summary of what this is?

    Title tag length and substance, meta description length and substance, and Open Graph tags. We check ranges rather than presence alone: a title under about twenty characters rarely states who you are and what you do, and one over sixty-five is likely to be truncated when quoted.

  3. 03Structure

    4 checks · 15% of readiness weight

    Can a machine find the section worth quoting?

    A single descriptive H1, a real H2 hierarchy with enough subheadings to create distinct quotable sections, internal linking that expresses how your pages relate, and image alt text. Four or more H2s is the threshold where a page reliably offers more than one extractable section.

  4. 04Schema

    6 checks · 28% of readiness weight

    Have you stated plainly what you are, or left it to inference?

    JSON-LD presence, then Organization, LocalBusiness, Product or Service, FAQ and Breadcrumb types specifically. Structured data is the only part of a web page where you declare facts in a machine-readable form instead of implying them in prose. Organization schema carries the highest weight in this category.

  5. 05Entity

    2 checks · 8% of readiness weight

    Is there a primary source for who you are and how to reach you?

    An About page linked from the homepage, and discoverable contact details — a contact page, an email address, a phone number or a postal address. About pages are disproportionately used as the basis for AI descriptions of a business, and contactability functions as a trust signal in its own right.

  6. 06Answerability

    3 checks · 15% of readiness weight

    Is this written in a shape an answer can be taken from?

    Question-style headings, answer-first writing, and overall content depth. The answer-first check measures the length of the first paragraph after your first H2 — a useful proxy for whether you state the answer immediately or make a reader wade through preamble. Machines do not wade.

  7. 07Authority

    1 checks · 5% of readiness weight

    Does anything independent corroborate your claims?

    sameAs profile links declared in your structured data, pointing to authoritative profiles elsewhere. Three or more independent profiles is the threshold for a Pass. This is the narrowest category on the site audit because most authority signals live off your own site and are assessed under Citations & Authority instead.

Weight shares are approximate and rounded — they describe each category’s portion of the readiness component, not of the overall score. See the full list of individual checks.

Sequencing

The order to fix things in

Some AEO fixes unblock others. Doing them out of order wastes effort on work that cannot take effect yet.

  1. First

    Unblock retrieval

    Any Fail in crawlability. A noindex tag, a disallow-all robots.txt or an unreachable homepage makes every other improvement invisible. This is usually a single-line change and it is always the first thing to do.

  2. Second

    Declare your identity

    Organization schema, then LocalBusiness if you serve a geographic market, then sameAs links to profiles that corroborate you. Until a machine can resolve who you are, it cannot confidently attribute anything it reads to you.

  3. Third

    State what you sell

    Product or Service schema, plus explicit service and location statements in body copy. Inference from marketing language is the most common source of inaccurate AI descriptions.

  4. Fourth

    Make the page quotable

    Question-style headings, an answer in the first paragraph after each one, FAQ schema where you genuinely answer questions, and enough subheadings that a page offers several distinct extractable sections.

  5. Fifth

    Tidy the metadata

    Title and description lengths, Open Graph tags, canonical declarations, alt text. Individually small weights, collectively a few points, and cheap to complete once the substantive work is done.

Scope

What the AEO audit does not cover

AEO is the controllable layer. Three things sit outside it, and each has its own page.

  • Retrieval and citation

    Whether generative systems actually fetch and cite you depends on sources beyond your own site. That is the GEO question, and it is far less well understood.

    Read more
  • What answers actually say

    Observed presence, recommendation and accuracy across platforms are measured in the full report, not in a site audit.

    Read more
  • Change over time

    A site audit is stable between deployments. AI answers are not, which is a separate measurement cadence altogether.

    Read more

Questions

Questions about AEO

Audit your answer-engine readiness

All twenty-four checks run in the free assessment, with the evidence shown for each one.