What is measured, what lands in the report, and the difference between a score and a fix list.
An AEO audit tests three things in order: whether AI assistants can reach your pages, whether there is anything on them worth quoting, and whether the model knows who you are well enough to name you.
Most things sold as an AEO audit return a number. A number tells you that something is wrong. It does not tell you which thing costs you the most citations, or what to do on Monday morning. The difference between those two outputs is the whole value of the exercise.
A citation requires a model to clear three separate gates. An audit that tests one and reports confidently on all three is guessing.
| Gate | What is checked | What failure looks like |
|---|---|---|
| Retrieval | Named AI agent permissions in robots.txt, HTTP status and render for each agent, JavaScript dependency, llms.txt | The model wants to cite you and cannot open the page |
| Extraction | Whether pages carry short, self-contained answers to the questions buyers actually ask; heading structure; tables and definitions | The page opens fine and contains nothing liftable |
| Attribution | Organization and Person schema with stable identifiers, naming consistency across the web, collisions with similarly named entities | Your words get used, your name does not |
Gate one is binary and cheap to fix. Gate three is slow and compounding. Ordering the findings by which gate they sit behind is what turns a list of problems into a plan.
AI agents split into two classes. Training crawlers — GPTBot, ClaudeBot, Google-Extended, CCBot — affect whether a model learns your brand exists over time. Retrieval agents — OAI-SearchBot, ChatGPT-User, PerplexityBot — fetch your page while the model answers a live question. Blocking retrieval is the expensive mistake. Blocking training is common, usually accidental, and far less damaging. An audit that reports “AI crawlers are blocked” without naming the agent and its class has not done the work. The full distinction →
Send a domain and you get a report back inside 48 hours. No call, no form-filling beyond the domain and an email, and the report is yours whether or not you ever reply.
What it deliberately does not include: multiple engines, repeated runs per prompt, a competitor benchmark, or the order of work. Those belong to the paid measurement protocol, and the reason is bluntly practical — the full version does not scale across everyone who asks.
| Free check | Full audit | |
|---|---|---|
| Buyer questions | 5 | 20+, designed against your ICP |
| Engines | 1 — Google AI Mode | ChatGPT, Perplexity, Google AI Overviews, Gemini |
| Runs per question | 1 | 3, with the spread reported |
| Findings | 3, plus 3 quick wins | Complete, ranked by cost of the gap |
| Competitor benchmark | Who was cited instead | Citation share against a named set |
| Order of work | Not included | Sequenced, with effort against each item |
| Turnaround | 48 hours | Part of the Foundation Sprint |
Assistant outputs are non-deterministic. Ask the same question twice and the cited sources can differ. A single run is a snapshot — useful for spotting an obvious absence, useless for claiming a trend. Any provider reporting a visibility score without stating how many times each prompt was run is reporting noise with a decimal point on it. Our measurement protocol is published in full, including what it cannot prove.
It cannot guarantee a citation, predict a model update, or attribute a sale. What it can do is establish whether you are mechanically capable of being cited, and whether, on a fixed set of questions on a fixed date, you were. That is a smaller claim than most of this market makes, and it is the one that survives contact with a sceptical buyer.
Yes. Send a domain, get the report inside 48 hours, no call required and no card. It is the short version of the paid protocol — five buyer questions on one engine, one run each, three findings and three quick wins — and it is yours to keep and act on whether or not you ever reply.
The free check is returned inside 48 hours. The full audit forms part of the two-week Foundation Sprint, because it is run alongside the fixes rather than handed over as a document to action later.
An SEO audit asks whether your pages can rank in a list of links. An AEO audit asks whether an AI assistant can fetch the page during a live query, lift a quotable statement from it, and identify you as the source. A site can pass one and fail the other, which is why the problem usually shows up as strong rankings alongside total absence from AI answers.
Ranking and citation are decided by different mechanisms. Retrieval agents are separately permissioned from Googlebot, models lift self-contained statements rather than ranked documents, and attribution depends on entity clarity rather than position. Strong rankings with no AI citations is the most common pattern we see.
The free check runs on Google AI Mode only, logged out and with the market pinned, so the result is reproducible. The full audit covers ChatGPT, Perplexity, Google AI Overviews and Gemini, because each retrieves from a different index and access has to be verified per named agent.