What an AEO audit is for
An AEO audit exists to tell you which of four separate problems is keeping you out of AI answers, because the fix for each one costs a different amount and takes a different length of time. That is the whole job. Everything else in the deliverable is supporting evidence for that one ranking.
This matters more than it sounds, because the four problems look identical from where you are standing. You ask an assistant a question your business should own, your name does not appear, and no message explains why. A site the crawler cannot reach and a site that is reached and passed over produce exactly the same silence. So does a site that gets retrieved but carries nothing quotable, and so does a business nobody outside its own domain has ever written about.
Four causes, one symptom, no error message. An audit that does not separate them has described your situation rather than diagnosed it, and you will spend the next quarter fixing whichever one the supplier happens to sell.
The trap is that all four are real problems. A supplier recommending more content is not lying about content mattering. They are answering a question you did not ask, which was not what is imperfect but what is binding. Those are different lists and usually only the second one is short.
The output you want is a ranked answer to which of the four is binding right now. Not a score, and not an inventory of every imperfection on the site.
The four areas, and why the order matters
They are not four equal columns on a scorecard. They stack, and the ones underneath decide whether anything above them can help you at all.
Crawler access
Whether the specific crawlers feeding AI answers are permitted to fetch your pages. This is binary, it is set in files you control, and a business can sit behind an accidental block for months with nothing appearing in its own analytics to suggest it.
It gets missed because the rule is frequently not written by anybody at the company. It arrives with a security product, a bot-management default, or a setting inside a hosting plan, and it is invisible until somebody goes looking specifically for it. This is the cheapest finding an audit can produce and the most awkward one to discover eighteen months late.
The check is not "is the site crawlable" in general. Search crawlers and answer crawlers are different clients with different names, and a site can be entirely open to one while closed to the other. A report that says the site is crawlable without naming which clients it tested has not done this.
Page answerability
Whether a page, once fetched, contains a sentence answering its own question directly enough to be lifted whole. Most commercial writing does not, and the reason is a habit rather than a failing. Marketing pages are built to hold attention and arrive somewhere, so the statement lands in the fourth paragraph once the context has been laid. A system composing a short answer needs the statement near the top, in one piece.
This is where most sites have genuinely easy ground, which is why an audit leads with it. It responds within weeks, it needs no engineering, and the work is moving a conclusion you already wrote rather than researching a new one. Our page on writing for LLMs goes through what a liftable paragraph actually looks like.
Subject depth
Whether there is enough on your site about a subject that reaching for you is a reasonable thing to do when the subject comes up. One page about a topic is a page. Fifteen connected pages answering the questions around it is a body of work, and the gap between those two shows up in whether you are retrieved at all rather than in where you place.
Depth runs on quarters and cannot be shortened much by spending faster, because the pages have to be genuinely different from one another to count for anything. This is also the area where an audit is most likely to hand you a number and least likely to hand you a plan.
Presence in sources you do not own
Whether your category is discussed in the places assistants draw on, and whether you appear in that discussion. You have influence here and no control. Any audit presenting this as a deliverable with a delivery date has misunderstood what it is.
What a good audit actually hands you
A useful AEO audit is short, specific and slightly uncomfortable to read. It names the one thing that is binding and says plainly which of the other findings you can leave alone for now.
Four artifacts are worth having. A list of which crawlers can and cannot reach you, with the exact rule that decides each one. A set of named pages with the specific sentence each should open with. An honest statement of how much depth exists on your priority subject next to what actually gets retrieved for it. And a baseline of what assistants say today when asked about your category.
That last one is the item most often missing and the one that makes every other item judgeable. Without a reading taken before the work, you cannot tell in three months whether anything moved, and you will end up arguing about impressions instead of reading a change. A baseline is also the cheapest thing on the list to produce and the easiest to skip, which is why it goes missing.
The finding that is usually wrong
Audits routinely report a single visibility percentage and treat a low one as the problem to solve. A percentage across a set of questions somebody chose is a summary of that question set, not a property of your business. Change the questions and the number changes without anything about you changing at all.
It is a useful thing to watch over time against a fixed set. It is a poor thing to receive as a verdict. Ask which questions produced it, how many times each was asked, and on which assistants. If those three answers are not in the document, the number is decoration. There is more on what a number like that can carry in our piece on the AI visibility score.
The second finding that is usually wrong
An audit that lists technical imperfections in descending order of how easy they were to detect has sorted by the wrong axis. Detectability and consequence are not the same thing, and a long list of small structural notes will crowd out the one access rule that is costing you everything.
The half you can run yourself this week
Two of the four areas are checkable without buying anything. Doing them first means you brief a supplier about the half you genuinely cannot see, rather than paying to be told what an afternoon would have told you.
For access, find every rule on your site and at your edge that names a crawler, and confirm what each permits. Read the rules rather than asking whoever set them up, because what you want is what the file says and not what somebody remembers intending. Check the edge as well as the site. A permissive file behind a restrictive edge is a closed door with a welcome mat on the far side of it.
For answerability, take your ten most commercially important pages and read only the first two sentences of each. If those two sentences do not answer the question in the page title, you have found the work. This exercise takes an hour and it tends to be sobering, because the pages that fail hardest are usually the ones somebody spent the most on.
For a baseline, write between six and ten questions a real buyer would ask in their own words, ask each one several times across more than one assistant, and write down what came back. Ask each question more than once. The same question put twice can return different names, and a single reading tells you almost nothing about a distribution. Our walkthrough of tracking brand mentions by hand covers how to do this without buying anything.
Arrive with your access rules, your ten opening sentences and your baseline. The conversation changes completely when a supplier is being asked about depth and sources rather than being asked to go and find something.
The honest trade-off
An audit is worth buying when you genuinely do not know which of the four is binding. It is a waste when you already do, and most businesses that have never looked are blocked or unanswerable. Both of those are findable in a day by somebody already inside the company.
The case for paying is a different case, and it is a real one. A baseline built properly is repetitive work that has to be done consistently across enough readings to mean anything, and consistency is exactly what an internal team drops first. An outside read on whether your subject coverage is thin is genuinely hard to produce from inside a business that has been writing about itself for a decade. And there is value in somebody competent telling you that the quarter of work you were about to commission is not your constraint.
What is not worth paying for is a document that scores you, inventories every technical imperfection, and closes by recommending more content. That is a report shaped like an audit, and you can tell before you sign because the proposal will describe deliverables rather than questions it intends to answer.
When to skip the audit entirely
There are two situations where commissioning one is the wrong move, and both are common enough to name.
The first is a business that has never checked its own access rules. Until that is done you are paying somebody to discover a fact you can establish before lunch, and the finding will arrive with an invoice attached to it. Check first. If you are blocked, fix it, wait for pages to be refetched, and take a reading afterwards. You may find the problem was the whole problem, which happens more often than the industry likes to admit.
The second is a business in the middle of a rebuild. An audit describes a site as it stands today, and a site being rewritten will invalidate half the findings before the document is read. Wait until the new pages exist, then audit those, because the interesting question about a new site is whether it answers well and nobody can tell you that in advance.
There is a third case worth mentioning that is less about timing and more about honesty. If the real question is whether a category is worth pursuing at all, an audit will not answer it. An audit tells you why you are not present. It does not tell you whether being present is worth what it costs, and those get confused because both arrive as documents with charts in them.
The way to tell which question you are actually asking is to imagine the audit coming back saying everything is fine. If your response would be relief, buy it. If your response would be "then why is nothing happening", your question is about demand rather than about visibility, and a different piece of work answers it.
Bottom line
Buy the diagnosis, not the score. The output that changes what you do next is a ranked answer to which of four problems is currently keeping you out, with the crawler rules and the opening sentences attached as evidence rather than as an appendix.
Run the top two areas yourself first. They are the cheap half, they are checkable in a day, and finding your own block costs considerably less than being told about it in a slide three weeks later. Then buy the half you cannot see from where you sit, which is the baseline and the honest read on depth.
If the audit you are offered does not include a baseline, ask for one before you agree to anything else. Everything the engagement claims afterwards depends on a reading taken before it started, and nobody can go back and take one later.