Why 9 in 10 Websites Fail AEO (And How to Fix It)

Most Websites Are Reachable by AI. Almost None Are Answerable.

Across 59 real websites that ran our answer engine optimization audit, the average score was 26 out of 100, and 9 in 10 scored below 40. Not one scored an A. AEO is, by a wide margin, the worst-performing dimension of AI visibility we measure.

That sentence is the whole problem in miniature. Most teams have quietly solved the first half of AI visibility: their sites can be reached by AI crawlers like GPTBot and PerplexityBot. They have not solved the second half: turning a reachable page into one an AI can actually understand, lift an answer from, and cite. Reachable is not the same as answerable, and the gap between the two is enormous.

The data here comes from real audits, not a survey. Between April 26 and June 3, 2026, 296 distinct domains ran 840 audits on our free Radar tools. This post zooms in on one tool, the AEO Auditor, and the 59 domains that used it. The numbers are anonymized and aggregate. No company is named. What they show is uncomfortable and consistent.

This is not a takedown of the sites that scored low. It is a snapshot of what AI answer engines see when they try to read the average website in mid 2026. The encouraging part: because the failure is structural, the fix is mechanical. You do not need better writing. You need the markup and structure that lets a machine find the answer you already wrote. We built two tools to test whether AI engines actually cite your pages.

What Does Answer Engine Optimization Actually Measure?

Answer engine optimization is the practice of structuring a page so an AI answer engine can extract a direct, citable answer from it. It is the on-page layer of getting cited by AI, and it is distinct from two things people often confuse it with.

Crawlability is whether an AI bot can fetch the page at all. It is a yes or no plumbing question answered by robots.txt rules, server responses, and whether your content exists in the HTML. Citation is the end result: whether an AI actually mentions you in an answer. AEO sits in the middle. It is everything that happens after the bot has your page and before the AI decides you are worth quoting. Can it find a clean answer? Can it trust the source? Can it lift a passage without dragging in noise?

Why this is its own discipline

A page can be perfectly crawlable and still score near zero on AEO. The bot fetches it, reads a wall of unstructured text with no schema, no clear heading hierarchy, and no front-loaded answer, and moves on. Nothing is broken in the traditional sense. The page just gives the answer engine nothing to grab.

AEO measures the signals that make a passage liftable: FAQ and Speakable schema, a strong Article and Organization graph, a clean H1 to H2 to H3 hierarchy with no skipped levels, a quotable answer at the top of each section, and freshness markers like dateModified. None of these are about content quality. They are about machine readability.

“Beautiful content an AI cannot parse is invisible content. The answer engine does not grade your prose. It grades whether it can find the answer fast and trust where it came from.”

Lloyd Pilapil, Pixelmojo

The Data: 9 in 10 Sites Score Below 40 on AEO

Of the 59 domains that ran the AEO Auditor, 55 landed below 40 and only two scored 55 or higher. Zero reached 85, the A threshold. The average was 26 out of 100.

That distribution is not a bell curve with a long tail of laggards. It is a wall. The overwhelming majority of sites cluster at the bottom, which tells you the problem is not a few bad actors. It is the default state of the web.

Correction: an earlier version of this post reported these results as letter grades on the AEO Auditor's cut-offs at the time (C from 55, D from 40, F below 40). Its grade table also listed only 57 of the 59 domains, with 53 below 40, while its text put 55 below 40; the table now follows the text. As of 25 September 2026, every Radar tool grades on one scale (A from 85, B from 70, C from 50, D from 30, F below 30), so this version reports the same June 2026 scores as score ranges instead of letters. No score changed. On the current scale every score below 40 is a D or an F, so at least 55 of the 59 domains still grade D or F. The Scored D or F column in the next table uses each tool's cut-offs at the time: below 55 for crawlability, AI Readiness and AEO, and below 50 for llms.txt.

The practical read for any individual team is oddly reassuring. The field is so weak that even a modest score puts you ahead of nearly everyone. Moving a single important page from below 40 to 55 does not require a heroic effort, and it lifts you past 95 percent of the domains in this dataset.

Reachable but Not Answerable: The 38-Point Gap

The clearest way to see the problem is to put AEO next to the other dimensions of AI visibility. When you do, a pattern jumps out: the more a dimension depends on machine-readable structure rather than basic access, the worse sites perform.

Crawlability averages 64 and only 29 percent of sites fail it. The plumbing mostly works. Bots can reach the pages. But AEO averages 26, with 93 percent scoring D or F. That is a 38-point gap between being reachable and being answerable, and it is the single most important number in this report.

This gap explains a frustration we hear constantly: "Our site is fine, GPTBot is allowed, so why are we never cited?" The honest answer is that crawl access is table stakes. It gets the AI to your door. AEO is whether you hand it a clean answer once it is inside. Most sites open the door and then hand the AI a filing cabinet with no labels.

“Crawl access gets the AI to your door. AEO is whether you hand it a clean answer once it is inside. The 38-point gap is the cost of skipping the second step.”

Radar audit data, June 2026

Why Do AEO Scores Land So Low?

The failures are remarkably consistent across the 59 domains. Five problems account for almost all of the lost points, and every one of them is a structure problem rather than a content problem.

Missing structured data

The most common failure is the simplest. No FAQ schema, no Speakable schema, no Article or Organization graph. The page has answers in it, but nothing tells the answer engine where they are or who is making the claim. Schema is the difference between a paragraph and a labeled, machine-readable answer with an author and a date attached.

No front-loaded answer

Most pages bury the answer. They open a section with context, history, and throat-clearing, then deliver the point three paragraphs down. Human readers tolerate this. Answer engines do not wait. The first one or two sentences under a heading should be a standalone, quotable answer to the implied question. If the AI has to read the whole section to find the point, it usually moves on to a competitor who front-loaded theirs.

Broken heading hierarchy

Skipped heading levels, multiple H1s, and decorative headings that are really styling break the document outline an AI uses to understand structure. A clean H1 to H2 to H3 nesting tells the engine how the page is organized and which passages answer which questions. When the hierarchy is broken, the page reads as undifferentiated text.

No freshness signals

AI engines weight recency, especially for anything that changes over time. Pages with no dateModified, no visible publish or update date, and no freshness markers read as undatable and therefore risky to cite. A simple, accurate dateModified in the Article schema is one of the cheapest points on the board.

Content rendered only in JavaScript

When the substantive content only appears after client-side JavaScript runs, the AEO parser often sees an empty shell. The answer exists in the browser but not in the HTML the engine reads. This single issue can take an otherwise strong page to near zero.

What a Passing AEO Page Looks Like

The two domains that scored 55 or higher in our dataset did not have better content than the others. They had better structure. Every section opened with a direct answer, the schema was complete, the heading hierarchy was clean, and the content was present in the HTML. That is the entire difference between an F and a passing grade.

A page that scores well on AEO does five things, in rough order of impact:

• It leads every section with a one or two sentence answer to the question that section addresses.
• It carries complete structured data: FAQPage schema for question and answer blocks, Speakable for the key passages, and Article plus Organization to establish authorship and source trust.
• It uses a clean H1 to H2 to H3 hierarchy with no skipped levels, and phrases headings as the questions a user would actually ask an AI.
• It exposes a dateModified and keeps it honest, so the engine can trust the page is current.
• It serves the core content as HTML, not as a JavaScript-only render.

None of that is a content rewrite. It is markup, structure, and the discipline of answering first. That is why a single page can usually move from an F to a C in an afternoon. The hard part, writing something worth citing, is the part most teams have already done.

How to Audit and Fix Your Own AEO

Start by measuring, because you cannot fix signals you cannot see. Run the page through the AEO Auditor, which scores it 0 to 100 and reports each missing signal: schema gaps, heading problems, missing front-loaded answers, and freshness issues. Then work the list in impact order.

First, add the answer at the top of each section, because front-loading is free and high-impact. Second, add FAQ and Speakable schema to the passages you most want cited. Third, repair the heading hierarchy. Fourth, add Article and Organization schema with an honest dateModified. Fifth, confirm your core content renders in HTML, not only in JavaScript.

If crawlability is also in question, check that first with the AI Crawl Check, since there is no point optimizing a page an AI bot cannot reach. But for most sites in our data, crawl access was already fine. The points were being lost downstream, on AEO.

“You cannot fix signals you cannot see. Measure the page first, then work the missing signals in impact order. Most of the gain is in the first afternoon.”

Lloyd Pilapil, Pixelmojo

Is AEO the Same as SEO or GEO?

AEO, SEO, and GEO are related but distinct, and the difference matters for where you spend effort. SEO optimizes for ranking in traditional search. GEO, generative engine optimization, is the broad practice of getting cited inside AI answers. AEO is the on-page layer of GEO: the page-level structure that makes a single page extractable and answerable.

Put simply, AEO is the work you do on the page, and GEO is the outcome you are chasing across engines. If you want the full picture of how the three fit together, our SEO vs GEO vs AEO guide breaks down each layer, and the GEO playbook covers the off-page side of getting cited. For the wider dataset behind this post, including how AEO compares to every other dimension across hundreds of audits, see our AI visibility benchmark report.

The reason AEO deserves its own focus is the 38-point gap. It is the dimension where the most points are being left on the table, by the most sites, for the most fixable reasons.

The Fix Is Cheaper Than the Problem

The headline number is bleak: 9 in 10 sites fail AEO, with an average of 26 out of 100. But the same data that exposes the problem also sizes the opportunity. The failure is structural, the field is weak, and the fixes are markup and structure rather than a content rewrite. A single important page can move from below 40 to 55 in an afternoon, and that alone clears 95 percent of the domains we measured.

If AI search matters to your business, AEO is where the most points are sitting unclaimed. Start with one page that should be getting cited and is not.

Ready to see where your pages actually stand?

• Run the AEO Auditor - $5 in Radar, scores any page 0 to 100 and lists every missing signal
• Check AI crawlability first - Confirm AI bots can reach the page before you optimize it
• Explore the full Radar platform - All thirteen AI visibility tools and the published methodology
• Talk to Pixelmojo - If you want the fixes shipped, not just the findings