AEO Audit
Deep checks, honest recommendations
A real audit.
Not a vibe check.
Moose fetches your page the same way OpenAI, Anthropic, Perplexity, Google, and Microsoft do, and shows you what each one can see. Every finding is graded by the evidence behind it.
Hi, Moose · AEO Audit
yourbrand.com/pricing
Page score
82
/ 100
Solid page. Three things worth fixing.
Extractability91
Answer placement64
Trust signals86
What I found
Your robots.txt blocks OAI-SearchBot. ChatGPT can't cite what it can't fetch.
Proven
The pricing table only renders with JavaScript. OpenAI's and Perplexity's crawlers won't run it.
Proven
Your direct answer sits 58% down the page. Engines mostly cite the top third.
Directional
The "43% faster" claim has no named source. Attributed numbers are easier to quote.
Best practice
4 recommendations. When a page is clean, I say zero.
What I check
First I check whether they can read you at all.
Every audit runs retrieval checks first - crawler access, rendering, working links, Bing indexing - then scores the content across eight categories. These are the ones that come up most.
AI crawler access
I read your robots.txt the way the crawlers do and test twelve of them by name across OpenAI, Anthropic, Perplexity, Google, and Microsoft. Some sites block these bots without knowing it.
JavaScript rendering
Most AI crawlers don't render JavaScript - Google's does, but OpenAI's, Anthropic's, and Perplexity's don't². I fetch your page with scripts off, then with them on, and measure how much of your content only exists after JavaScript runs.
Bing index presence
Copilot answers come from Bing's index, and ChatGPT leans on it too¹. I run a real Bing search for your page and read the results. When the check can't tell, the report says inconclusive - I don't guess.
Link health
When Vercel measured AI crawler traffic, over a third of OpenAI's and Anthropic's fetches hit 404 pages², and Ahrefs found AI assistants send readers to dead links at almost three times Google's rate³. I sample your internal links and flag the broken ones first.
Answer placement
Engines overwhelmingly cite the top third of a page. I check where your direct answer sits, and whether someone skimming - human or machine - would reach it.
Related-question coverage
One question fans out into a dozen related ones. On a managed plan, I test your page against the adjacent queries the engines ask and show you the gaps.
Graded, not guessed
Every recommendation shows its evidence.
Every recommendation carries one of three labels. The label tells you how strong the evidence is, so you know what to fix first and what can wait.
Proven
Retrieval preconditions with direct evidence: crawler access, server-side rendering, working links, index presence. Fix these first - the other checks assume the engines can read you.
Directional
Backed by large-scale observational data: answer-first structure, freshness, clean formatting. Good signals to act on, and I'll tell you where the data thins out.
Best practice
Plausible but unproven: statistics, quotations, sourcing polish. Cheap to do and good for readers - just not sold to you as a guaranteed citation lever.
Under the hood
Where the depth is.
Robots.txt and rendering checks are table stakes - plenty of tools run those. These are the parts that took the engineering.
Twelve crawlers, told apart
Each AI company runs separate bots for search, live fetching, and training. I test them separately and tell you which blocks matter: blocking a training bot is a policy choice, blocking a search bot costs citations. Most tools lump them together.
The citation landscape
On a managed plan, I run your target query and capture the live answer: which domains got cited, whether yours is among them, and how much of the answer comes from sources you can't edit.
Rewrites anchored to your page
Every proposed edit carries the exact text it replaces, matched against your live page, and a second model reviews each recommendation before it reaches you. A suggestion that doesn't line up with what's really on the page is dropped, not shipped.
Your Google rankings, protected
If you've connected Search Console and the page already ranks well for the query, I keep the changes additive - I won't rewrite the title that earned the position to chase an AI citation.
Re-audits that keep score
Your first audit becomes a baseline. Run it again after the fixes and I report which issues were resolved and how each category moved.
Reads your language
Answer-readiness scoring is language-aware across English, German, Spanish, French, Italian, and Portuguese - not English rules applied to everyone.
What I won't tell you
The advice you won't hear from me.
Some common AEO advice has been tested by now and didn't hold up. When the research says something doesn't work, I don't recommend it.
"Add an llms.txt file."
In the largest study to date, 97% of published llms.txt files received zero AI traffic. Google says it isn't planning to support it.
"Schema markup boosts AI citations."
The only treatment-and-control study in the space found no citation uplift on any platform. Keep schema for classic SEO - it isn't an AI lever.
"Write longer content."
There is no ideal word count for AI search, and padding rewards no one. I never recommend length for its own sake.
"Ten fixes for every page."
A good page gets one or two recommendations, sometimes none. Zero is a real result.
Then I help you fix it
The audit is step one. The repair is the point.
For each finding, I draft the fix in your voice. You review it, I publish it to your CMS, and then I re-check the page to see if it worked.
1Audit finds the gap
2I draft the fix, in your voice
3You review and give the nod
4I publish to your CMS
5I re-check and report back
See how the loop runs end to end: Workflows →
Run your first audit
See your page
the way the engines do.
Download Hi, Moose and audit a page today. If the page is clean, that's what the report says.