What a Technical AEO Site Audit Actually Checks (and the Fixes That Move Citations)

A technical AEO site audit checks whether AI answer engines can reach your pages, render them, understand them, and trust them enough to quote. In practice that means five things: crawlability for AI bots, render completeness, structural clarity (headings and schema), freshness signals, and performance. The fixes that move citations fastest are almost always the unglamorous ones: an inherited noindex on your best guide page, a robots.txt line blocking Perplexity, a hero section that only exists after JavaScript executes, and Core Web Vitals stuck in the red.

Most B2B SaaS sites we audit are not invisible to ChatGPT because their content is bad. They are invisible because a page from 2022 still says noindex, or the FAQ block renders client-side, or the LCP is 4.8 seconds. AI retrieval layers are less forgiving than Googlebot. If they cannot parse a clean chunk on the first pass, they move on.

Here is what a real audit looks at, and where the leverage actually is.

Can AI crawlers reach the page at all?

The first check is the boring one and it costs the most citations. Robots.txt, meta robots tags, and User-Agent rules get inherited from an old SEO playbook and never revisited. GPTBot, PerplexityBot, ClaudeBot, and Google-Extended all read robots.txt, and they read it differently than Googlebot. A blanket Disallow: / for anything that "looks like a scraper" locks you out of the answer layer entirely.

What to check:

• Robots.txt for explicit allows or disallows on GPTBot, PerplexityBot, ClaudeBot, CCBot, and Google-Extended. Silent is not the same as allowed on some setups.
• Meta robots and X-Robots-Tag on your highest-value guide, comparison, and Q&A pages. Legacy noindex on a "thin" page from three years ago is the single most common suppressor we find.
• Server-side blocks at Cloudflare or your WAF that rate-limit or 403 AI user agents.
• Redirect chains and 404s on internally linked content. Broken chains bleed authority the same way they do in classic SEO, and retrieval systems drop the trail.

Fix these first. A single noindex removed from a well-written FAQ page often shows up in monitoring within two to four weeks.

Does the page actually render for a retrieval bot?

AI retrievers pull the DOM. If your answer, your headings, or your product detail load only after a hydration step or a lazy scroll, that content might as well not exist. Audit whether the raw HTML response contains the substantive text of the page, or whether it is a shell waiting for React to hand it a body.

Check:

• Server-rendered or statically generated critical content, not client-only.
• Fonts, CSS, and fetch requests unblocked in robots.txt (yes, blocking /api/ or /_next/ can partially break rendering for some crawlers).
• Lazy-loaded FAQ, testimonial, and comparison sections that hold your most quotable copy.

Is the structure legible enough to quote?

Retrieval layers slice pages into passages using heading structure and semantic markup. If your page has three H1s, a jumped heading hierarchy, and no publication date, a model has to guess where an answer starts and ends. Assume it will not.

• One H1 per page, question-shaped where the page is answering a question.
• Nested H2/H3 that mirror the sub-questions a buyer would actually ask an assistant.
• JSON-LD schema: Article and FAQPage on guide content, Product and Organization site-wide, BreadcrumbList for context. Missing schema does not kill citations, but it lifts confidence, especially on Claude and Perplexity.
• Visible publication and updated dates, plus a lastmod in sitemap.xml and a truthful Last-Modified header. Recency is a real weight.

You can spot-check individual pages with the free schema validator and an llms.txt generator before you invest in a full pass.

How fast, and how fresh?

Lighthouse scores under 70 correlate with lower inclusion in retrieval sets, particularly on Perplexity, which leans harder on freshness and load. Getting LCP under 2.5s, CLS near zero, and TBT under 200ms is not just an SEO tax anymore. It is a citation lever.

Freshness is the twin. A guide with no visible update date, a stale sitemap, and Cache-Control set to "cache forever" reads as an old page to an AI engine. Add real updated timestamps when you materially edit, and let the sitemap reflect it.

The fixes that move the needle

If you had to rank the impact from an audit, in our experience:

1. Removing inherited noindex or robots blocks on high-value Q&A, guide, and comparison pages.
2. Moving critical content from client-rendered to server-rendered.
3. Adding real publication/update dates and clean heading hierarchy.
4. Lifting Core Web Vitals from poor to good.
5. Adding Article and FAQPage schema where the content already answers questions.

Everything else is polish.

How Clark handles it

Auditing a site by hand is fine once. Doing it every week across a growing content library is where founders quietly give up. Clark, the agent behind Clarity Search AI, runs the Site Audit job on a schedule: it crawls the connected site, runs Lighthouse, groups issues by severity, and ships fixes as GitHub pull requests you can review and merge, single fix, section, or batch. The point is not to hand you another dashboard. The point is to close the gap between "audit found it" and "the fix is live," which is where most site health work dies.

If you want to see what a real audit surfaces on your site before you commit to anything, the free AI readiness check is a fair place to start. Rank is what everyone reports. Revenue is what we prove.