AI SEO Audit Checklist: 18 Checks for AI Search Visibility
A classic technical audit asks one question: can search engines crawl, index, and rank this site. An AI audit asks a different one: can AI assistants reach this content, understand what it means, and decide it is worth citing. Those are related jobs, but they are not the same job. A page can pass every item on a technical SEO audit checklist and still be invisible to ChatGPT, Perplexity, and Google AI Overviews, because the failure is happening one layer up.
This checklist covers 18 items in five groups. For each one, you get a short reason it matters and a concrete thing to look for. Some of these overlap with traditional SEO, which is the point: the foundation still has to hold. The rest are specific to how AI systems find and quote sources. You can run many of these checks by hand, but on a site of any size a crawler surfaces them in a single pass.
Can AI Reach Your Content
If an AI crawler cannot fetch your page, nothing else on this list matters. This is the layer people skip most often, because a page that looks fine in a browser can be closed to the exact bots that feed AI answers.
1. AI Bot Access in Robots.txt
AI engines use named crawlers: GPTBot for ChatGPT, ClaudeBot for Claude, PerplexityBot for Perplexity, and Google-Extended for Google’s AI training. Your robots.txt decides which of them may read your content. Many sites block these bots without realizing it, sometimes through a copy-pasted rule, sometimes through a security plugin’s default.
What to look for: Open yourdomain.com/robots.txt and check for Disallow rules targeting these user agents. Decide deliberately: blocking them protects content from AI training but also removes you from AI citations. Our complete robots.txt guide for AI bots covers the tradeoff in detail.
2. Llms.txt Presence and Accuracy
An llms.txt file is a proposed convention that points AI systems to your most important, clean content. Adoption is uneven and no engine treats it as a ranking input, so it is a low-cost signal rather than a fix. If you publish one, it needs to be accurate and current, not a stale list of moved URLs.
What to look for: Whether the file exists, whether every URL in it returns 200, and whether it actually reflects your key pages. If you are weighing whether to bother, do AI engines actually read llms.txt walks through the real evidence.
3. Content in the Initial HTML
AI crawlers are far less patient than Googlebot about running JavaScript. If your main content only appears after client-side rendering, many AI systems will see a near-empty page. Server-side rendering or static generation puts the text where any crawler can read it on the first request.
What to look for: Fetch your page without JavaScript (view source, or a crawler’s raw HTML view) and confirm the primary content and headings are present. If the body is empty until scripts run, that is a citation blocker. See JavaScript SEO and rendering for the fix.
4. Firewall and CDN Rules
Beyond robots.txt, a WAF or CDN can silently return 403 or challenge pages to AI user agents while serving humans normally. This is common with bot-management defaults that treat every non-browser agent as a threat. The result is an invisible block that robots.txt inspection will not reveal.
What to look for: Request your pages using an AI bot’s user-agent string and confirm you get a 200 with real content, not a block page or a JavaScript challenge.
Can Machines Understand the Page
Reaching the page is not the same as parsing it. AI systems extract a clean, structured version of your content before they can summarize or cite it, and vague markup makes that extraction unreliable.
5. Structured Data Validity
Schema markup gives machines an explicit statement of what a page is: an article, a product, an FAQ, an author. AI systems use it to disambiguate content and attribute it correctly. Invalid or mismatched schema is worse than none, because it introduces noise.
What to look for: Every important page type should carry the right schema, the JSON-LD should validate, and it should match the visible content. Our schema markup guide for SEO and AI search covers which types matter and how to avoid the common errors.
6. Semantic HTML and Heading Structure
A single descriptive <h1>, a logical heading order, and real semantic elements (lists, tables, paragraphs) let a model reconstruct your page’s structure. A wall of <div> tags styled to look like headings gives it nothing to hold onto.
What to look for: One <h1> per page, no skipped heading levels, and content marked up with the element that fits its meaning rather than generic containers.
7. Self-Contained Answers
AI systems tend to quote passages that stand on their own. A paragraph that answers a specific question in a few sentences, without depending on the three paragraphs above it, is far more quotable than a point buried inside a long narrative build-up.
What to look for: For each key question your page addresses, check that there is a direct passage a model could lift verbatim and have it still make sense.
8. Descriptive Metadata
Titles, meta descriptions, and Open Graph tags are among the first things a crawler reads. They frame what the page is about before the body is even parsed. Vague or duplicated metadata forces the model to guess.
What to look for: Unique, descriptive titles and descriptions on every page, matching the actual content, with no duplication across the site.
Is the Page Worth Citing
Access and clarity get you into the candidate pool. Whether a model actually cites you comes down to trust and substance. This is where AI visibility diverges most sharply from traditional ranking.
9. Factual Density and Specificity
Models favor sources that state concrete, checkable facts: numbers, dates, named entities, specific steps. Generic advice that could have been written about anything rarely gets quoted, because it adds nothing a model could not already generate.
What to look for: Passages that make specific, verifiable claims. If a section reads like filler, it is unlikely to earn a citation.
10. Freshness Signals
Recency matters more to AI answers than to classic rankings, especially for anything time-sensitive. A visible, honest publish or update date helps a model judge whether your page is current. After a shift like the completed June 2026 spam update, stale pages that never get revisited lose ground fastest.
What to look for: Accurate publishDate and updatedDate on your content, and a real process for revisiting time-sensitive pages rather than backdating them cosmetically.
11. Source and Author Clarity
Clear authorship, an about page, and consistent entity information all feed the trust signals AI systems weigh when choosing between competing sources. Anonymous content on an unclear site is a weaker candidate than the same content with a named, credentialed author behind it.
What to look for: Named authors with real bios, a substantial about page, and consistent organization details across the site and its structured data.
12. Unique Information Gain
Information gain is what your page adds that the ten pages already saying the same thing do not. AI answers synthesize many sources, so a page that only restates the consensus gives a model no reason to reach for it specifically. Original data, first-hand testing, or a genuinely different angle is what earns the quote.
What to look for: At least one thing on the page a reader (or a model) could not get from the top existing results. Our guide to citation-worthy pages in AI search goes deeper on this, and the broader strategy sits in the generative engine optimization guide.
Technical Hygiene That Still Counts
None of the AI-specific work saves a page that is technically broken. These classic checks still gate whether your content is a reliable source.
13. Canonicals and Duplicates
When several URLs serve the same content, a model may see a weaker duplicate instead of your canonical version, or split the signals between them. Clean canonical tags keep attribution pointed at one authoritative URL.
What to look for: Self-referencing canonicals on indexable pages, no canonicals pointing at redirected or missing URLs, and no unresolved duplicate variants.
14. Broken Links and Redirect Chains
Dead internal links and long redirect chains waste crawl attention and can strand content before it is read. They also read as neglect, which is exactly the opposite of the reliability signal you want a citing model to see.
What to look for: Internal links returning 4xx, and redirects with more than one hop. Point links directly at final URLs and repair or remove the broken ones.
15. Page Speed and Core Web Vitals
Slow pages get crawled less and abandoned more, and the underlying performance problems often correlate with the rendering issues from item 3. A fast, stable page is simply easier for every automated reader to process.
What to look for: LCP under 2.5 seconds, INP under 200 milliseconds, CLS under 0.1, starting with your highest-value pages.
16. Crawlable Internal Linking
Internal links are how any crawler, AI or otherwise, discovers depth and understands which pages you consider important. Orphaned pages and shallow linking hide your best content from the systems meant to surface it.
What to look for: No orphan pages, important content within a few clicks of the homepage, and descriptive anchor text that signals what each linked page covers.
Measure and Maintain
An AI audit is a snapshot of a fast-moving target. The engines, their crawlers, and their citation behavior all change on a scale of months, so a single pass is a starting point, not a finish line.
17. Track Your AI Visibility
You cannot improve what you do not watch. Google Search Console now reports AI Overview performance, and third-party tools track how often assistants cite your domain. One caveat worth internalizing: recent analysis found only about a quarter of cited sources overlap between ChatGPT’s different reasoning modes, so visibility in one engine, or even one mode, does not guarantee it elsewhere.
What to look for: A baseline from the GSC AI Overview performance report, plus whatever citation monitoring you can sustain. Watch trends, not single data points.
18. Re-Audit on a Schedule
Because AI crawling and citation rules shift quickly, the gap between audits should be shorter than for classic technical SEO. New bots appear, robots.txt conventions evolve, and engine behavior changes without announcement. A checklist run once and forgotten drifts out of date fast.
What to look for: A recurring audit, quarterly at least, plus a targeted re-check after any major site change or any confirmed change in how the AI engines crawl.
Running the Audit
Most of these 18 items are things a crawler can check for you in one pass: bot access, rendering, schema validity, broken links, redirects, canonicals, thin or dateless pages, and orphaned content. A tool like Seodisias runs these checks locally on your own machine, across Windows, macOS, and Linux, so your site data never leaves your computer. That local, cross-platform approach matters more when you are auditing content you may not want uploaded to a third-party service.
A practical order of work:
Fix access first. If GPTBot or ClaudeBot cannot reach a page, or it renders empty without JavaScript, no amount of citation-worthy writing will help. Clear the blocks before anything else.
Then earn the citation. Once the page is reachable and parseable, the leverage moves to substance: specific facts, a real update cadence, clear authorship, and information the consensus does not already contain.
Fix at the template level. If schema is missing or metadata is duplicated across hundreds of pages, the cause is in a template, not in individual posts. One fix there resolves it everywhere.
Re-crawl and watch. Verify each fix with another crawl, then track AI visibility over weeks. Attribution behavior is noisy, so look for direction over time rather than reacting to any single answer.
Summary
AI visibility is built in layers. First the content has to be reachable by the named AI crawlers and present in the initial HTML. Then it has to be understandable, through valid schema, clean structure, and self-contained answers. Then it has to be worth citing, through specific facts, freshness, clear sourcing, and genuine information gain. Underneath all of it, the classic technical foundation still has to hold.
Work the layers in that order. A page that is reachable but hollow will not get cited, and a brilliant page that AI crawlers cannot fetch will not either. This checklist keeps both problems in view at once, and pairs naturally with the classic technical SEO audit checklist: run that for the foundation, and this one for the layer AI search added on top.