A site that ChatGPT, Perplexity, Google AI Overviews, and Claude all read cleanly is built differently from a site that only optimizes for Google’s blue links. Pages with a First Contentful Paint under 0.4 seconds average 6.7 AI citations, while pages loading over 1.13 seconds average 2.1, and the gap is widening as AI search platforms take a larger share of discovery. The checklist below covers the technical, structural, and editorial items that determine if a site shows up when a buyer asks an AI assistant a question instead of typing into a search bar.
Each section closes with a short list of items to verify on the live site, in the order they should be addressed.
Crawlability and Indexing Foundations
AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended) read the same pages search crawlers read but treat the content differently. The first preparation step is making sure the site is reachable and parseable by all of them.
Verify these items on the live site. The robots.txt file does not block GPTBot, ClaudeBot, PerplexityBot, or Google-Extended unless an intentional decision was made to do so. An XML sitemap exists at the conventional location and is submitted to Google Search Console. Canonical tags point to one definitive URL per piece of content. HTTPS is enforced site-wide with no mixed content warnings. No important pages return soft 404s.
The single most-missed item is the `Google-Extended` user agent token in robots.txt. A site that allows Googlebot but forgets to allow Google-Extended is invisible to AI Overviews specifically.
HTML and Semantic Structure
Most AI crawlers do not execute JavaScript. Content that loads after page paint is invisible to them. Server-rendered HTML or static generation fixes this for sites built on modern frameworks.
Verify these items. View source on any important page and confirm the body content appears in the initial HTML response. The page has one and only one H1, with H2s for major sections and H3s for subsections, in source order matching visual order. Semantic landmark elements (`main`, `nav`, `article`, `section`, `aside`, `footer`) wrap the appropriate content regions. Lists use `ul` or `ol’ rather than styled `div` containers. Tables hold only tabular data. Layout is handled with CSS.
Heading hierarchy is the single most useful structural signal for AI summarizers, which use it as the table of contents they cite from.
Structured Data Markup
Schema markup gives crawlers explicit facts about the page that text alone leaves implicit. JSON-LD is the format Google and most AI systems prefer, and it goes in the page head. Yoast’s structured data with schema ultimate guide explains the modular approach of describing the WebPage, the WebSite, and the Organization in one connected graph.
Verify these items. Organization schema appears on every page identifying the business by legal name, address, and phone. LocalBusiness schema is on contact and location pages. Product or Service schema is on the offering pages. FAQPage schema is on any page with a question-and-answer block. BreadcrumbList schema is on every page below the homepage. Every JSON-LD block validates against the Schema.org Validator and the Google Rich Results Test.
Schema accuracy matters as much as presence. AI systems compare structured data against visible page content and ignore (or penalize) pages where the two disagree.
Performance and Page Quality Signals
Speed acts as a tiebreaker in Google ranking and a deprioritization signal in AI summaries when other factors are close.
Verify these items. The LCP score is under 2.5 seconds on mobile. The INP score is under 200 milliseconds. The CLS score is under 0.1. Server response time (Time to First Byte) is under 200 milliseconds. Hero and above-the-fold images are served as WebP or AVIF, sized to the viewport, with explicit width and height attributes. Critical CSS is inlined in the head. Non-critical JavaScript is deferred or async-loaded.
For sites where the current foundation is too far gone to retrofit, the alternative is to build a website with AI tooling that handles these performance defaults from the first draft, then add the content and brand layers afterward.
Accessibility for Machines and Humans
Large language models behave like screen readers. They read code rather than visuals. The accessibility work that benefits people with visual impairments also benefits AI extraction. The A11Y Project checklist is the most-used quick reference for the items below.
Verify these items. Every image has descriptive alt text that describes what the image actually shows in context. Every form input has an associated label. Color contrast between text and background is at least 4.5:1 for normal text and 3:1 for large text. Keyboard navigation reaches every interactive element in logical order. Focus indicators are visible on every focused element. Skip links allow keyboard users to bypass repeated navigation.
The intersection of accessibility and AI visibility is one of the highest-leverage items on the whole checklist. Fixing one fixes the other.
Content Depth and Update Cadence
AI systems prefer content that answers a specific question well, with a definitive statement up front and supporting evidence near it. Skim-readable structure beats narrative arc for AI extraction.
Verify these items. Each important page opens with a one-sentence or two-sentence direct answer to the question the page targets. Specific numbers, named examples, and dated facts replace vague qualifiers like “many” or “experts say.” A “Last updated” date is visible on every page where freshness matters. Internal links from authoritative pages point to important secondary pages. Author bios with credentials appear on content where authority matters.
E-E-A-T signals are reinforced through what the content actually demonstrates, rather than asserted in marketing copy.
Off-Site Entity Signals
AI systems gain confidence in a brand when independent sources agree about it. Your own site is a biased source by definition. The signals that move an AI from “probably real” to “confidently cited” come from elsewhere on the web.
Verify these items. The business name, address, and phone number are consistent across at least 10 industry directories. The company has a Wikipedia entry where notability supports it. Profiles exist on the review platforms most cited by AI systems for the business’s category (G2 for software, Yelp for local services, TripAdvisor for hospitality). The business has been quoted in at least 3 trade or general-interest publications in the last 12 months. SEMRush’s guide to generative engine optimization covers the off-site work that builds entity consensus.
The off-site work is slow and unglamorous. It is also where the largest gap between mediocre and AI-cited sites opens up.
Pre-Launch Sweep
Before declaring the site AI-ready, run through the items below. The Moz beginner’s guide to SEO is a useful complementary reference for any item below where the underlying SEO logic needs context.
Verify these items in order. Run the page through Google’s Rich Results Test and the Schema.org Validator. Run the page through PageSpeed Insights and confirm Core Web Vitals targets are met on mobile. Test the page with a screen reader (VoiceOver on macOS, NVDA on Windows). Check server logs to confirm AI crawlers are accessing the site without 403 or 404 errors. Read the page out loud and confirm a stranger could leave with the answer to the question the page promises.
A site that passes every item on this checklist is the site an AI system has the easiest time recommending. Most sites pass three or four sections cleanly and fail the rest. The order to fix is the order above.
