geo checklist

Generative engines retrieve content through the same crawling, indexing, and ranking pipeline as classic search, and Google states there are no additional requirements to appear in AI Overviews or AI Mode beyond being indexed and eligible for a snippet. Work through the full SEO checklist first: crawlable pages, clean architecture, and helpful content are the foundation every GEO item below builds on.

See our SEO best practices checklist

Research the prompts your buyers actually type into ChatGPT, Perplexity, Gemini, and Google's AI Mode, the AI equivalent of keyword research. AI answers fan one prompt out into many sub-queries, so monthly search volume understates the real opportunity: map the questions, comparisons, and recommendation-style prompts where your brand should appear, then shape content to answer them. Track your share of those answers over time with an AI visibility tracker such as Ahrefs Brand Radar.

Query fan-out research (Seer Interactive) Why search volume is declining in importance (Seer Interactive)

Verify that robots.txt and your CDN or firewall are not blocking AI crawlers such as GPTBot, ClaudeBot, PerplexityBot, and Google-Extended; a site these bots cannot fetch cannot be read or cited by their engines. Run your domain through the AI bot analyzer to see which bots are allowed or blocked, and also review CDN bot-protection rules, which can block crawlers at the network level even when robots.txt allows them.

AI bot analyzer

AI agents increasingly browse websites on behalf of users to compare options and complete tasks. Run your site through the Is It Agent Ready test to score how well an autonomous agent can read and operate it, then fix what it flags, such as content that only appears after JavaScript runs, unlabeled controls, and blocked bots.

Is It Agent Ready test

Browser agents gather data by reading the DOM, the accessibility tree, and visual renderings, so a site that works well for assistive technology also works well for agents. Follow the web.dev agent-friendly guidance: semantic markup, forms and buttons with real accessible labels, key content reachable without complex interactions, and a stable, descriptive page structure.

Build agent-friendly websites (web.dev)

Scan key pages with an LLM content visibility scanner to check how much of your copy large language models can actually see and extract. Use the results to find content that is invisible to AI systems, typically copy rendered by JavaScript, locked inside images, or buried in weak structure, and fix those pages first.

LLM content visibility scanner

Every page has two versions: the raw HTML response the server sends and the DOM after JavaScript runs. Most AI crawlers do not execute JavaScript, so headings, body copy, and internal links that exist only after rendering are invisible to them; serve that content in the initial HTML response via server side rendering or static generation. Spot-check single pages with the Ahrefs SEO Toolbar and audit at crawl scale with Sitebulb's rendered hints.

Response vs render hints (Sitebulb) The rise of the AI crawler (Vercel)

Give every page a unique title tag, a meta description, and a single H1 that states the page topic plainly; these are the first fields both search engines and AI retrieval systems read when deciding what a page is about. Put the primary keyword or entity in the title and H1, keep titles roughly 50 to 60 characters, and write the description as a clear one or two sentence summary the page actually delivers on.

Pass the Core Web Vitals assessment on real-user field data: LCP of 2.5 seconds or less, INP of 200 milliseconds or less, and CLS of 0.1 or less, with TTFB under 0.8 seconds and FCP under 1.8 seconds as supporting metrics. Test every key page on PageSpeed Insights and apply the fixes its Diagnostics and Insights sections name, then monitor field data over time with the CrUX Dashboard or Treo instead of relying on one-off lab tests.

PageSpeed Insights Core Web Vitals overview (web.dev)

Google lets signed-in users mark sites as preferred sources so their content appears more often in Top Stories and related surfaces. Confirm your site is eligible per Google's documentation, then invite loyal readers to select you, for example with a small preferred-source prompt or button on the blog, so your coverage surfaces more prominently for the audience that already trusts you.

Preferred sources (Google Search Central)

Plan every article from an outline built for AI retrieval before writing: question-form headings that mirror real prompts, a short extractable summary after the H1, an FAQ block at the end, and slots for statistics, expert quotes, and cited sources. Outlining first keeps each section self-contained and quotable, which is exactly what generative engines lift into answers.

How to rank in AI search (Search Engine Journal)

Write in concise subject-verb-object sentences (semantic triples) and name specific entities close together; embeddings-based retrieval matches dense, unambiguous statements far better than vague prose. Support claims with sourced statistics, direct quotes from named experts, and inline citations: the GEO research paper (arXiv 2311.09735) found that citing sources, adding statistics, and adding quotations are the top-performing methods for getting content into generative answers.

GEO research paper (arXiv) Semantics for search and LLM visibility (iloveseo)

Package content into self-contained, scannable units instead of long prose: an FAQ section answering real user questions in one or two sentences each, bulleted lists for parallel items, numbered steps for any process, and real HTML tables for comparative data. These structures are what search engines lift into featured snippets and what AI answer engines can quote cleanly, all sitting under a proper H1 to H3 heading outline.

Structure each page so machines can read meaning from the markup itself: one main, header, and footer; the H1 and page summary inside a header within main; each H2 topic wrapped in its own section nested in an article; time for dates, figure and figcaption for images, and full table markup (thead, th, tbody, caption). Keep the DOM under about 900 nodes and audit pages with a semantic HTML5 markup inspector.

Semantic HTML5 markup inspector Why semantic HTML still matters (Jono Alderson)

Show a visible published date and last-updated date on every article, and back them with machine-readable datePublished and dateModified values in structured data so Google and AI systems can establish freshness. Keep the visible dates and the markup consistent, and only change dateModified when the content meaningfully changes.

Providing publication dates to Google

Keep the server's average response time to crawlers under 200 milliseconds; slow responses waste crawl budget and throttle how much of the site search and AI bots fetch. Check the trend in Google Search Console under Settings, then Crawl stats, then the Average response time chart, and treat sustained values above 200ms as a server or hosting problem to fix.

Add JSON-LD structured data that matches the visible content of each page: Organization and WebSite site-wide, Article or BlogPosting on posts, plus FAQPage, Product, HowTo, Review, Person, or Event where they genuinely fit. Schema gives engines an unambiguous, machine-readable summary of your entities and relationships; keep the markup consistent with the on-page text, since mismatches undermine eligibility.

Give every meaningful image descriptive alt text (for example, "founder speaking at a marketing conference in Mumbai"), a keyword-rich filename such as blue-ceramic-coffee-mug.jpg instead of an auto-generated hash, and natural captions with contextually relevant surrounding text. On the technical side, declare width and height, compress to under 100KB where possible, serve WebP or AVIF, lazy-load below-the-fold images, and set og:image so shares carry the intended visual.

Google Images best practices Writing good alt text (Google)

Give every page exactly one H1 stating the topic, H2s for the main sections, and H3s nested under their parent H2, never skipping levels. Pick heading tags for meaning and style them with CSS rather than choosing tags for font size; a quick audit is to read only the headings top to bottom and check that they form a coherent outline of the page.

A post's page template carries trust signals independent of the copy: an author byline and box near the top linking to a full author page, a table of contents, breadcrumbs, both published and updated dates, visibly cited sources, and sharing buttons. Keep the layout clean, with a white background, generous margins, and hero images around 450px tall on desktop so the opening copy stays above the fold, all built on semantic HTML that passes Core Web Vitals.

Table of contents research (Nielsen Norman Group) Blog design best practices (Siege Media)

Google's quality rater guidelines call trust the most important part of E-E-A-T: untrustworthy pages score low no matter how expert they look. Answer the four trust questions on the site itself: who is responsible for it (an about page, a contact page with a real phone number and work email, team and leadership pages, legal pages in the footer), who created the content (credential-rich author boxes and author pages), how it was created (editorial guidelines and an AI policy), and why it exists. Structure the homepage so AI systems can extract what you are expert in, who recognizes that expertise, and what you offer.

Creating helpful, reliable content (Google) What is E-E-A-T in SEO (Hobo Web)

Generative engine optimization is making content citable by AI Overviews, ChatGPT, Perplexity, and Gemini. The foundational research (arXiv 2311.09735) found the top-performing methods are citing authoritative sources, adding statistics, and adding expert quotations; layer those onto a technically ready site with AI crawler access, sub-200ms responses, passing Core Web Vitals, server side rendering, semantic HTML, and structured data. Publish original research, keep content fresh, and track AI visibility monthly with a tool such as Ahrefs Brand Radar.

GEO research paper (arXiv) How to rank in AI search (Search Engine Journal)

WebMCP is Chrome's early preview of the Model Context Protocol for the web: it lets a site declare structured tools that in-browser AI agents can call directly instead of scraping the interface. Follow the Chrome early preview program and evaluate exposing key actions (search, quotes, bookings) as WebMCP tools once the spec stabilizes; for now treat it as a watch-and-prototype item.

WebMCP early preview (Chrome)

Cloudflare can serve a clean markdown version of your pages to AI agents through content negotiation, cutting token cost and parsing errors for crawlers that prefer text over full HTML. If your site runs behind Cloudflare, enable this and verify that key pages convert cleanly; it is a cheap way to make content easier for agents to consume.

Markdown for agents (Cloudflare)

llms.txt is a proposed plain-text file at the site root that gives language models a curated map of your most important content. Google Search ignores it entirely (it neither helps nor harms your Google visibility), but some other AI systems and tools do read it, so treat it as a cheap, low-priority addition: publish one if it is easy to maintain, and never at the expense of higher-impact items on this list.

Review robots.txt with AI user agents in mind: confirm no disallow rules or leftover staging blocks shut out the crawlers you want citing you (GPTBot, ClaudeBot, PerplexityBot), and make a deliberate decision about Google-Extended, which controls AI training and grounding for Gemini separately from Googlebot's normal search crawling. Reference your XML sitemap in the file so every crawler can find it.

Maintain an XML sitemap that lists every indexable URL with accurate lastmod dates, submit it in Google Search Console, and reference it from robots.txt. AI systems that ground answers in search indexes discover and refresh content through the same pipeline, so a clean sitemap keeps new and updated pages visible to them quickly.

Publish a human-readable HTML sitemap page that lists the site's important pages grouped by section, and link it from the footer. It gives crawlers, including AI bots that do not run JavaScript, one flat plain-HTML path to every key URL, and it helps users and agents find pages the main navigation does not surface.

Gemini's Personal Intelligence can draw on a user's Gmail when answering, which means a brand's email newsletters can surface inside a client's AI answers. Run a genuinely useful newsletter your audience opts into, and study the early experiments on how email content gets retrieved; the inbox is becoming another AI visibility surface worth a presence.

Personal Intelligence announcement (Google) Personal Intelligence experiment (iPullRank)