The SEO & GEO Audit Prompt: How to Make Any Website Readable by Google and AI Assistants

Short answer

What is GEO, and why does it need its own audit?

Generative Engine Optimization (GEO) is the practice of making content more likely to be used and cited inside answers written by AI systems. The term was introduced by researchers from Princeton University, Georgia Tech, the Allen Institute for AI and IIT Delhi, who built a benchmark of real queries and found that targeted content changes can boost a source's visibility in generative engine responses by up to 40%.^[1] Among the methods they tested, adding citations to credible sources, relevant quotations and statistics gave the strongest gains.^[1]

AI assistants that search the web generally follow the retrieval-augmented generation (RAG) pattern: they first retrieve relevant documents, then generate an answer based on them.^[2] If a crawler can't retrieve and read your page, it can't cite it. That's why GEO starts with the same technical foundations as SEO. Google itself states that there are no extra requirements to appear in its AI Overviews or AI Mode, and that the same foundational SEO best practices apply.^[3]

So the audit below does both jobs at once: it fixes classic SEO foundations for Google and Bing, and adds the extra signals that help other AI tools understand and quote your site.

The full prompt (copy-ready)

Use it with an AI coding assistant that can read and edit your website's files, such as Claude Code or Cursor. See how to run it below.

Act as a technical SEO and GEO (Generative Engine Optimization) specialist. Audit and improve this website so it is easy for search engines (Google, Bing) and AI assistants (ChatGPT, Perplexity, Gemini, Claude) to crawl, understand and cite.

RULES
- Do not delete or rewrite existing content, text, or features. Only add or improve markup.
- Keep the design and layout visually identical.
- Before changing anything, produce an audit report listing every issue found per page. After changes, produce a summary of exactly what was changed and where.

1. CRAWLABILITY & RENDERING
- Check whether page content is present in the raw HTML (view-source) or only rendered by JavaScript. If it's JS-only, flag it as critical and propose pre-rendering / static generation / SSR.
- Add or fix robots.txt (allow all important pages, reference the sitemap).
- Generate sitemap.xml with every public page, including <lastmod>.
- Add a canonical URL to every page. Check for broken links and redirect chains.

2. SEMANTIC HTML
- Exactly one <h1> per page describing its main topic; logical H2 → H3 hierarchy with no skipped levels.
- Use <header>, <nav>, <main>, <section>, <article>, <footer> where appropriate; text in <p>, lists in <ul>/<ol>.
- Descriptive alt text on every meaningful image (alt="" for decorative ones); width/height set on images.
- Descriptive link text (no "click here"); <html lang="..."> set correctly.

3. ON-PAGE META
- Unique <title> (50–60 chars) and meta description (140–160 chars) per page, including the main keyword naturally.
- Open Graph and Twitter Card tags (title, description, image, url).
- Favicon and viewport meta tag.

4. STRUCTURED DATA (JSON-LD)
- Organization (name, logo, url, contact, sameAs social links) and WebSite on the homepage.
- LocalBusiness if there is a physical location; Service for each service page; FAQPage where there are FAQs; BreadcrumbList on inner pages.
- Validate everything against Google's Rich Results Test format.

5. GEO / AI READABILITY
- Create /llms.txt (Markdown): a short summary of the business, its services, and links to key pages with one-line descriptions.
- Ensure each service page states clearly, in plain text near the top: what it is, who it's for, and the outcome — so AI tools can quote it directly.
- Add an FAQ section where useful, with concise question-and-answer pairs.
- Make sure the business name, location, and contact details are consistent across all pages.

6. PERFORMANCE & MOBILE
- Lazy-load below-the-fold images, use modern formats (WebP/AVIF), preload key fonts, and minimise render-blocking scripts.
- Report likely Core Web Vitals issues (LCP, CLS, INP).
- Confirm the site is fully responsive.

Finish with a prioritized list (critical / important / nice-to-have) of anything you could not fix yourself and that I need to do manually (e.g. Google Search Console submission, Google Business Profile, backlinks).

The rules: why “don't delete anything” comes first

RULES - Do not delete or rewrite existing content, text, or features. Only add or improve markup. - Keep the design and layout visually identical. - Before changing anything, produce an audit report listing every issue found per page. After changes, produce a summary of exactly what was changed and where.

AI assistants are eager to “improve” things. Without limits, they may rewrite copy that already ranks, rename URLs, or change the layout. These three rules keep the job to adding markup, which carries little risk.

• No deletions or rewrites: Google recommends content written for people first, with original information and expertise.^[4] Your existing text is that work, so the audit keeps it.
• Same design: semantic tags and metadata are invisible to visitors, so nothing they see should change.
• Report before and after: you get a reviewable list of issues and an exact change log, so you can check every edit before publishing.

1Crawlability & rendering

- Check whether page content is present in the raw HTML (view-source) or only rendered by JavaScript. If it's JS-only, flag it as critical and propose pre-rendering / static generation / SSR. - Add or fix robots.txt (allow all important pages, reference the sitemap). - Generate sitemap.xml with every public page, including <lastmod>. - Add a canonical URL to every page. Check for broken links and redirect chains.

Is your content in the HTML, or built by JavaScript?

This is the most important check in the whole prompt. Google processes JavaScript sites in three phases (crawling, rendering and indexing), and rendering can be deferred until resources are available.^[5] Most AI crawlers don't render JavaScript at all. A 2024 analysis of real crawler traffic by Vercel and MERJ found that none of the major AI crawlers render JavaScript, including OpenAI's GPTBot, OAI-SearchBot and ChatGPT-User, Anthropic's ClaudeBot and PerplexityBot. They download script files but don't execute them.^[6]

So if your page only says “loading…” until JavaScript builds it, most AI assistants see an empty page. That's why the prompt calls it “critical”. Google recommends server-side rendering, static rendering or hydration as the solution, and describes the older “dynamic rendering” approach as a workaround, not a long-term fix.^[7]

robots.txt

robots.txt is a plain-text file at the root of your domain that tells crawlers which paths they may access. It's an IETF standard, the Robots Exclusion Protocol (RFC 9309).^[8] Google points out that it manages crawler traffic but is not a way to keep a page out of search results. Use a noindex directive for that.^[9] The sitemap protocol also lets you list your sitemap's address in robots.txt with a Sitemap: line, so every crawler can find it.^[10] AI companies publish their own crawler names, for example OpenAI's GPTBot and OAI-SearchBot, which you can allow or block separately.^[11]

sitemap.xml and <lastmod>

An XML sitemap lists the URLs you want crawled, in a format defined by the sitemaps.org protocol and supported by Google, Bing and others.^[10] Google's documentation is precise about the fields: it ignores <priority> and <changefreq>, and uses <lastmod> only when it is “consistently and verifiably accurate”.^[12] That's why the prompt asks for <lastmod> specifically. Only change it when a page's content really changes.

Canonical URLs, broken links and redirects

The same page is often reachable at several addresses (with and without www, with tracking parameters, and so on). A canonical link tells search engines which one is the preferred version. It's an IETF standard (RFC 6596),^[13] and Google treats rel="canonical" as a strong signal when choosing which URL to show.^[14] Broken links waste crawls, and chains of redirects slow down both users and crawlers. Google recommends permanent server-side redirects when a URL changes for good.^[15]

2Semantic HTML

- Exactly one <h1> per page describing its main topic; logical H2 → H3 hierarchy with no skipped levels. - Use <header>, <nav>, <main>, <section>, <article>, <footer> where appropriate; text in <p>, lists in <ul>/<ol>. - Descriptive alt text on every meaningful image (alt="" for decorative ones); width/height set on images. - Descriptive link text (no "click here"); <html lang="..."> set correctly.

Semantic HTML means using tags for what the content is, not how it looks. The HTML Living Standard defines elements like <article>, <nav>, <header> and <footer> so that software can tell main content apart from navigation and boilerplate.^[16] The same structure helps machine readers. The GEO study found that changing how content is presented can change how often AI engines use it as a source.^[1]

Headings: one H1, then H2 → H3 in order

The W3C's accessibility guidance recommends headings that follow the content's structure, with a single main heading and nested levels that aren't skipped. This is how screen readers and other software build an outline of the page.^[17] WCAG 2.2, the international accessibility standard, requires that structure shown visually is also present in the markup (success criterion 1.3.1) and that headings describe their topic (2.4.6).^[18] Google's SEO Starter Guide also recommends headings that help readers move through the text.^[19]

Alt text and image dimensions

Google uses alt text, together with computer vision and the surrounding page, to understand images. It recommends short, descriptive alt text that fits the page's context.^[20] WCAG requires a text alternative for every meaningful image (criterion 1.1.1),^[18] while purely decorative images should have an empty alt="" so assistive technology skips them.^[21] Setting width and height lets the browser reserve space before the image loads, which prevents layout shift. Layout shift is measured by the CLS metric covered in step 6.^[22]

Link text and page language

Google recommends anchor text that is “descriptive, reasonably concise, and relevant” to the page it links to, because it uses that text to understand the target page.^[23] WCAG makes the same point for accessibility: a link's purpose should be clear from its text (criterion 2.4.4).^[18] Finally, the lang attribute on <html> tells browsers, screen readers and translation tools which language the page is in.^[24] WCAG requires it (criterion 3.1.1).^[18]

3On-page meta tags

- Unique <title> (50–60 chars) and meta description (140–160 chars) per page, including the main keyword naturally. - Open Graph and Twitter Card tags (title, description, image, url). - Favicon and viewport meta tag.

Title and meta description

The <title> is the main source Google uses for the clickable headline in search results. Google asks for distinct, descriptive titles for every page and warns against repeating the same boilerplate across a site.^[25] The meta description can be used as the snippet under that headline. Google recommends a unique, accurate summary for each page.^[26]

Open Graph and Twitter / X Cards

When someone shares your link on LinkedIn, Facebook, WhatsApp or X, those platforms read Open Graph tags (og:title, og:description, og:image, og:url) to build the preview card.^[27] X also supports its own twitter:card tags.^[28] Without them, platforms guess, and often pick the wrong image or text.

Favicon and viewport

Google can show your favicon next to your site name in results, provided it is defined in the page head and can be crawled.^[29] The viewport meta tag makes pages display correctly on phones. That matters because Google uses the mobile version of a site for indexing and ranking (mobile-first indexing).^[30]

4Structured data (JSON-LD)

- Organization (name, logo, url, contact, sameAs social links) and WebSite on the homepage. - LocalBusiness if there is a physical location; Service for each service page; FAQPage where there are FAQs; BreadcrumbList on inner pages. - Validate everything against Google's Rich Results Test format.

Structured data is a standard, machine-readable description of what a page is about, written with the Schema.org vocabulary. Google uses it to understand page content and gather information about the world, and recommends the JSON-LD format.^[31] Think of it as an ID card for your business that every machine reads the same way.

• Organization: tells Google your official name, logo, contact details and social profiles (sameAs). This can be used in knowledge panels and to tell your brand apart from others.^[32]
• LocalBusiness: for businesses with a physical location. It can include your address, opening hours and phone number for Google Search and Maps.^[33]
• Service: a Schema.org type for describing a service, its provider and the area it serves.^[34] Google offers no rich result for it, but it still describes your offer clearly to any machine reader.
• BreadcrumbList: shows the page's position in your site, which Google can display in search results.^[35]
• FAQPage: marks up question-and-answer pairs. Since August 2023, Google only shows FAQ rich results for well-known, authoritative government and health websites.^[36] For other sites the markup won't create a visual rich result, but it still describes the questions and answers in machine-readable form.

5GEO & AI readability

- Create /llms.txt (Markdown): a short summary of the business, its services, and links to key pages with one-line descriptions. - Ensure each service page states clearly, in plain text near the top: what it is, who it's for, and the outcome — so AI tools can quote it directly. - Add an FAQ section where useful, with concise question-and-answer pairs. - Make sure the business name, location, and contact details are consistent across all pages.

llms.txt

llms.txt was proposed by Jeremy Howard of Answer.AI in September 2024. It's a Markdown file at the root of a site that gives AI tools a short, clean summary and a curated list of links, instead of making them dig through navigation, scripts and ads.^[40] The format is simple: an H1 with the site name, a short summary in a blockquote, then sections of links with one-line descriptions.^[41]

Quotable “what, who, outcome” statements

RAG systems retrieve passages and generate answers from them.^[2] A short, self-contained sentence near the top of a page, such as “We provide X for Y so that they get Z,” is easy to retrieve and quote accurately. The GEO study found that content written with clear statistics, quotations and cited sources was used more often by generative engines.^[1] This matches Google's own advice to write helpful, people-first content that clearly shows its expertise.^[4]

FAQ sections

Real customer questions, each with a short and direct answer, match how people phrase prompts to AI assistants. They also give retrieval systems clean question-and-answer passages to work with.^[2] Write them for readers first. As noted in step 4, the FAQ rich result itself is now limited to authoritative government and health sites.^[36]

Consistent name, address and contact details

If your website, structured data and Google Business Profile give different names, phone numbers or addresses, machines can't be sure they describe the same business. Google lists relevance, distance and prominence as the main factors in local ranking, and recommends complete, accurate business information.^[42] Your Organization markup should use exactly the same details.^[32]

6Performance & mobile

- Lazy-load below-the-fold images, use modern formats (WebP/AVIF), preload key fonts, and minimise render-blocking scripts. - Report likely Core Web Vitals issues (LCP, CLS, INP). - Confirm the site is fully responsive.

Core Web Vitals

Core Web Vitals are Google's metrics for real-world user experience. Google recommends these thresholds, measured at the 75th percentile of visits:^[43]

INP replaced First Input Delay as a Core Web Vital in March 2024.^[44] Google says its ranking systems use Core Web Vitals, and recommends achieving good scores for success in Search and for a good user experience in general.^[45]

Images, fonts and scripts

• Lazy-loading: use the browser's built-in loading="lazy" for images below the fold,^[46] but never for content visible on first load. Google warns that lazy-loading must not depend on scrolling or clicks, because Googlebot doesn't interact with pages.^[47]
• Modern formats: WebP and AVIF usually compress much better than JPEG or PNG at similar quality.^[48]
• Fonts: preload the fonts you need immediately and use font-display so text shows while fonts load.^[49]
• Render-blocking resources: scripts and stylesheets in the <head> that block the first paint should be deferred, made asynchronous or inlined if critical.^[50]

Responsive design

Because Google indexes the mobile version of your site,^[30] content that is missing or broken on mobile is effectively missing from Google. Check every page on a real phone, and make sure the mobile version has the same content, headings and structured data as the desktop version.^[30]

The final report: what you still do by hand

Finish with a prioritized list (critical / important / nice-to-have) of anything you could not fix yourself and that I need to do manually (e.g. Google Search Console submission, Google Business Profile, backlinks).

An AI assistant can edit your files, but it can't log into your accounts. The prompt ends by listing those jobs for you, sorted by priority:

• Google Search Console: submit your sitemap and track indexing and errors in the Sitemaps report.^[51]
• Bing Webmaster Tools and IndexNow: Bing has its own guidelines,^[52] and the IndexNow protocol lets you notify participating search engines the moment a URL changes.^[53]
• Google Business Profile: complete and verified information improves local visibility. Reviews and links to your business add to its prominence.^[42]
• Backlinks and mentions: earned links and third-party mentions are signals no file change can create. They come from useful content and real relationships.^[19]
• Re-architecture: if step 1 found JavaScript-only rendering, moving to static generation or SSR is a project decision for you to approve, not an automatic edit.^[7]

How to run the prompt

1. Back up your site, or make sure it is under version control (Git), so every change can be reviewed and undone.
2. Open the project folder in an AI coding assistant that can read and edit files, such as Claude Code, Cursor or a similar tool.
3. Paste the prompt and add your domain and business details (name, address, phone, social profiles) so the structured data is accurate.
4. Read the audit report first. Approve fixes before the assistant applies them.
5. Review the change summary, test the pages (view source, Rich Results Test, PageSpeed Insights), then publish.
6. Do the manual list: submit the sitemap in Search Console and update your Business Profile.

Want to see the results first? Our SEO Dashboard shows the kind of visibility data we track, and our article From Keywords to Vectors: The Science of Ranking in AI Search Engines goes deeper into how AI systems retrieve content.