How the Google + AI visibility score works

What goes into the 100 points?

Seven weighted buckets add up to 100 points. Experimental ai.txt and llms.txt checks remain visible but do not add or remove points. If Search Console is not connected, the score is normalized over the other 75 points and clearly marked as missing index coverage.

• Discovery crawler access (10 pts): Can Googlebot, bingbot, OAI-SearchBot, PerplexityBot, and Claude-SearchBot read your site? Training-only crawlers do not affect this score.
• Page schema (15 pts): Whether sampled pages contain valid, relevant JSON-LD that matches the visible content. Schema can support classification and eligible rich results, but special schema is not required for AI features.
• Answer clarity (20 pts): Whether mirrored page content gives clear, direct, complete answers.
• Recent updates (8 pts): How many distinct mirrored URLs had meaningful content updates in the last 30 days. 10 or more earns 8 points, 3 to 9 earns 5, and 1 to 2 earns 2. Do not change dates without changing the content.
• Site basics (12 pts): The average of technical SEO sanity and internal link graph health. Optional llms.txt quality is reported separately and does not affect the score.
• Google indexing (25 pts): The share of inspected, tracked URLs that Google Search Console confirms as indexed. This is measured outcome data, not a readiness guess.
• Page speed (10 pts): The sampled Core Web Vitals and PageSpeed score. A temporary provider outage is excluded rather than awarded free points or counted as a failure.

How is answer clarity calculated?

Answer clarity is a geodeck diagnostic, not a Google or Bing ranking factor. It scores the Markdown body of each tracked URL for patterns that usually make answers easier to scan, verify, and reuse. These eight core signals contribute up to 70 points:

• Question-style headings: At least one H2-H4 that is a natural question (+14). Three or more question headings adds +4.
• Definition-style opening: A 30-120 word lead paragraph that explains the topic directly (+10).
• Lists or structured items (+8).
• Short, scannable paragraphs: Median paragraph length of 90 words or fewer (+8).
• Sensible heading hierarchy without skipped levels (+6).
• Authority markers: Sourced claims, statistics such as "30%", or relevant dates (+6).
• Direct language: At least two uses of "you" or a related form (+4).
• Content depth: 300-3,000 words (+10), or 150-299 words (+6). This is a geodeck heuristic, not a preferred Google word count.

The current model can add points for a 40-80 word bottom-line-up-front opening (+8), numeric fact density (+3 from 0.3 facts per 100 words, +6 from 0.6, or +8 from 1), outbound primary or authoritative sources (+12), attributed quotations (+6), readable prose (+3), self-contained paragraphs (+4), and sameAs entity links (+3 when structured data is available to the scorer). Content older than 90, 180, or 365 days can lose 3, 6, or 10 points. The final score is capped at 100.

This emerging technology, which we formalize under the unified framework of generative engines (GEs), can generate accurate and personalized responses, rapidly replacing traditional search engines like Google and Bing.

— Aggarwal et al., “GEO: Generative Engine Optimization” (arXiv:2311.09735)

How is freshness measured?

We count distinct mirrored URLs whose tracked content was updated in the last 30 days. 10 or more earns 8 points, 3 to 9 earns 5, and 1 to 2 earns 2. Coverage caps the result so a few refreshed pages cannot represent a large site. Update a date only after a meaningful content change; Google explicitly warns against artificial freshness.

How do I push my score from 90 to 100?

The score plateaus around 90 once bot access and schema are clean. The remaining gains usually come from answer clarity, freshness, and internal linking:

1. Open the Audits tab and click Recompute.
2. Sort URLs by answer score ascending and pick the lowest five.
3. Add a question heading and a 30-120 word definition lead.
4. Refresh those URLs from the Index tab.
5. Recompute again after roughly 90 seconds.

Crawler access: let search and AI tools read your site

Search engines and AI tools (ChatGPT, Perplexity, Claude, Gemini) send automated programs called bots to read content. If your robots.txt file blocks a bot, that bot cannot crawl your site. The fix is usually one line of text.

1. Find your robots.txt file. Open a browser and go to yourdomain.com/robots.txt. If you see a page that says "User-agent: *" or lists bots, that file exists and controls who can crawl your site.
2. Check discovery crawlers separately from training crawlers. A site-wide Disallow under Googlebot, bingbot, OAI-SearchBot, PerplexityBot, or Claude-SearchBot can reduce discovery. Blocking GPTBot, ClaudeBot, or Google-Extended is a separate training and model-use choice and does not reduce the geodeck score.
3. Allow the discovery crawlers you want to reach. Use explicit groups only if a broader rule blocks them, and test the complete robots.txt precedence after editing:

   User-agent: Googlebot
   Allow: /
   
   User-agent: bingbot
   Allow: /
   
   User-agent: OAI-SearchBot
   Allow: /
   
   User-agent: PerplexityBot
   Allow: /
   
   User-agent: Claude-SearchBot
   Allow: /

Using WordPress? Install the Yoast SEO or Rank Math plugin. Both have a built-in robots.txt editor under SEO → Tools → File Editor. Add the lines above and save.

Using Squarespace? Go to Settings → Advanced → External API Keys or Custom Code. Squarespace controls robots.txt automatically - contact their support to allowlist specific bots if blocked.

Using Wix? Open the Wix SEO Wiz or go to Marketing & SEO → SEO Tools → Robots.txt. The editor shows your current file; add the Allow entries above.

Full score (10/10): none of the five measured discovery crawlers is blocked. Each blocked discovery crawler removes 2 points. A failed or incomplete audit earns no points until all five are checked. Training crawler policy is shown separately and is score-neutral.

Page schema: help AI identify what your pages are about

Schema markup is hidden code (JSON-LD) that gives search systems explicit clues about what a page contains: an article, product, organization, person, or breadcrumb trail. It does not guarantee rankings, but it can improve classification and eligibility for supported rich results.

1. Use a plugin - the easiest path. For WordPress: install Yoast SEO (free) or Schema Pro (paid). Yoast automatically adds Article and WebPage schema to posts and pages. Enable the structured data feature in the plugin settings.
2. Add an Organization block to your homepage. This anchors your brand as a real entity. Example (paste into your page's HTML, inside a <script> tag):

   <script type="application/ld+json">
   {
     "@context": "https://schema.org",
     "@type": "Organization",
     "name": "Your Company Name",
     "url": "https://yourdomain.com",
     "logo": "https://yourdomain.com/logo.png",
     "sameAs": [
       "https://www.linkedin.com/company/yourcompany",
       "https://twitter.com/yourhandle",
       "https://en.wikipedia.org/wiki/Your_Company"
     ]
   }
   </script>

3. Add sameAs links. The "sameAs" field (shown above) links your brand to official profiles such as LinkedIn, Wikidata, Crunchbase, Wikipedia, or other authoritative pages you control. Add only real profiles.
4. Use FAQPage only when a page is genuinely a question-and-answer page, and do not expect a rich result for a normal commercial site. Choose Article, Product, Organization, WebSite, BreadcrumbList, or another supported type only when it accurately matches the visible page. Google no longer shows HowTo rich results.

   <script type="application/ld+json">
   {
     "@context": "https://schema.org",
     "@type": "FAQPage",
     "mainEntity": [{
       "@type": "Question",
       "name": "What is your product?",
       "acceptedAnswer": {
         "@type": "Answer",
         "text": "Our product is..."
       }
     }]
   }
   </script>

5. Add datePublished and dateModified to articles. If your site publishes blog posts or news, include dates in the Article schema. This gives crawlers explicit freshness metadata.

Check your work. Paste any page URL into Google's Rich Results Test (search for it) to see what schema the page returns and whether it parses correctly.

Schema.org provides a collection of shared vocabularies webmasters can use to mark up their pages in ways that can be understood by the major search engines: Google, Microsoft, Yandex and Yahoo!

— schema.org, “Getting started with schema.org using Microdata”

Full geodeck schema diagnostic (100/100 per page): valid JSON-LD, an Organization or WebSite entity, a relevant supported content type, BreadcrumbList, and no missing properties in the types geodeck checks. This is a completeness heuristic, not a Google requirement or a promise of rich results.

Answer clarity: write content AI quotes

Answer clarity measures whether your pages contain clear, self-contained, verifiable answers. It is geodeck's transparent editorial heuristic, not an official Google, Bing, or OpenAI score. Use it to find weak pages, then apply each suggestion only when it improves the answer for a real reader.

• 1. Question headings (+14 pts, plus +4 for three or more). Rewrite an H2-H4 as a direct question only when the section genuinely answers it. Instead of "Our Features", write "What features does {product} include?". Descriptive non-question headings are also valid when they serve the reader better.
• 2. Definition lead (+10 pts). Open with a 30-120 word paragraph that explains the topic directly when the page needs a definition. For a task page, lead with the outcome or next action instead. The word range is a diagnostic threshold, not a universal writing rule.
• 3. Lists and structured items (+8 pts). Use bullets, steps, or a table when readers need to compare or scan several related items. Keep connected prose as prose.
• 4. Short paragraphs (+8 pts). The audit passes when the median paragraph is 90 words or fewer. Prefer one coherent idea per paragraph, but do not split sentences merely to earn points.
• 5. Clean heading hierarchy (+6 pts). Use headings to reflect real sections and avoid skipping levels when opening a subsection. W3C notes that moving back to a higher-level heading can correctly close a subsection.
• 6. Verifiable claims (+6 core pts, with up to +26 supporting pts for facts, sources, and attributed quotations). Link to primary sources, explain your evidence, and use dates only when they add relevant context. A bare year or invented statistic may trigger the heuristic but does not make a page trustworthy.
• 7. Direct language (+4 pts). Use "you" where it makes instructions clearer, but do not force second-person wording into definitions, legal text, or reference material. This is a readability heuristic, not an official AI requirement.
• 8. Useful depth (+10 pts). The audit rewards 300-3,000 words and gives partial credit from 150 words. Google explicitly says it has no preferred word count, so stop when the page has answered the reader's question completely.

Maximum score: 100/100 per page after the core and supporting signals are combined. Treat 70 as a review threshold, not a guarantee of rankings or citations. Fix missing evidence and incomplete answers before chasing formatting points.

Which checks match official best practices?

Official guidance supports the goal of clear, useful, well-sourced content, but it does not endorse most of geodeck's numeric thresholds. As of 2026, the practical comparison is:

• Strongly supported: people-first completeness, original expertise, clear sourcing, evidence, descriptive headings, accurate updates, and crawlable pages. Google and Bing both recommend these outcomes.
• Supported as structure, not as a quota: headings, lists, tables, and FAQ sections can make information easier to navigate and reference. W3C also recommends meaningful heading hierarchy. No official source requires three question headings or paragraphs below 90 words.
• geodeck-only heuristics: a 30-120 word definition lead, two uses of "you", Flesch targets, fact-density cutoffs, and the 300-3,000 word band. They are useful review prompts, not ranking requirements.
• Do not game freshness or schema. Update dates only after meaningful changes, cite claims that need support, and use structured data that matches visible page content. Google says AI features need no special schema or additional optimization beyond sound SEO and helpful content. Optional llms.txt and experimental ai.txt policy reports therefore do not affect the geodeck visibility score.

Primary sources: Google's generative AI guide, Google's people-first content guidance, Bing's AI Performance guidance, W3C heading guidance, OpenAI crawler guidance, Anthropic crawler guidance, and Perplexity crawler guidance.

The best practices for SEO remain relevant for AI features in Google Search (such as AI Overviews and AI Mode).

— Google Search Central documentation, “AI features and your website”

SEO basics: the foundation every page needs

SEO basics covers the technical signals every page needs before AI and search engines treat it as credible. geodeck checks your sitemap, page titles, meta descriptions, canonical URLs, and hreflang tags. All of these can be managed without writing code if you use a CMS.

• Page title. Give each indexable page a concise, descriptive title that distinguishes it from other pages. geodeck flags titles outside 20-65 characters as a search-display review prompt, not as a ranking failure; Google can truncate or rewrite titles at different lengths.
• Meta description. Summarize the page accurately and make the result useful to searchers. geodeck flags descriptions outside 50-160 characters for review, but Google has no fixed character requirement and may generate a different snippet from the page.
• Canonical URL. A canonical tag identifies the preferred representative among duplicate or very similar URLs. It helps consolidate crawling and signals; it does not prevent a mythical duplicate-content penalty. Most CMS platforms add canonicals automatically. Verify that each indexable page points to the intended absolute URL.
• Sitemap with lastmod dates. A sitemap.xml file lists all your pages with their last-updated date, helping crawlers find and prioritize fresh content. WordPress: install Yoast or Google XML Sitemaps - they generate sitemaps automatically. Submit the sitemap URL to Google Search Console (Search Console → Sitemaps → Add new sitemap). geodeck auto-detects and submits your sitemap daily.
• Hreflang (multi-language sites only). If your site serves content in more than one language, add hreflang link tags so search engines serve the right language to the right audience. Your CMS multilingual plugin (WPML, Polylang, Weglot) should add these automatically - check the plugin settings to ensure it is enabled.

Score 100/100: a reachable sitemap, indexable sampled pages, intended canonicals, reciprocal hreflang where the site is multilingual, and accurate lastmod values when supplied. Optional lastmod and x-default values are not required for full credit. Title and description length flags are informational display heuristics.

Internal links: connect your content so nothing is orphaned

Internal links are links from one page on your site to another page on the same site. They help AI crawlers discover all your content and understand which pages are most important. An orphan page - one with no links pointing to it - is invisible to crawlers that start from your homepage.

What is an orphan page? Any page that cannot be reached by following links from your homepage is an orphan. geodeck detects orphans by mapping every link in your mirrored content. The audit shows the number of orphan pages - the goal is zero.

1. Link from high-traffic pages to lower-traffic ones. Open your most visited pages (check Google Analytics or Search Console) and add 2-3 links to related pages that have fewer visitors. Use descriptive anchor text - "learn more about our pricing" is better than "click here".
2. Add a Related Articles or Related Posts section. Most CMS themes support this out of the box or as a plugin. Every article that shows related posts automatically links to other content, eliminating orphans quickly.
3. Update your navigation menu. Ensure every important page appears in your main menu or footer menu. Pages reachable from global navigation are never orphans.
4. Fix broken internal links. A link to a page that no longer exists is a broken link. geodeck counts these in the audit. Search your CMS for any redirect or 404 error log, or ask your developer to run a broken-link checker.
5. Use your sitemap as a checklist. Export your sitemap URL list. For any URL not linked from at least one other page, find the most relevant existing page and add a link there.

Score 100/100: zero orphan pages (every page linked from at least one other), zero broken internal links, and a healthy hub structure where important pages receive links from many others.