Skip to main content
Some content formats are specifically the ones AI models reach for when grounding answers to high-intent queries. Comparison pages answer “X vs Y,” buying guides answer “how do I choose,” best-of lists answer “what should I get,” and structured spec tables answer “what are the differences.” Pages that fit these molds get cited more reliably than prose-only competitors for question-driven queries — and sites without any of them lose the high-intent traffic AI sends.

​
Methodology

Cited samples up to 5 pages from your site and classifies each against four AI-friendly content patterns. The signal counts distinct patterns found across the entire sample, not occurrences — finding a comparison page on one URL and a buying guide on another counts as two patterns. The same pattern repeated on multiple pages still counts as one. The four patterns:
  • Comparison page — title or H1 contains “vs”, “versus”, or “compare/comparison”. These pages answer head-to-head queries that AI models route to grounded sources almost exclusively (training data alone can’t reliably compare current product versions).
  • Buying guide — title or H1 contains “buying guide”, “how to choose”, or “what to look for”. These pages answer purchase-intent queries where AI defers to expert-framed criteria.
  • “Best of” page — title or H1 contains “best” or “top N” (where N is a number). These pages answer ranking queries; AI models cite them for shortlist-shaped answers.
  • Structured specifications — page contains a <table> with more than 3 rows OR a <dl> definition list with more than 3 terms. AI models extract structured spec data more reliably than prose specs.
Scoring rewards breadth — having multiple pattern types signals to AI that your site is a citation-ready source across query shapes:
  • 3 or more pattern types found → 7/7. Full citation-format coverage.
  • 1 or 2 pattern types found → 4/7. Partial coverage — some query shapes covered, others not.
  • 0 pattern types found → 0/7. None of the AI-friendly molds present on sampled pages.
The signal scores out of 7. Status thresholds: pass at 5/7, partial at 3/7, fail below. Sitemap evidence adds context but not score. Even when a pattern isn’t found in the 5-page sample, the scanner scans the sitemap for URLs matching pattern shapes — /compare-X-vs-Y, /buying-guide, /best-X, /top-10-X. Matches are reported as evidence (“18 best-of pages in sitemap but not in this scan’s sample”) so brands with deep catalogs see their full content surface even when the sample missed it. Sitemap matches don’t lift the score because the scanner can’t verify the pages actually have AI-friendly structure without crawling them.

​
Verification

You can verify our finding yourself in a browser. Step 1: Open the pages we sampled. Cited reports the URLs we tested. Open each in a new tab. Step 2: Check the title and H1 against the patterns. For each page, look at the <title> tag (browser tab) and the first <h1> (usually the page’s main heading). Compare against the four patterns:
  • “vs”, “versus”, “compare”, “comparison” → comparison page
  • “buying guide”, “how to choose”, “what to look for” → buying guide
  • “best”, “top 10”, “top 25” → best-of page
Step 3: Look for structured spec tables or definition lists. In the Console, run: 
({ tablesWith4PlusRows: [...document.querySelectorAll('table')].filter(t => t.querySelectorAll('tr').length > 3).length, dlWith4PlusTerms: [...document.querySelectorAll('dl')].filter(d => d.querySelectorAll('dt').length > 3).length })
Either value greater than 0 means the page satisfies the structured specs check. Step 4: Scan your sitemap for additional matches. Open https://yoursite.com/sitemap.xml. Search the URL list for /vs-, /-vs-, /compare/, /buying-guide, /how-to-choose, /best-, /top- followed by a digit. URLs matching these patterns are pages your site likely has but the 5-page sample didn’t include. Cited reports these in the signal’s evidence. If your verification disagrees with Cited’s finding, that’s a bug — let us know.

​
Technical detail

AI citation studies (across Perplexity, ChatGPT search, Gemini grounding) consistently show that pages structured around answer-shaped content earn citations more reliably than prose-only pages for the same topic. Cited’s scanner doesn’t measure citation rates directly — that’s tracked separately by the AI Visibility surface — but the four patterns here are derived from observed citation behavior across thousands of grounded responses. Pattern detection logic. For each crawled page, the scanner runs four checks:
  • Comparison page — regex /\bvs\.?\b|\bversus\b|\bcompar(e|ison)\b/i against the lowercased concatenation of document.title and the first <h1> text. Matches “X vs Y”, “X vs. Y”, “compare X and Y”, “X comparison”.
  • Buying guide — regex /buying guide|how to choose|what to look for/i against the same combined string.
  • Best-of page — regex /\bbest\b|\btop\s+\d+/i against the same combined string. Matches “best X”, “top 10 X”, “top 25 X” but not “top X” without a number.
  • Structured specifications — DOM query on the rendered page: count of <table> elements with more than 3 <tr> children OR <dl> elements with more than 3 <dt> children. Either condition satisfies.
Patterns are tracked as boolean flags per page, then aggregated to a Set across all sampled pages. The Set’s size determines the tier. Sitemap supplementary detection. When a pattern isn’t found in the crawled sample, the scanner checks the sitemap URL list against URL-path regexes:
  • Comparison: /(compare|vs-)|(-vs-)/i — catches /compare-x-y, /x-vs-y, /vs-x slug patterns
  • Buying guide: /(buying-guide|how-to-choose)/i
  • Best-of: /(best-|top-\d)/i — catches /best-coffee-makers, /top-10-laptops slug patterns
Matches are listed as evidence (“Additional AI-friendly content detected in sitemap but not in this scan’s 5-page sample”) with up to 5 example URLs. The sitemap pass never changes the score — the scanner can’t verify the pages actually have the right title/H1/structure without crawling them, and sitemap URL slugs alone aren’t sufficient evidence. Edge cases the scanner handles:
  • Multiple pages with the same pattern — five product comparison pages still count as one Comparison pattern. The Set deduplicates. The signal rewards breadth across pattern types, not depth within a type.
  • Patterns in unrelated copy — a blog post titled “What I Learned at the Best AI Conference” matches the best-of regex via the word “Best”. The scanner accepts this false positive in favor of catching the common case (genuine “best of” lists). AI models filter at consumption time.
  • Pattern in title but not H1 — the scanner checks the concatenation of both, so either alone satisfies. Pages with <title>Comparison: X vs Y</title> and an H1 of just “X” still match.
  • Definition lists for FAQs — sites that use <dl><dt>Question?</dt><dd>Answer.</dd></dl> for FAQ blocks satisfy the structured-specs check. This overlaps with the Answer-Block Formatting signal, which detects FAQs separately. Cited intentionally allows the overlap because the two signals score different things — FAQ presence vs. structured-data discipline.
  • Sitemap URL false positives — a URL like /blog/the-vs-debate matches vs- via substring. The sitemap pass is evidence-only and doesn’t score, so false positives are reported but don’t affect the result.
What this signal does not measure:
  • Citation rate per content type. The scanner doesn’t measure how often each page type is actually cited by AI models. That’s tracked by the AI Visibility surface, not GEO Score. Studies show these formats correlate with citation but the signal here is structural presence, not measured outcome.
  • Quality within the format. A 200-word “Best CRMs for SMBs” page with no analysis scores the same as a 2,000-word version with detailed comparisons. Content depth is measured separately.
  • Comparison freshness. “X vs Y 2019” matches the comparison pattern but is unlikely to be cited for current queries. Content Freshness scores this independently.
  • Pattern intent depth. “Best Sunday for hiking” matches the best-of regex via “best” but doesn’t carry buyer intent. The scanner accepts the false positive.
For brands scoring 0-3/7, the highest-leverage fix is identifying one query in your category that AI models field daily (“best CRM for SMBs”, “Hubspot vs Salesforce”, “how to choose a CRM”), and publishing a single page in the matching format. One well-structured comparison or buying guide lifts the signal score from 0 to 4/7. Adding two more pattern types brings it to 7/7. Most brands underperform on this signal because their existing content is prose-shaped — the fix isn’t more content; it’s structuring existing content into AI-friendly molds. See also: Answer-Block Formatting, llms.txt Presence, JSON-LD Structured Data.