An AI citation audit tells you whether AI systems mention your brand when users ask questions your brand should answer. That’s it. No mystery. If someone asks ChatGPT “best CRM for small businesses” and your CRM isn’t named, you have a citation problem. If Gemini defines your industry category and cites 4 competitors but not you, that’s a citation gap you can measure and fix. We started running these audits at ScaleGrowth.Digital, a growth engineering firm based in Mumbai, in late 2025 when clients noticed something unsettling: their Google rankings were stable, their traffic was decent, but leads from “I heard about you from AI” were going exclusively to competitors. One B2B SaaS client with 47 first-page rankings wasn’t cited in a single ChatGPT response across 30 industry-relevant prompts. Zero. Their competitor with weaker SEO appeared in 22 out of 30. The reason was structural. It wasn’t about content quality. It was about how that content was formatted, referenced, and connected across the web. AI models don’t rank pages. They extract entities, match definitions, and cite sources that give them clean, attributable answers. This 15-point framework is what we use internally. Every check maps to a specific signal that AI models use when deciding whether to cite your brand. We’ve run it across 40+ brands in financial services, SaaS, healthcare, and ecommerce. The average score on first audit? 38 out of 150. Most brands fail 9 of the 15 checks outright.

“The brands getting cited by AI aren’t necessarily the ones with the best content. They’re the ones whose content is structured for extraction. There’s a difference between writing a great article and writing an article that an LLM can confidently attribute to you. This framework tests for the second thing.”

— Hardik Shah, Founder & Digital Growth Strategist at ScaleGrowth.Digital

Check 1: Entity Consistency

AI models build entity profiles from every mention of your brand across the web. If your About page says “Acme Solutions,” your LinkedIn says “Acme Solutions Inc.,” and your Crunchbase profile says “Acme,” the model has 3 conflicting entries for what should be 1 entity. Inconsistency reduces confidence. Lower confidence means fewer citations. What to audit: Compare these five data points across all profiles:

• Brand name
• Founding year
• Founder name
• Headquarters location
• Core description

Check your website, Google Business Profile, LinkedIn, Crunchbase, Wikipedia (if applicable), and your top 10 third-party mentions. Count discrepancies. Pass criteria: Fewer than 2 discrepancies across all profiles. Brand name is identical everywhere. Core description uses the same language within 90% similarity.

Check 2: Definition Blocks

When an AI answers “what is [your brand]?” it needs a clean, extractable definition. This is usually a 2-3 sentence block that states what you are, what you do, and who you serve. If this block doesn’t exist on your homepage or About page, the AI has to assemble one from scattered fragments, which usually means it just skips you and cites a competitor who has a clear definition. What to audit: Check your homepage, About page, and any “What is [brand]” page for a clear definition block within the first 300 words. The block should be a standalone paragraph, not buried inside a longer narrative. Pass criteria: A clear, extractable 2-3 sentence definition exists on at least 2 pages. It includes: brand name, category, primary offering, and target audience.

Check 3: Schema Markup Presence

Schema.org structured data gives AI models machine-readable entity information. Organization schema tells them your company details. FAQ schema tells them your Q&A content is verified. Article schema tells them who authored your content and when. Without schema, you’re asking AI models to parse your HTML and guess. They will. They’ll just guess wrong more often. What to audit: Run your homepage, About page, and 5 key content pages through Google’s Rich Results Test. Check for these schema types:

• Organization
• WebSite
• Article
• FAQ
• Product
• BreadcrumbList

Note which are present and which are missing. Pass criteria:

• Organization schema on homepage with name, URL, logo, founding date, and sameAs links
• Article schema on blog posts with author, datePublished, and dateModified
• FAQ schema on at least 3 pages

Score 8+ requires all 6 schema types present where applicable.

Check 4: AI Crawler Access

If your robots.txt blocks GPTBot, Google-Extended, ClaudeBot, or PerplexityBot, you’ve told those AI systems not to read your site. They’ll respect that. They also won’t cite you. We see this surprisingly often: brands that blocked AI crawlers in a knee-jerk reaction to ChatGPT’s launch in 2023 and never unblocked them. They’re invisible to AI and don’t know why. What to audit: Check your robots.txt for user-agent blocks on:

• GPTBot and ChatGPT-User
• Google-Extended and Googlebot (which also affects AI Overviews)
• ClaudeBot and anthropic-ai
• PerplexityBot
• Applebot-Extended

Also check for overly broad disallow rules that might accidentally block AI crawlers. Pass criteria: All major AI crawlers are allowed to access your content pages. Blocking specific directories (like /admin/ or /staging/) is fine. Blocking entire domains or all content pages is a fail.

Check 5: llms.txt File

The llms.txt standard (published at llmstxt.org) is a plain-text file at your domain root that tells AI models what your site is about, what your key pages are, and how to cite you. Think of it as robots.txt for AI understanding rather than crawling. As of March 2026, roughly 12% of the top 10,000 websites have implemented it. Early adopters report 15-25% higher citation rates in Perplexity specifically. What to audit: Check if yourdomain.com/llms.txt exists. If it does, verify it includes: brand description, key pages with descriptions, preferred citation format, and contact information. If it doesn’t exist, that’s a 0. Pass criteria: llms.txt file exists, is properly formatted per the spec, includes at least 10 key pages, and has been updated within the last 90 days.

Check 6: Answer Block Format

AI models preferentially cite content that’s already structured as a direct answer. This means: a question as a heading (H2 or H3), followed immediately by a 40-60 word answer paragraph, followed by supporting detail. If your content buries the answer in paragraph 4 of a 2,000-word section, the AI model has to extract it. Content where the answer leads is 3x more likely to be cited, based on citation testing we’ve done across 1,200 content pages. What to audit: Review your top 20 content pages. For each, check: Does the first paragraph after each H2 contain a direct, standalone answer? Could you extract that paragraph and it would make sense on its own? Or does the content start with background, context, and preamble before eventually getting to the point? Pass criteria: At least 15 of 20 pages lead with direct answers after headings. Answer paragraphs are 40-80 words and self-contained.

Check 7: FAQ Structure

FAQ pages and FAQ sections within content pages are citation gold. When a user asks an AI a question that matches one of your FAQs, and you have FAQ schema marking it up, the AI can pull a verified, structured answer directly. 67% of Perplexity citations in our testing linked to pages with FAQ schema, compared to 23% for pages without it. What to audit: Count how many pages on your site have FAQ sections. Check if those sections use proper FAQ schema markup. Verify the questions match real user queries (check Google Search Console’s query data or AnswerThePublic for question patterns). Pass criteria: At least 10 pages with FAQ sections. All FAQ sections have valid FAQPage schema. Questions match actual user search queries, not internally invented questions that nobody asks.

Check 8: Author Entity

AI models evaluate source credibility partly through author identity. Content published by “Admin” or with no author attribution gets lower trust signals than content by a named author with a LinkedIn profile, a consistent publishing history, and schema markup connecting them to the content. Google’s E-E-A-T framework feeds directly into AI Overview citations, and author entity is a core component. What to audit: Check your blog posts and key content pages for:

• Visible author name and bio
• Author schema (Person schema linked via Article schema)
• Dedicated author page on your site
• Consistent author presence across the web (LinkedIn, industry publications, speaking engagements)

Pass criteria: All content pages have named authors with bios. Author schema is present with sameAs links to external profiles. At least 1 author has published content on 3+ external sites in the past 12 months.

Check 9: Third-Party Mentions

AI models don’t just read your website. They read what other websites say about you. If 15 industry publications mention your brand in “best [category]” articles, the AI has 15 independent signals that you belong in that category. If zero external sources mention you, the AI only has your own claims, which carry less weight. This is the AI equivalent of backlinks, but for entity recognition. What to audit: Search Google for your brand name (in quotes) minus your own domain. Count third-party mentions in the past 12 months and categorize them:

• Listicle mentions (“best X” articles)
• News coverage
• Review sites
• Industry directories
• Social mentions from authoritative accounts

AI models weight authoritative sources more heavily than random blog mentions. Pass criteria: At least 20 third-party mentions in the past 12 months. At least 5 from authoritative sources (industry publications, major news outlets, established directories). At least 3 “best [category]” listicle inclusions.

Check 10: Content Freshness

AI models consider when content was last updated. A “Best CRM Software 2024” page that hasn’t been touched in 18 months is less likely to be cited than one updated in January 2026. This is especially true for AI Overviews, where Google explicitly factors in freshness signals. Our data across 200 tracked pages shows that pages updated within the last 6 months are cited 2.4x more than pages older than 12 months, controlling for all other factors. What to audit: Check your top 30 content pages for last-modified dates. Verify the dates are accurate (not just auto-generated by your CMS on every cache clear). Look for visible “Last updated” dates on the page, and dateModified in Article schema. Pass criteria: 80% of key content pages updated within the last 6 months. All pages show a visible last-updated date. dateModified schema matches the actual last-updated date.

Check 11: Citation Testing — Direct Prompts

This is where you find out the truth. Run your 20 test prompts across ChatGPT (GPT-4o), Gemini Advanced, and Perplexity Pro. Record whether your brand is mentioned in each response. This isn’t theoretical. This is your actual citation rate right now. What to audit: Run all 20 prompts on each platform (60 total responses). For each response, record:

• Was your brand mentioned?
• If yes, was it the primary recommendation or a secondary mention?
• Was the information about your brand accurate?
• Did the AI link to your website? (Perplexity does this; others generally don’t.)

Pass criteria: Brand mentioned in at least 40% of responses across all platforms (24 out of 60). Primary recommendation in at least 15% of responses (9 out of 60). Zero factual errors about your brand in any response.

Check 12: Platform Coverage

Different AI platforms have different citation behaviors. ChatGPT tends to cite well-known brands and recently crawled content. Gemini leans on Google’s own index and Knowledge Graph data. Perplexity cites web sources explicitly with links. AI Overviews pull from ranked pages and structured data. A brand might score well on Perplexity but poorly on ChatGPT, or vice versa. What to audit: From your Check 11 data, calculate your citation rate per platform. Also test 5 prompts on Claude and 5 on Apple Intelligence (if accessible) for additional coverage data. Identify which platforms cite you most and least. Pass criteria: Cited on at least 3 of 4 major platforms (ChatGPT, Gemini, Perplexity, AI Overviews). No platform has a 0% citation rate. Cross-platform citation variance is less than 30 percentage points (e.g., not 60% on Perplexity and 5% on ChatGPT).

Check 13: Content Structure for Extraction

AI models extract information in chunks. They grab headings, grab the paragraph below, and evaluate whether that chunk answers the query. If your content uses clear H2/H3 hierarchy, short paragraphs (under 150 words), and topic sentences that summarize the paragraph, extraction is easy. If your content is 500-word paragraphs with no subheadings, the model struggles and moves on to a source that’s easier to parse. What to audit: Select 10 key content pages. For each, check:

• Average paragraph length
• Heading density (at least 1 H2/H3 per 300 words)
• Presence of summary sentences
• Use of bold for key terms
• Whether each section could stand alone as an answer

Run a readability check (aim for Grade 8-10 reading level). Pass criteria: Average paragraph length under 100 words. Heading density of 1 per 250-350 words. Key terms bolded consistently. Reading level between Grade 8 and Grade 11.

Check 14: Internal Linking for Entity Reinforcement

Internal links tell AI models how your content relates to each other and which pages are most important. A “What is [your product category]” page that links to your product page, your case studies, and your comparison pages creates an entity cluster. The AI sees that you don’t just mention the topic once; you’ve built an interconnected knowledge base around it. We’ve measured that pages with 8+ internal links pointing to them get cited 40% more than pages with fewer than 3 internal links. What to audit: Use Screaming Frog or your CMS analytics to map internal links to your top 20 pages. Count inbound internal links per page. Check if your cornerstone content pages (product pages, category definitions, comparison pages) are well-linked from supporting blog posts and resource pages. Pass criteria:

• Top 10 priority pages each have 8+ internal links pointing to them
• Every blog post links to at least 2 cornerstone pages
• No orphan pages (pages with 0 internal links) among your top 50 content pages

Check 15: Competitor Citation Comparison

Your citation rate means nothing in isolation. What matters is your citation rate relative to competitors. If you’re cited in 30% of responses but your top competitor is cited in 70%, you’re losing the AI visibility war. This check benchmarks you against 3-5 direct competitors across the same prompt set. What to audit: Run your 20 test prompts again (or use the data from Check 11). This time, track every brand mentioned in every response. Build a frequency table with these columns:

• Brand Name
• Times Cited
• % of Responses
• Avg. Position (primary vs. secondary)

Rank yourself against competitors. Pass criteria: Your citation rate is within 20 percentage points of the top-cited competitor. You are not the least-cited brand among your direct competitors. You appear as primary recommendation at least once across the 60 responses.

Your total falls into one of four tiers. Each tier maps to a general citation rate range we’ve observed across the 40+ audits we’ve completed. A useful benchmark: across our client base, scores improved by an average of 43 points in the first 90 days of focused work. The biggest gains came from:

• Check 2 (definition blocks)
• Check 4 (AI crawler access)
• Check 7 (FAQ structure)

These are quick fixes with outsized impact. Check 9 (third-party mentions) takes longest to improve since it requires external outreach and PR.

Five patterns show up in nearly every audit we run.

Failure 1: No definition block anywhere on the site

The brand has 200 pages of content but not a single clear, extractable statement of what they are. Their About page is a brand story. Their homepage is benefit-driven marketing copy. Neither gives an AI a clean answer to “what is [brand]?” Fix: Add a 2-sentence definition to your homepage and About page. Takes 15 minutes. Impact shows up in 4-6 weeks.

Failure 2: AI crawlers blocked in robots.txt

We found this in 23% of audits. The fix takes 30 seconds. The cost of not fixing it is complete invisibility in AI responses. Check your robots.txt right now. Seriously. If you see User-agent: GPTBot followed by Disallow: /, that’s your problem.

Failure 3: Content buried behind preamble

Pages that take 200+ words to get to the actual answer. AI models scan the first few paragraphs after a heading. If those paragraphs are context-setting rather than answer-delivering, the model moves on. This is the biggest gap between traditional SEO content (which often front-loads context) and AI-optimized content (which front-loads the answer).

Failure 4: Zero third-party mentions

Startups and mid-market brands suffer from this most. They have good content but no external references. The AI has only the brand’s own claims to go on. Without independent confirmation from external sources, the model’s confidence stays low. Fix: An active digital PR and guest publishing program. This is a 6-12 month play, not a quick fix.

Failure 5: Schema exists but is incomplete or invalid

The site has Organization schema, but it’s missing the sameAs links that connect it to social profiles and external references. Or they have Article schema but no author information. Partial schema is better than none, but it leaves citation performance on the table. Run every schema type through the Schema Markup Validator and fix all errors.

You can absolutely run it yourself. That’s why we published it. Checks 1-10 require basic technical knowledge (checking robots.txt, validating schema, reviewing page structure). Checks 11-15 require patience and a spreadsheet. Nobody needs to hire us or anyone else to run this checklist. Where a specialist adds value is in three areas.

Interpretation

Knowing your score is 52 out of 150 is useful. Knowing that your specific pattern of failures (strong on Checks 1-5, weak on Checks 6-10) indicates a content structure problem rather than a technical problem is what turns a score into an action plan. Pattern recognition across hundreds of audits matters.

Fix prioritization

With 15 checks and limited resources, sequencing matters. Fixing Check 4 (AI crawler access) before Check 9 (third-party mentions) is obvious. But should you invest in Check 7 (FAQ structure) or Check 8 (author entity) first? That depends on your industry, competitive field, and existing content assets. A specialist has the data to make that call.

Ongoing optimization

The audit is a snapshot. Improving and maintaining your citation performance is continuous work that involves content production, technical implementation, PR outreach, and cross-platform monitoring. Most in-house teams don’t have the bandwidth to sustain this alongside their existing SEO programs. If you want to start with the automated version, the AI Visibility Checker gives you a scored report covering 12 of these 15 checks in under 10 minutes. It’s free for a single-domain scan.