When a user asks ChatGPT "what's the best project management tool for remote teams?" or Perplexity "which CRM should I use for a small agency?" — the answer doesn't come from a ranked list of blue links. The AI reads, interprets, and synthesizes content from across the web, then generates a recommendation.
The sites that get mentioned aren't necessarily the ones ranking #1 on Google. They're the ones that AI systems can find, understand, extract from, and describe accurately when asked — not just on your site, but everywhere the AI looks.
We built AIVerdict to measure exactly this. After auditing thousands of websites, we identified 10 distinct pillars that determine whether AI will recommend a site. They split into two layers:
- CORE (5 pillars) — what AI can see, render, and extract from your site itself
- AURA (5 pillars) — what AI actually says about you on real user queries against ChatGPT, Gemini, and Perplexity
- + Risk — a modifier that subtracts when AI surfaces controversies
Your overall score is the geometric mean of CORE and AURA — both halves have to work. A site can pass every CORE check and still be invisible if AI doesn't recognise the brand. A site with great AI brand recognition can still get nothing if its content is locked behind JavaScript.
Why traditional SEO metrics aren't enough
Google's algorithm evaluates hundreds of ranking signals: backlinks, page speed, keyword relevance, user engagement. These still matter. But AI systems like ChatGPT, Perplexity, and Gemini operate fundamentally differently:
- They don't rank pages — they synthesize answers from multiple sources
- They need structured, extractable content — not just keyword-optimized copy
- They can't execute JavaScript in most crawling and live-grounding scenarios — if your content is JS-rendered, it might be invisible
- They evaluate trust through training-data presence — not backlinks. Being in Wikipedia, Common Crawl, and news cycles matters more than how many sites link to you
- They pick winners on category queries — "best X for Y" answers surface 3–5 named brands. If you're not in that shortlist on category prompts, you don't exist to the AI's user
A site can rank #1 for its target keywords and still be completely invisible to AI recommendation engines. That's the gap the 10 pillars measure — both the onsite gap (CORE) and the offsite gap (AURA).
The 5 onsite pillars
CORE measures whether AI can find, render, and extract from your site at all. These are the levers you control with code changes — robots.txt edits, SSR, schema, structured content.
Do AI training crawlers allow-list your site?
Training Access is the foundation. We check robots.txt allow/deny status for 10 AI training crawlers: GPTBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot (Perplexity), Google-Extended (Google AI training), CCBot (Common Crawl), Bytespider (TikTok), Applebot (Apple Intelligence), Amazonbot (Alexa/Rufus), FacebookBot (Meta AI), and Cohere-AI.
Common failures: a default-deny robots.txt that leaked from a staging environment, Cloudflare "Super Bot Fight Mode" silently blocking crawler IPs even when robots.txt allows them, cookie walls that crawlers can't pass. A single missing Allow: / line can drop your score 60+ points — and the fix is also one line.
This pillar is part of the free CORE check — runnable on any URL with no signup.
Can live AI agents reach and render your pages?
Different from Training Access. When a user asks ChatGPT to "go check this site" or uses Perplexity for research, a live AI agent fetches your page on demand. These agents ignore robots.txt (REP applies to indexers) but they do fail on: JavaScript-only SPAs, full-page cookie walls, slow TTFB, multi-hop redirects, and DDoS-protection challenges.
Common failures: React/Vue/Angular SPAs without SSR ship an empty <div id="root"> as the initial HTML — live agents see nothing. GDPR consent overlays that block the DOM until accepted. Region redirects that loop the agent. Cloudflare's "I'm Under Attack" mode killing the fetch silently.
Also part of the free CORE check. Together with Training Access, it answers the most-asked question agencies get: "is AI blocked from our client's site?"
Can AI extract specific, citable facts?
Content Quality uses LLM-based semantic analysis to evaluate clarity, citation readiness, competitive positioning, and topical depth across 8–12 sampled pages. It's not a grammar or readability check — beautifully written prose can score lower than plain text if the plain text contains concrete facts AI can quote.
Common failures: marketing pages full of "empower your team" buzzwords with no specific claims, feature pages that list names without explanations, pricing hidden behind contact-sales forms (AI gets "how much does X cost?" constantly — if you don't answer it, a competitor will), no comparison content to help AI position you against alternatives.
This is where agencies get the richest actionable insights: each finding maps to a specific page-level recommendation with an estimated score impact.
Can AI generate direct answers from your content?
Answerability scores five sub-checks: definition clarity (are key terms defined as "X is Y"?), structured lists, parseable tables, FAQ/Q&A blocks with schema markup, and how-to step sequences. The exact sub-dimension names vary by site type (SaaS, e-commerce, news, local, blog, generic) so the score reflects what AI actually looks for in your category.
A site might have great content a human can easily understand but if it's all dense paragraphs with no structural cues, AI systems struggle to extract specific answers. FAQ schema is one of the fastest CORE wins — it directly increases the rate at which AI cites you in generative answers.
Is your content readable without executing JavaScript?
Six sub-checks: schema (20%), semantic HTML (25%), alt text (20%), content structure (20%), OG/meta tags (10%), and consent barriers (5%). We test the ratio of meaningful content in raw HTML vs rendered HTML across sampled pages. A site built entirely on a JS framework with no SSR can score near-zero on Extractability — meaning AI systems literally can't read any of the content, regardless of how good it is.
This is often the highest-leverage CORE fix because it's near-binary: either AI can read your content or it can't. Migrating to SSR (Next.js, Remix, Nuxt) or SSG (Astro, Eleventy) lifts the score immediately.
The 5 offsite pillars
AURA measures what AI actually says about you when probed. Every AURA pillar fires 10 user-realistic prompts across ChatGPT, Gemini, and Perplexity (30 LLM calls per audit) on category-defining queries you'd actually be asked, then parses the responses. These pillars shift slower than CORE because they depend on third-party signals.
Do AI systems already know your brand without being shown?
We probe ChatGPT, Gemini, and Perplexity with grounding/web-search OFF, then ask them to describe your brand from training data alone. The score measures whether each model recognised you, how accurately it described you, and whether it confused you with another entity sharing your name.
The slowest pillar to move (3–12 months for changes to surface in next-gen models) but the most durable competitive moat. Levers: Wikipedia article (highest single signal), heavily-scraped press coverage (TechCrunch, Verge, Wired, Hacker News, Reddit), canonical category lists (G2, Capterra, ProductHunt).
How often does AI mention you vs competitors on category queries?
Probes 10 prompts across 7 intent buckets: compare ("X vs Y"), recommend ("best X for Y"), explain, troubleshoot, integrate, price-research, alternatives. We count brand mentions in each response and calculate your share. The bucket weights are calibrated from real probe research — "recommend" was originally weighted 50% but downweighted to 10% after data showed only a 2% vendor-cite rate.
Most actionable AURA pillar because content moves it within weeks rather than months. Lever priority: own category content (fair "X vs Y" comparison pages tend to surface alongside whoever they compare), Reddit/HN/forum participation, presence in canonical "best X" listicles.
When AI cites you, does it cite your own pages or only third-party sources?
Same 10 prompts as Share of Voice but with live grounding ON. We parse each AI response for source URLs and classify each as your-domain, third-party-positive, third-party-neutral, or third-party-negative. High Citation Depth = you own the narrative (AI quotes your docs/blog/product pages). Low = AI cites Stack Overflow, Reddit, or reviews to describe you.
Lift it with structured docs (FAQ schema, comparison tables, clear definitions), high Agent Access (so live grounding can fetch you), and answer-shaped content phrasing ("X is Y because Z").
Does third-party content mirror your brand's positioning?
Crucial caveat: Alignment is not a sentiment score. A critical review that accurately describes your positioning scores higher than a glowing review that mischaracterises what you do. We extract your stated positioning from your site (hero, About page) using LLM analysis, then compare against how AI describes you when grounded with third-party sources.
If your site says "fast-onboarding CRM for agencies" but live AI says "developer tooling," Alignment drops. The lever isn't sentiment manipulation — it's narrative consistency. Engage with reviews, contribute to forum threads, write a Wikipedia article that matches your positioning.
When AI lists vendors, where does it rank you?
Different from Share of Voice. SoV counts mentions; Competitive Rank measures ordering. AI tends to surface 3–5 named brands per query. Position 10/10 is still better than not being mentioned at all, but position 1–3 wins clicks. We parse numbered lists, mention-order in prose, recommended-winner statements, and comparison-table column order to compute your average rank.
To lift Rank: get strong AI Brand Recognition (recognised brands surface earlier), high Citation Depth (first-party data improves AI's confidence in placing you high), and specialise instead of generalise. "Best CRM for SaaS agencies under 50 people" is a winnable narrow category. "Best CRM" puts you against Salesforce.
The modifier
Does AI surface controversies about your brand?
Risk is not a 6th AURA pillar — it's a subtractor on the AURA composite score that visually tints the AURA ring orange. Categories tracked: lawsuits, security incidents, regulatory actions, layoffs covered as crisis events, executive scandals, product safety issues, and sustained negative review cycles.
Time-decayed: incidents in the last 90 days weigh full strength; 90 days–1 year weigh half; 1–3 years quarter; older than 3 years near zero. A 5-year-old breach won't tank your score; a 3-month-old breach will. The biggest lever is time, but transparent post-incident write-ups also partially offset the penalty.
How the pillars work together
The 10 pillars aren't independent. Your overall AI Visibility Score is the geometric mean of the CORE composite and the AURA composite — not a simple average. That matters: if either half is weak, the overall drops faster than a simple average would suggest. Both halves have to work.
Common failure patterns we see:
- High Training Access + Low Agent Access = robots.txt is fine but live agents can't render your JS-only SPA. Common in modern React/Vue sites without SSR.
- High CORE + Low AI Brand Recognition = perfect technical site but AI doesn't know your brand. Common in well-built new startups — they haven't existed long enough to be in training data. Time + Wikipedia + press coverage fix it.
- High AURA + Low Citation Depth = AI mentions you a lot but always via third parties. You don't own the narrative. Fix: structured first-party docs + high Agent Access so live grounding can fetch them.
- High everything + High Risk = great visibility but recent controversy is dragging AURA down. Wait for time decay; publish transparent post-mortems.
- High Share of Voice + Low Competitive Rank = AI mentions you but always last. Lift Brand Recognition and Citation Depth to climb the order.
The individual pillar scores are where the actionable insights live. The composite tells you how visible you are; the pillars tell you why.
What this means for agencies
If you're an SEO agency, AI visibility is a new service line waiting to happen. Your clients are already asking (or will be asking) "why isn't ChatGPT recommending us?"
The 10-pillar framework gives you:
- A concrete audit deliverable — the $29 paid audit produces a white-label PDF with all 10 pillars, the per-dim fix plan, and the live ChatGPT/Gemini/Perplexity probe transcripts
- Clear optimization roadmap — each pillar has specific, actionable fixes with score-impact estimates so you can sequence work by ROI
- Measurable progress — every paid audit includes a free automatic 90-day drift re-audit, so you can verify the offsite needle moved without spending another credit
- Pay-per-audit economics — $29 single, $99 for 5 ($19.80/audit), $299 for 20 ($14.95/audit), $999 for 100 ($9.99/audit). No subscription means you can offer this to a single client without committing to a tool budget
- A differentiated service — most agencies still aren't offering AI visibility audits. The ones that start now will compound the advantage as AI search continues to grow
The agencies that start measuring and optimizing AI visibility now will have a significant competitive advantage as AI search continues to grow.
Try it yourself
You can run a free CORE check on any website right now — Training Access + Agent Access scored in 10–30 seconds, no account needed. The full $29 audit unlocks the other 8 pillars (3 deeper CORE dims + 5 AURA dims + Risk modifier) with the live brand-blind probes against ChatGPT, Gemini, and Perplexity.