English Version

Not Cited by AI? Walk This 6-Stage Diagnostic Tree needs a clear distinction between crawler access, search retrieval, model training, and customer-facing answers. This guide uses reachable sources to compare tradeoffs, implementation checks, measurable decisions, operational ownership, and documented rollback conditions before a team changes its robots.txt policy.

Your content is live, but AI still won't cite it — the problem is rarely the writing. Somewhere between discovery and absorption, a single broken link in the chain renders months of work invisible.

Don't Blame the Model First: The Problem May Start Earlier

When teams notice that AI platforms like ChatGPT, Perplexity, or Gemini never cite their content, the instinct is to rewrite copy or add keywords. That instinct addresses the wrong layer. Before a model can cite anything, the content must clear three prior gates: it must be discoverable, crawlable, and retrievable. Fail any one of them, and even precisely written content never enters the model's candidate pool.

The AI citation chain works roughly like this: a crawler finds the page, reads its content, passes it to an index, a retrieval system surfaces it when a question matches, the model selects from candidates, and finally absorbs the specific claims before citing. That's six distinct stages. A failure at stage two looks identical from the outside to a failure at stage five — the brand simply never appears.

Google's own documentation makes the infrastructure dependency explicit:

"Googlebot uses a web rendering service to render JavaScript before it can index the page content, but this process can be delayed or skipped if resources are constrained." — Google Search Central: JavaScript SEO Basics

That delay isn't theoretical. Pages served as client-rendered JavaScript shells may sit unrendered in queues for days, or never get rendered at all for AI crawlers that don't queue. The practical consequence: a visually normal page can be completely unreadable to every machine that matters.

Accepting this reframes the diagnosis. "AI doesn't cite us" is a symptom. The six-stage diagnostic tree below is the method for finding the actual cause.

Problem Explained: Why the AI Citation Chain Breaks

The chain breaks because each of the six stages has independent failure modes, and later-stage failures are routinely misattributed to content quality. Three early-stage failures account for the majority of invisible-brand cases.

Discovery failure happens when crawlers never find the page at all. A missing or malformed sitemap, an internal link structure more than three levels deep, or a page with zero inbound links means crawlers have no path to the URL. If the page can't be found, nothing downstream matters.

Crawl failure is subtler and more damaging precisely because it's harder to spot. The crawler reaches the URL but cannot read the content — either because robots.txt blocks the request, or because the page requires JavaScript execution that the crawler skips. The critical nuance is that robots.txt can block not just pages but the script and stylesheet resources those pages depend on. Block the JavaScript file, and the crawler renders an empty shell.

"If you block Googlebot from crawling JavaScript or CSS files, it might have trouble understanding and indexing your pages." — Google Search Central: robots.txt Introduction

Retrieval failure occurs after indexing. The page made it into the index, but when an AI constructs a response to a specific query, the retrieval system doesn't surface this page as a candidate. The most common cause is low semantic density: no structured markup, no clear entity definitions, no FAQ schema. Retrieval models match query intent to content structure — a page that reads like a brochure but contains no explicit question-answer pairs is easy to skip.

These three failures compound. A page that is partially crawled, sparsely indexed, and structurally thin will fail all three stages simultaneously, producing the same symptom as a page blocked outright.

Method Framework: Walk a 6-Stage Diagnostic Tree

The diagnostic tree's value is in forcing sequential verification. You cannot credibly diagnose a stage-four failure (selection) without first ruling out stages one through three. Each stage has a diagnostic question, observable failure signals, and a concrete check.

Stage 1 — Discovery Failure Diagnostic question: Can a crawler find a path to this page? Failure signals: Page absent from submitted sitemap; not linked from any other page on the site; Google Search Console reports it as "not indexed — not found." Check: Verify the page appears in sitemap.xml and that it's reachable within three internal link hops from the homepage.

Stage 2 — Crawl Failure Diagnostic question: When the crawler reaches the page, can it read the content? Failure signals: Google Search Console URL Inspection shows a blank or near-empty rendered screenshot; robots.txt contains Disallow rules covering /js/, /assets/, /_next/, or /static/; server logs show 403 or 5xx responses to known crawler user-agents. Check: Use the URL Inspection "view crawled page" screenshot. Per JavaScript SEO Basics, critical content should be available in server-rendered HTML, not loaded exclusively by client-side scripts.

Stage 3 — Retrieval Failure Diagnostic question: Does the content surface when AI systems answer questions in this topic area? Failure signals: Competitors are cited consistently; this domain never appears despite covering the same topics; page has no JSON-LD schema, no clear FAQ structure, no explicit entity definitions. Check: Add Article, FAQPage, and Organization JSON-LD. Confirm the page contains at least one explicit question-answer pair matching a likely user query.

Stage 4 — Selection Failure Diagnostic question: The content is in the candidate pool — why isn't it chosen? Failure signals: The page occasionally appears in retrieval but is never cited; competing pages on the same topic carry stronger authority signals (more external links, entity mentions in Wikipedia or Crunchbase, industry publication references). Check: Compare the authority profile of consistently cited competitors. Identify whether the gap is in external links, named entity associations, or the presence of sourced statistics that make the content more citable.

A 2025 study of AI citation behavior (arXiv:2506.11097) examined how large language models select sources when generating answers, finding that selection probability correlates with page structure and authority signals rather than keyword density alone. This is the numeric underpinning for why stage-four fixes look different from stage-two fixes.

Stage 5 — Absorption Failure Diagnostic question: The model selected this page — did it extract the intended claim accurately? Failure signals: AI answers cite the URL but misstate the brand's product, category, or positioning; the brand is placed in the wrong competitive set. Check: Confirm the page's above-the-fold content contains a single, unambiguous definitional sentence ("X is a tool that does Y for Z"). Verify a deployed llms.txt file provides a structured brand description the model can parse without inference.

Stage 6 — Conversion Failure Diagnostic question: Citations are occurring — are they generating accurate, attributable brand mentions? Failure signals: The brand name appears in AI answers without a link or within a generic category list; the description used is a competitor's framing; brand positioning is diluted or inverted. Check: Run structured prompts across ChatGPT, Perplexity, Gemini, and Claude. Evaluate whether the brand description in each answer matches the intended positioning. If not, update the llms.txt brand definition and FAQ entries to close the gap.

The tree enforces discipline: every "AI doesn't cite us" investigation starts at stage one, not at content. The three numeric checkpoints — rendering delay documented in JavaScript SEO Basics, script-blocking documented in robots.txt Introduction, and structure-over-keyword selection documented in arXiv:2506.11097 — each correspond to a distinct stage and require a distinct fix.

BrandGEO Solution: Map the Tree to Audit, Fix, and Recheck

Knowing which stage is failing matters only if there's a clear path from diagnosis to deployed fix. BrandGEO's workflow is structured to match the six stages directly: enter a public URL, run a six-gate audit, receive a fix package, deploy it, and recheck.

The six-gate audit checks the full chain: robots.txt configuration, sitemap status, JavaScript render readability, structured markup completeness, FAQ coverage, and brand description accuracy. Results are presented as itemized scores so a team can see exactly which stage is failing and how severely.

The fix center generates the specific artifacts needed per failing stage:

• llms.txt — a structured plain-text file deployed at the site root that gives AI crawlers a parseable brand description, product definition, and FAQ. This directly addresses stage-five absorption failures.
• robots.txt revision — removes rules that inadvertently block JavaScript or CSS resources, clearing stage-two crawl failures.
• JSON-LD schema snippets — adds Article, FAQPage, and Organization markup to improve stage-three retrieval matching.
• FAQ generation — produces question-answer pairs aligned to likely user queries, improving both retrieval density and absorption accuracy.

Each artifact comes with a deployment verification step. The fix is considered complete only after verification confirms the crawler can now read what was previously blocked, or that the schema is valid and indexed.

BrandKit stores the canonical brand description, product definitions, and common-question answers that feed both llms.txt generation and FAQ content. When a model reads the page during stage-five processing, BrandKit's content determines whether the brand is understood correctly or approximated from surrounding context.

Recheck tracking runs the six-gate audit again after deployment and presents a before/after score comparison alongside changes in citation rate. This replaces subjective evaluation with a measurable signal: did the fix move the score at the relevant stage?

Anonymous CSR Shell Case: Why Months of Content Left AI Blind

A corporate social responsibility team spent several months producing content — sustainability reports, community program descriptions, policy statements — each piece professionally edited and SEO-reviewed. When the team queried multiple AI platforms on topics squarely within their coverage, the company never appeared. Competitors were cited repeatedly. Their own content was invisible.

The initial investigation revealed that every CSR content page was served through a single-page application architecture. Content loaded dynamically via client-side JavaScript; the initial HTML response contained no body text. A crawler hitting any of these URLs received a structurally valid but completely empty page — title tag present, content absent.

This is a compound stage-two and stage-three failure. The content itself was accurate and well-structured. The failure was in the rendering layer. Crawlers received shells, indexed shells, and retrieval systems had nothing to match against.

As JavaScript SEO Basics documents, Googlebot queues JavaScript rendering separately from the initial crawl, introducing delays that can extend days or longer. AI crawlers used by Perplexity, ChatGPT's browsing features, and similar systems typically do not queue — they parse what the initial HTTP response provides and move on.

The fix was not producing more content. It was adding server-side rendering so the initial HTML response contained actual text. After SSR implementation and sitemap resubmission, content began appearing in AI citations within weeks.

The lesson: months of content investment produced zero AI visibility because of a single infrastructure failure at stage two. No keyword optimization, content expansion, or authority-building effort would have changed that outcome. The diagnostic tree catches this at stage two and stops teams from investing in the wrong fix.

Frequently Asked Questions

Q1: My content is already indexed by Google. Why doesn't AI still cite it?

Google indexing confirms you've cleared stages one and two — discovery and crawl. AI citation requires three additional stages: retrieval (does the content surface for relevant queries?), selection (does it win against competing candidates?), and absorption (does the model extract the intended claim accurately?). Being indexed is necessary but not sufficient. Start your continued diagnosis at stage three: check for JSON-LD schema, explicit FAQ structure, and clear entity definitions.

Q2: How do I verify that my robots.txt isn't accidentally blocking crawlers?

Visit yourdomain.com/robots.txt directly. Look for any Disallow rule that covers /js/, /assets/, /_next/, /static/, or similar resource paths. Per the robots.txt Introduction, blocking these paths prevents crawlers from completing page rendering, leaving them with empty content. For confirmation, use Google Search Console's URL Inspection tool and review the "view crawled page" screenshot — if it shows a blank or skeleton layout, a resource is being blocked.

Q3: What is llms.txt and does it actually affect AI citations?

llms.txt is a plain-text file placed at your site root that provides AI crawlers with a structured, parseable description of your brand: what it is, what it does, who it serves, and answers to common questions. It targets stage-five failures — cases where the model reaches your page but misidentifies or misrepresents the brand because the page's primary content is ambiguous or requires too much inference. A well-formed llms.txt reduces the probability of the brand being placed in the wrong category or described in a competitor's terms.

Q4: How often should I run through the full six-stage diagnostic?

Run a complete diagnostic whenever: the site undergoes a technical architecture change (migrating from static to SPA, or the reverse); brand positioning changes significantly; or AI platform citations show a measurable drop or start containing inaccurate descriptions. For routine maintenance, a monthly check of stages one through three — sitemap validity, crawl accessibility, and schema presence — is sufficient. Full-chain diagnostics are event-triggered, not calendar-triggered.

Ready to find out which stage is breaking your AI citation chain?

Enter your public URL and BrandGEO runs a six-gate audit automatically, generates a deployable fix package — llms.txt, robots.txt revisions, JSON-LD snippets, and FAQ entries — and provides a recheck comparison after deployment so you can see exactly where the score moved.

FAQ

Who should use this guide?

It is for teams evaluating Not Cited by AI? Walk This 6-Stage Diagnostic Tree who need clear steps, evidence, and risk boundaries.

What should I confirm before acting?

Confirm the target audience, public evidence, citable site pages, and the structured content that needs attention first.

How do I tell whether the work is effective?

Track brand mentions in AI answers, cited sources, indexed pages, structured-data status, and the content quality-gate results.

Source-backed data points

• Not Cited by AI? Walk This 6-Stage Diagnostic Tree source 1：Manual archive: tasks/2026-07-12-manual-digest-source.md.