Rules by topic

Pure AI content may carry risks of brand penalties and lower citation quality; human review is used to support accuracy and brand voice.

Vendor guidance suggests LLMs may prioritize credible brands and verifiable information. Expert attribution and cited sources are associated with a higher likelihood of being selected as a cited answer. Content relevance signals are also described as depending on demonstrating genuine expertise rather than thin content.

Platforms are rejecting AI slop; AI cannot fabricate named-source expertise, original quotes, or proprietary data — these are the signals that earn third-party credibility as AI content floods the web.

Trust is often described as a constraint on AI adoption. Source links and authority signals are used for verification—63% of AI searchers still want receipts.

AI models may infer authority from named experts. Attribution is associated with trustworthiness when a model considers quoting a page.

Practitioner guidance describes E-E-A-T signals as among the clearest reliability indicators for AI systems choosing sources to cite.

YouGov found these two signals were the trust signals AI searchers cited most often (clear source links 16%, official-source framing 15%), while 49% of non-AI searchers say no trust signal would change their mind.

Opinion and vendor guidance suggest AI systems may establish source credibility before quoting. Author transparency is described as easing that resolution.

SME expertise is a commonly cited differentiator, and author schema identifies a verifiable individual as the author.

AI search synthesizes from diverse off-site sources; if your brand is invisible in forums, review sites, and social platforms, it will be absent from AI-generated answers even if your site is strong.

Visible citation share for Reddit is small and moved sharply in 2026, so do not treat a community post as a citation channel. Coordinated promotional activity on these platforms can be flagged and banned by moderators and detection systems.

Primary sources create net-new information that AI systems cite across channels and earn backlinks from trusted domains; secondary interpretation competes with the original sources it references.

AI search can draw on multiple sources; observational research found off-site mentions, especially third-party mentions, correlated with visibility in AI answers.

AI engines use cross-platform corroboration to establish trust; a network of average content distributed across trusted sources outperforms isolated exceptional content.

AI search cites third-party domains (Reddit, YouTube, LinkedIn, news) in addition to your site; brand mentions off-site are associated with AI visibility.

AI systems may not paraphrase what only you possess; proprietary data is often a distinctive reference point for AI systems and journalists alike. One analysis found that articles based on original data or stats accounted for 50% of AI referral clicks versus 5% from organic search.

Practitioner guidance describes AI systems as synthesizing a brand's identity from many sources, and links coherent, detailed content to greater visibility and contradictions to weaker credibility in AI-generated answers.

AI systems often draw on on-site content, off-site mentions, owned properties, and earned media; consistent brand presence from multiple angles is associated with higher citation likelihood; 84% of citations come from earned, third-party sources.

Observational data shows third-party pages that define categories and reflect consensus are associated with AI mentions more often than owned domains; ~90% of third-party mentions come from listicles, comparisons, and review roundups.

Ahrefs research found AI answers often featured other brands mentioned in the text rather than the author's brand, meaning the listicle may become free marketing for competitors.

Entity optimization and cross-platform consistency never went mainstream in SEO but matter more in AI search because the model's account of you is aggregated from all readable sources.

ChatGPT visible Reddit citation share fell from about 3.83% to under 1% in August 2026. A 2025 retrieval study found Reddit in many search slots and few visible citations; a later pass by the same lab said that gap no longer describes the engine. Reddit flags spam at scale and bans offenders, so do not post artificial brand recommendations.

SurferSEO and 2026 AirOps studies show content addressing only the head term misses the passages actually retrieved; 95% of ChatGPT fan-out queries had zero monthly search volume. One study found pages with strong fan-out coverage were cited in Google AI Overviews 161% more often, and 32.9% of cited pages showed up only in results for fan-out queries.

Louise's correlation study across traditional SEO factors found YouTube mentions correlated at 0.737 with AI visibility; YouTube was #1 cited domain in AI Overviews and AI Mode and a close second to Reddit in ChatGPT/Gemini/Copilot/Perplexity, likely because YouTube transcripts are major training data.

Do not advertise. Expert participation is the practice. Reddit visible share of ChatGPT citations fell under 1% in August 2026, so this is not a dependable citation channel.

AI citations and AI referrals serve different purposes; a page can excel at one while failing the other. Prioritize content types that score high on proprietary data, verified facts, decision support, or action completion.

AI tools tend to cite trusted, consistently mentioned sources, so domain authority established through traditional SEO remains relevant.

The source explains LLMs lean on a small set of sources per category and that mapping cited domains directs outreach 'where citations come from instead of where backlinks used to.'

In an analysis of 3,554 retrieved pages across 57 conversations, cite rates were 6.2% with two domain pages but 1.7% at six or more; the analysis reports that ChatGPT groups results by domain and by claim, not keyword.

Freshness strongly affects AI citations, especially for commercial intents.

Freshness is a documented pattern in AI citation behavior; stale content loses visibility over time.

Practitioner guidance suggests AI engines may favor recency, and one report found prompts with freshness signals were associated with web search 81% of the time.

Cited pages are on average 25.7% fresher than traditional Google rankings, according to observational data.

Observational data links freshness with citation: content from the past two years accounts for 85% of AI Overview citations, and 76.4% of ChatGPT's most-cited pages were updated within the previous 30 days.

HubSpot observed 106% organic growth associated with updating old posts. AirOps data shows stale pages are 3x more likely to lose AI citations. Updating existing pages is associated with measurable results faster than building from scratch.

Practitioners report that AI systems scan for passages they can lift and restate inside generated answers. Content that requires context from surrounding paragraphs may be skipped.

Transformer attention and passage retrieval bias toward lead paragraphs. Sources cited in answers must be identifiable from opening sentences alone.

Longer queries with informational intent are associated with AI Overviews and citations more often.

Observational analyses found that a substantial share of AI citations come from the opening section of a page.

AI systems pull from early passages to generate answers; the first sentences carry highest extraction probability.

Vendor guidance suggests answer engines surface lead paragraphs as standalone citations without needing surrounding context.

AI users ask detailed conversational questions; one-size-fits-all keyword content won't satisfy these nuanced requests.

AI systems favor scannable content; lists improve readability for both AI and users and match AI response formats.

AI systems scan content higher up on the page and prefer sentences semantically close to the query that appear early.

AI systems tend to retrieve content that mirrors the phrasing in user queries, and mismatch between content language and query language is associated with lower extractability even for authoritative pages.

Microsoft states assistants can often lift Q&A pairs word for word into AI-generated responses. Q&A format is native to how AI extracts and presents information.

Practitioner guidance describes AI systems as evaluating whether content answers the broader intent; 37% of AI searchers use assistants to verify information and expect completeness.

AI engines are reported to extract individual passages to answer subqueries, and explicit answers may be associated with higher citability.

Guidance indicates AI engines retrieve discrete answer units, and modular blocks may support precise extraction. FAQ structure mirrors how AI presents answers in results.

AI systems evaluate each chunk independently. Sentences that only make sense in context may be extracted incorrectly or skipped entirely.

Guidance suggests AI systems may skip content that is confusing, vague, or filled with marketing language, while clear, direct content is more likely to be quoted in an AI answer.

Daily AI search users click through at 50% rate; they may find little value in content that merely repeats the AI Overview summary.

Google's AI systems favor content that provides unique expertise beyond what any AI model could generate or summarize.

The Princeton GEO study found content with attributed expert quotes was associated with roughly 41% higher visibility in LLM citations.

ChatGPT cites 44.2% from the top 30% of a page; the bottom 10% of any page earns only 2.4-4.4% of citations across all verticals, and the first 10% is typically navigation and intro fluff AI skips.

The Princeton/Delhi GEO paper found citations, quotations, and statistics increased source visibility by over 40%, outperforming keyword stuffing and unique wording.

HubSpot claimed a 642% rise in AI citations from semantic triples, but overuse makes content feel artificial and NLP can already extract relationships from complex sentences.

The article's AI-ingestion design principles place key information first for extraction.

An analysis of 13,000+ ChatGPT conversations found the average prompt was approximately 350 words.

Each AI platform draws from different citation ecosystems and source types. A single 'AI-optimized' content piece aimed at all three platforms is aimed at none.

Analysis of 815,000 query-page pairs found that moderate coverage (26-50%) outperforms exhaustive coverage in ChatGPT citation rates. The retrieval system may act as the gatekeeper rather than content volume.

LLMs break content into chunks for retrieval; passages extracted without surrounding context must retain meaning on their own.

AI systems tend to cite content that demonstrates clear expertise on focused topics; wide-ranging articles are associated with lower citation odds.

Observational research and practitioner guidance report that answer engines and RAG systems often cite passages from early in the page.

Guidance describes question-shaped headings as helping map user prompts to retrievable sections in conversational search.

AI search favors structured, credible, and interpretable content. Leading with direct answers and using concise lists can make content easier for users and models to read, interpret, and reuse.

Use one H1 and nest H2 and H3 in order so each section can be read on its own. The multiplier often attached to this advice is the same AirOps comparison of cited URLs against Google top results, not a test that adding headings raises citations. A controlled head-to-head found formatting changes did not decide which source was cited.

AI engines retrieve at the passage level, not the page level. Each section must stand alone for AI to extract and cite it.

Guidance suggests AI engines favor content with clear structure matching query intent. Headings that mirror question phrasing are associated with higher extraction probability.

AI Mode processes content at the passage level, pulling modular chunks. Content that can stand alone gets selected for summaries and direct answers.

Practitioner guidance suggests AI engines favor extractable structured content. Dense tabular facts with explicit units can be quoted without interpretation.

AI systems often surface extractable answers to specific questions; answers buried in prose may be harder for AI to surface.

AI Mode processes content at passage level, pulling modular chunks to build responses; modular sections read as complete even without surrounding context.

Query fan-out indicates AI systems draw on multiple sub-queries, so addressing the entire topic cluster may help.

Listicles are the dominant format AI cites for comparison and recommendation queries.

Evertune's analysis of ~25,000 URLs found 63% of LLM citations pointed to listicles; ranked lists accounted for 71-86% of cited listicles.

The article prescribes clear H1/H2/H3 hierarchy with descriptive self-contained sections and front-loaded answers for AI extraction.

AI search is built around synthesized answers; content supporting the full answer path is associated with more citations than content answering only the first query.

Google (and likely other AI systems) crawls pages but excludes those that are duplicative of existing content; commodity content is the primary reason for crawled-but-not-indexed status.

FTC Consumer Review Rule prohibits presenting company-controlled content as independent reviews; penalties reach $53,088 per violation and Google is cracking down.

In 2026 some brands lost 30–50% of their AI visibility from such tactics, and Google's crackdown on self-promotional listicles removes the citations AI engines draw on.

Lily Ray research found sites relying on thin promotional listicles lost 30-50% of traffic after Google stopped ranking that content.

Gary Illyes said at Search Central Live that scaled content is now becoming more of a problem than link spam, and Google expects more spam updates and new spam thresholds.

Official crawler docs separate training vs search use; wrong blocks remove you from retrieval.

ACP emerged in late 2025 with documentation from Stripe and OpenAI and is the core of Instant Checkout on ChatGPT; merchants can apply at chatgpt.com/merchants.

Google's AI optimisation guide points ecommerce and local brands to these feeds as the structured facts its systems answer from, and feed accuracy usually sits outside the SEO team.

The added detail is designed to help AI systems determine when the product fits, how it compares with alternatives and which pre-purchase questions it answers.

Ecommerce answers use retailer and marketplace domains as commercial evidence, with competitor/alternative domains at 36.6% of the citation mix.

Product pages carry the revenue, yet a study of retail sites found about a third of a typical product page could not be read by a model, with site-level scores ranging from 82.5% down to 54.2%.

Many AI crawlers do not execute JavaScript; hidden facts are invisible to retrieval.

Practitioner guidance indicates AI retrieval bots often fetch raw HTML rather than execute JavaScript, and an empty application shell may stop citation at the first request. Googlebot is reported to render JS but queue it, which can add delay and cost.

ChatGPT may pull live results from Bing; IndexNow notifies Bing of content changes, which can make new content discoverable more quickly and improve citation chances.

Guidance and crawler analyses report that most AI crawlers do not execute JavaScript and only see initial HTML; JS-loaded content may not be visible to answer engines.

Bing's webmaster guidelines say prompt URL notifications cut stale or wrong links in Copilot answers and grounding results.

Fabrice Canel of Microsoft explicitly recommended the IndexNow API for notifying engines of new or changed content.

When heading levels skip or a page has several H1s, a model has a harder time telling which sections belong together.

Anthropic never named its index but added Brave Search to its subprocessor list and a BraveSearchParams parameter was found in the API; Profound measured the overlap on a 15-result sample and found an 86.7% overlap.

If an answer engine's first crawl finds thin content or rendering issues, it may take much longer to return, if it returns at all.

Most AI tools crawl raw HTML like search engines and miss content that requires interaction to load.

The speakers state AI bots don't render JS, so single-page-app content is stripped to a blank or incomplete page and never enters the retrieval pipeline.

Logical headings and one-idea paragraphs can make extraction easier for AI systems that parse content by following HTML structure.

AI agents rely on the accessibility tree, and unnamed controls may be unusable to them; one fix in a global header applies across every page.

LLMs are probabilistic generators, not databases; RAG systems are described as needing an organized, navigable, authoritative data source, which SEO practices aim to provide.

The study found AIO on 97.7% of alternative prompts and 92.9% of category prompts, but only 21.5% of brand-name searches.

Practitioners report that AI systems assemble their entity picture from sources across the web, not just your homepage, and argue that inconsistent NAP or product specifications may weaken the confidence models need to cite you.

Guidance suggests LLMs act as consensus engines, and conflicting facts (e.g., different hours on Yelp vs your site) are associated with lower AI confidence and omission to avoid hallucination.

AI evaluates relevance in context; location-specific answers help AI match nuanced user intent with authoritative responses.

Citation gaps define the AEO opportunity score. Position 1 in AI answers gets cited 58% of the time versus 14% for position 10.

Traditional SEO metrics don't reflect AI visibility; different layers require different measurements to identify where optimization is working or failing.

AI search removes the click step by providing answers in the interface. Traffic is no longer a reliable proxy for influence or presence in AI-generated answers.

Only 22% of AI-influenced purchases convert inside AI tools. 86% of consumers verify AI recommendations on Google (68%) or brand sites (48%) before buying. AI influence is real but invisible in standard analytics—it appears as organic or direct traffic.

Citations measure evidence; referrals measure next action. Blending them can obscure which layer is underperforming and is associated with false confidence in 'AI visibility' numbers. One analysis found that up to 67% of AI-referred traffic lands on pages that were never cited.

~40% of AI citations don't name the source brand (ghost citations); Perplexity reaches 52%. Tracking only citations hides where your brand disappears from answers.

Traditional SEO KPIs (CTR, rankings) do not measure AI visibility. Citation data is required to justify AEO investment to leadership.

AI answers shift as models update; without tracking, you are optimizing blind; most brands do not know their AI visibility.

Single prompt tracking misses that 85% of categories have no dominant brand; appearing in one prompt does not mean owning the topic.

Brands named in fan-out queries are cited 68.9% of the time versus 2.1% for pages only retrieved; if your brand never enters the query, you have a visibility gap.

Branded visibility is inflated by default; unbranded prompts reveal where you are truly competitive. Tracking topics disconnected from published content is associated with irrelevant insights.

It is the first dedicated dashboard showing which URLs are cited in AI-generated answers and which queries trigger citations, though metrics reflect citation frequency only — not ranking, prominence, or clicks.

The FAQ gives the exact GA4 configuration steps and source domain list for tracking AI referral traffic.

An aggregate blends different engines into one misleading number, like averaging Google rank with Bing rank; high vs low reasoning alone opens an 18 percentage point citation-rate gap.

Pew found only ~1% click citations in AI Overviews, a ChatGPT leak shows 0.69% CTR, and 70.6% of AI-referred traffic appears as Direct with the referrer stripped; custom channel groupings recover only 50-70%.

Conductor's 2026 benchmark of 13,770 enterprise domains put AI referrals at 1.08%, ranging 0.25% (Communication Services) to 2.80% (Information Technology), and attribution leaks across apps and surfaces; there is no defensible multiplier.

Profound's ~80,000-prompt-per-platform study measured this volatility; changing prompts confounds your work with background movement.

The report (live for everyone as of August 11, 2026) counts impressions only, and the Search Analytics API type parameter still ends at googleNews with no AI-related searchAppearance; the UI export button is the only data exit.

EMARKETER's AIVI methodology uses consistent analyst-led prompts like 'best [product]' or 'top [service] for [specific need]' so every brand is measured against the same standard.

Semrush survey shows AI-tool discovery mentions are the most commonly observed change (43.1%), ahead of direct AI search traffic (40.8%).

Conductor's report finds conversions/leads from AI search is the top 2026 metric and frames AEO as a quality game vs. a numbers game.

Many AI-influenced conversions show up as direct or branded organic visits after someone first meets the brand in an AI answer, so analytics miss them; Ahrefs reported that self-reported attribution put about 3% of conversions on AI, and that AI visitors converted far better than organic ones.

GSC question queries reveal genuine user demand at the intent level; 80%+ of AI Overview queries are informational, so FAQ content may be relevant to AI visibility.

Bot traffic dwarfs human traffic (~98% of requests). Client-side analytics miss all bot traffic. Without IP verification, spoofed bots may inflate false metrics. One comparison found GPTBot can make 470x more requests than Googlebot on fresh domains.

ChatGPT and Claude share only ~8% of citations; brand overlap ranges from 40% (Purchase) to 63% (Comparison); optimizing for the wrong intent type may waste resources.

AI models crawl pages they consider relevant without citing them; this crawl-without-citation gap is an early warning signal of authority deficits.

Qwairy's analysis of 118,000 AI responses across ChatGPT, Perplexity, Google AI Mode, and Claude found minimal overlap in cited domains; only 11% of cited domains appeared across multiple platforms, while 89% were platform-specific.

Fan-out queries seek new information while grounding queries verify and fact-check existing information, but practically there is good overlap; Bing also shows citations per URL.

It is the only owned page type that never drops below double-digit citation share on either platform; an observational study found it cited at 19.7–26.6% on ChatGPT and 14.8–17.1% on AI Mode.

Request frequency is the only signal of whether your content is being regularly considered in these environments.

The AI Visibility Index methodology uses standardized prompts like "best body lotion" or "shampoo for damaged hair" varied across those dimensions to reflect real-world consumer questions, developed from primary shopper research.

The AIVI framework explicitly lists monitoring competitive share of voice and identifying emerging category trends as its core uses.

The article states the questions you track decide your number more than anything else, and a citation rate with no prompt set attached doesn't mean anything.

SparkToro's study collected 2,961 responses across 12 prompts and three AI tools and found substantial variation in recommendation lists; Envoyra independently found the same list of brands repeated fewer than 1 time in 100.

Structured identity and offers are intended to help machines reconcile brand and product facts.

Keep heading levels in order (H1, then H2, then H3) and add schema that matches the visible page. The heading multiplier and the schema percentage quoted for this rule come from one AirOps comparison of pages ChatGPT cited against Google top results. That comparison was not replicated, and a controlled test of adding JSON-LD found no meaningful citation lift.

Structured data is often described as a bridge between content and how AI interprets it. Schema markup following Google's guidelines is intended to help AI understand page content and create multiple entry points in search.

Structured formats are easier for AI systems to lift and reproduce accurately than flowing prose. They signal clear organization and reduce extraction errors.

AI engines extract facts from tables more reliably than prose. Structured markup reduces hallucination risk and enables rich results.

Schema declares entities and relationships in the language AI systems already parse; it is infrastructure for machine understanding, not a citation hack.

Schema makes it easier for LLMs to parse content purpose and pull structured data into responses.

Practitioners describe structured data as the format AI systems tend to favor when deciding what to cite.

Vendor guidance suggests AI systems use entity recognition to verify and surface brands. Organization schema with sameAs can signal cross-platform credibility and help AI systems build a consistent understanding of your brand.

FAQ sections map directly to how users ask AI questions and are naturally chunkable for LLM retrieval.

AI agents can use structured data rather than parsing HTML, and inconsistencies between site data and feeds may hinder agent-mediated transactions.

JSON-LD separates concerns, supports @id graph references, and Google's documentation states 'we recommend using JSON-LD for structured data' with tools optimized to parse it.

Some practitioners report schema helps AI systems map entities; others see no evidence that JSON-LD reaches LLM answer engines directly.

AI crawlers that don't execute JS see only the shell of SPAs; even crawlers that can render face rate limits, timeouts, and headless-browser differences producing incomplete renders.

Practitioner guidance suggests explicit labeling plus FAQ schema is associated with AI systems more easily parsing and repurposing content as answer snippets.

The source links accurate schema with pages that AI and Google feature more readily in Overviews, rich results and voice answers, and recommends validating the markup this way.

ChatGPT's mentions and the sources its search layer reads may not always align.

Google states AEO/GEO are still SEO; vendor chunking tactics conflict with that guidance.

AI visibility is a parallel surface, not a replacement. Traditional rankings and AEO citations must both be optimized.

Critchlow argues that, in SEO and GEO alike, results depend mostly on brand, product, PR and editorial work rather than on the SEO team.

Practitioner guidance describes connected content and internal linking via topic clusters as supporting authority in ways standalone pages do not, and describes LLMs as drawing on query fan-out across angles.

Lily Ray states that drops in organic visibility tend to coincide with drops in AI search visibility, as language models may retrieve less content.

Marie Haynes observed ChatGPT referrals plateaued and declined since August 2025 while millions use AI search inside Google daily.

Multimodal search blends text, images, voice, and video; AI systems cannot synthesize what they cannot segment, and transcripts make audio/video readable and searchable.

OtterlyAI's YouTube Citation Study 2026 shows popularity doesn't predict citation value; structure does — long-form accounted for 94% of AI citations and 40.83% of cited videos had fewer than 1,000 views.

52% of content marketers are increasing video investment, and video can earn AI citations when it is readable for LLMs and AI bots.