Google AI Overviews & AI Mode: optimization patterns and citation engineering

A comprehensive installation and audit reference for winning citation on Google AI Overviews and Google AI Mode. The two surfaces share an engine family (Gemini 3 Pro since January 2026) and 13.7…

Google AI Overviews and AI Mode: Citation Architecture for the Decoupled Surface

A comprehensive installation and audit reference for winning citation on Google AI Overviews and Google AI Mode. The two surfaces share an engine family (Gemini 3 Pro since January 2026) and 13.7 percent citation overlap, but operate as distinct optimization targets. Citation decoupling means classic ranking position no longer determines visibility on these surfaces. This document specifies the structural, schema, entity, and freshness signals that drive AI Overview and AI Mode citation, the page pattern that earns eligibility, the audit rubric that measures it, and the maintenance cadence that defends against regeneration volatility. Dual purpose: installation manual and audit document.

Cross stack implementation note: the code samples in this framework are written in plain HTML for clarity. For React, Vue, Svelte, Next.js, Nuxt, SvelteKit, Astro, Hugo, 11ty, Remix, WordPress, Shopify, and Webflow equivalents of every pattern below, see framework-cross-stack-implementation.md. For pure client rendered SPAs (no SSR/SSG) see framework-react.md.

1. Document Purpose and How to Use This Document

1.1 What This Document Is

This is the canonical operational reference for citation on Google AI Overviews (AIO) and Google AI Mode. AI Overviews appear on approximately 48 percent of all Google searches in Q1 2026 and over 70 percent of informational and how to queries. Position 1 organic CTR drops up to 61 percent on queries that show AI Overviews. Citation on these surfaces drives roughly 35 percent more clicks than non cited top 10 results, and cited visitors convert at approximately 23 times the rate of standard search visitors.

Citation has decoupled from organic ranking. A Surfer SEO December 2025 study of 173,902 URLs found 68 percent of AI Overview citations come from outside the top 10 organic. An Ahrefs February 2026 study of 863,000 keywords found only 38 percent of AI Overview cited pages also rank in the top 10, down from 76 percent in mid 2025. The signal that determines AI Overview selection is structural readability and entity authority, not ranking position.

1.2 Three Operating Modes

Mode A, Install Mode. Build AI Overview ready infrastructure on a new or existing site. Follow Sections 2 through 14 in order.

Mode B, Audit Mode. Evaluate an existing site for AIO citation eligibility and current performance. Skip to Section 13.

Mode C, Hybrid Mode. Audit first, then install for failing items.

1.3 How Claude Code CLI Should Consume This Document

1. Read Section 2 and collect client variables.
2. Run Section 13 audit on representative pages to baseline current state.
3. Apply Section 6 page pattern to priority pages.
4. Install Section 7 schema stack across the site.
5. Establish Section 10 tracking before measuring lift.
6. Apply Section 9 citation defense cadence.
7. Generate the Section 14 report.

1.4 Conflict Resolution Rules

1.5 Required Tools

• Google Search Console (AI Overview impressions in standard web search since 2024)
• curl with custom user agent for substrate verification
• Google Rich Results Test and Schema.org Validator for schema validation
• An AI Overview tracking tool: one of Profound, Otterly, Athena HQ, BrightEdge AI Catalyst, or Semrush AI Toolkit (tool comparison in framework-aicitations.md)
• Manual sampling discipline: weekly for priority queries, monthly for long tail

1.6 Relationship to Neighboring Frameworks

This framework covers Google AI Overviews and Google AI Mode specifically. The broader AI citation surface across ChatGPT, Claude, Perplexity, Bing Copilot, and Meta AI lives in framework-aicitations.md. The multi engine surface map and the four pillars architecture (SEO, AEO, AIO, GEO) live in SEO-Search-Appearance.md Section 14. The feature level CTR economics for AI Overview as one of 30 SERP features live in SERP-Optimization.md Section 9.1. This document is what those references point to when AIO specific guidance is required.

2. Client Variables Intake

# AI OVERVIEWS FRAMEWORK CLIENT VARIABLES

# --- Business and Site Identity (REQUIRED) ---
business_name: ""
primary_domain: ""
business_industry: ""
ymyl_classification: ""              # "full_ymyl", "partial_ymyl", "lite_ymyl", "non_ymyl"

# --- AI Overview Presence Baseline (REQUIRED) ---
queries_currently_cited_in_aio: 0
queries_currently_cited_in_ai_mode: 0
priority_queries: []                 # 10 to 25 queries the site targets for AIO citation

# --- Content First Baseline (REQUIRED, see framework-contentfirst.md) ---
contentfirst_score: 0                # Out of 30 from content first audit
substrate_in_first_byte: false
schema_in_first_byte: false
js_required_for_primary_content: false

# --- Schema Coverage (REQUIRED, see framework-schema.md) ---
core_graph_present: false            # Organization, WebSite, WebPage, Person
article_schema_present: false
faqpage_schema_present: false        # Where FAQ content exists
howto_schema_present: false          # Where procedural content exists

# --- E-E-A-T and Author Signals (REQUIRED, see framework-eeat.md) ---
author_bylines_on_every_content_page: false
author_pages_with_credentials: false
reviewer_credit_for_ymyl: false      # Required if ymyl_classification above "non_ymyl"
eeat_self_assessment_score: 0        # Out of 130

# --- Information Gain Capacity (REQUIRED, see framework-infogain.md) ---
percentage_pages_with_information_gain: 0
information_gain_categories_active: []

# --- AI Crawler Access (REQUIRED) ---
gptbot_allowed_in_robots_txt: false
claudebot_allowed_in_robots_txt: false
google_extended_allowed_in_robots_txt: false
llms_txt_present: false

# --- Freshness Posture (REQUIRED) ---
content_refresh_cadence: ""
date_modified_visible_and_in_schema: false

# --- Tracking (REQUIRED) ---
gsc_property_verified: false
gsc_ai_overview_impression_baseline_28d: 0
third_party_aio_tracking_tool: ""
manual_sampling_cadence: ""

Citation defense work cannot start until contentfirst_score is at least 22, eeat_self_assessment_score is at least 90, and the core graph schema is server rendered. Sites failing those dependencies route back to those frameworks first.

3. What AI Overviews Are

3.1 Definition

An AI Overview is a generative summary at the top of a Google search results page, synthesizing an answer from multiple cited sources. AI Overviews run on Gemini 3 Pro since January 2026. They cite sources approximately 84.9 percent of the time and mention brands by name approximately 61 percent of the time.

AI Mode is Google's standalone conversational interface accessible from the homepage and from the AI Overview "Ask follow up" affordance. Also Gemini 3 Pro powered. Serves 75 million daily active users processing over 1 billion queries per month as of January 2026. Cites sources 76.3 percent of the time, mentions brands 37.6 percent.

3.2 AI Overview vs AI Mode

The 13.7 percent overlap means optimization for AI Overview does not automatically deliver AI Mode citation. Both surfaces are covered in this framework. Sections 5, 8, and 11 call out differentiation where it matters.

3.3 The 2026 State

Three numbers define the surface:

1. AI Overviews on approximately 48 percent of all Google searches, over 70 percent on informational and how to queries.
2. Position 1 organic CTR drops up to 61 percent on AIO queries.
3. Cited visitors convert at approximately 23 times the rate of standard search visitors. AIO citation drives approximately 35 percent more clicks than non cited top 10 results.

Classic position 1 is increasingly position 5 below the AI Overview, PAA, Local Pack, and Top Stories carousel. A citation in AIO is the single most valuable placement available on Google in 2026.

3.4 Citation Decoupling from Organic Ranking

Two independent studies establish the pattern:

• Surfer SEO December 2025, 173,902 URLs: 68 percent of AIO citations are from outside the top 10 organic.
• Ahrefs February 2026, 863,000 keywords: 38 percent of AIO cited pages also rank in top 10, down from 76 percent in mid 2025.

A page at position 47 in classic organic can be cited in AIO. A page at position 3 may not be cited if the substrate fails to deliver structured extractable content in the first byte. This is why AIO optimization is a separate framework rather than a layer on top of classic SEO.

3.5 Four Pillars Context

AIO optimization is the central work of the AIO pillar in the four pillars visibility architecture: SEO (classic ranking, ten blue links), AEO (Answer Engine Optimization, featured snippets, voice), AIO (AI Overview Optimization, this framework), GEO (Generative Engine Optimization, broader AI citation, see framework-aicitations.md). The four pillars are independent surfaces. A page can win SEO and fail AIO, or win AIO and fail GEO. Each requires its own optimization pass.

4. The AI Overview Citation Mechanics

4.1 Query Fan Out

A user query does not retrieve a single result set on these surfaces. The Gemini engine generates 8 to 12 sub queries (AI Overview) or 9 to 16 sub queries (AI Mode) that decompose the original query into specific information needs, then retrieves candidates per sub query.

A candidate page does not need to match the head query exactly. It needs to match one of the sub queries. Pages with comprehensive sub topic coverage have more surface area for citation than pages answering only the literal head query.

Sub query inspection is possible in AI Mode, which exposes the sub queries panel for some prompts. Use that to reverse engineer fan out for priority head queries and write content addressing the revealed sub queries directly.

4.2 Retrieval Weighting

Each sub query retrieves candidates from the Google index. Retrieval weights:

1. Structural extractability (first byte HTML, semantic sectioning, schema completeness)
2. Entity match between query entities and page declared entities
3. Information Gain score
4. E-E-A-T pillar signals (author authority, reviewer credit, citation infrastructure)
5. Freshness for time sensitive queries
6. Brand mention frequency across the open web

Classic ranking signals (backlinks, CTR, dwell time, domain authority) influence retrieval but do not gate it. A low authority page with strong structural and entity signals can outrank a high authority page with weak structure for AIO citation.

4.3 Synthesis and Citation Selection

The engine synthesizes a single answer from the top candidates per sub query. Citation picks the sources that contributed most. Typical AI Overview cites 3 to 8 sources. AI Mode cites 5 to 12. Documents contributing uniquely citable claims (specific dates, prices, statistics, quoted experts, named processes) earn citation at higher rates than documents paraphrasing common knowledge.

4.4 Regeneration Volatility

AI Overview content is not stable:

• 70 percent of AI Overview content changes when the same query is run twice.
• 45.5 percent of cited sources are replaced on regeneration.
• AI Mode self overlap on the same query run three times: 9.2 percent.

Volatility is the operating state of the surface. Citation defense (Section 9) treats it as a constant, not an anomaly. The measurement frame is probability of citation across a rolling four week window, not binary win or lose on a given day.

4.5 Reading Mode Prevalence

Reading mode is the plain HTML parse mode AI engine bots use. ChatGPT bot begins crawls in reading mode approximately 46 percent of the time, with a 63 percent bounce rate when the page requires JavaScript for primary content. AI Overview parsing mirrors this. The first byte HTML is what the engine sees.

Three implications: server rendered substrate is non negotiable, heading hierarchy carries citation weight (reading mode parses by structure not rendered layout), and JavaScript injected schema is invisible. The deeper tactical treatment of reading mode optimization across all AI engines lives in framework-aicitations.md. For AIO, pass the curl -A "GPTBot" substrate test (see framework-contentfirst.md Section 9.1) and reading mode is largely handled.

4.6 Why Position Based Optimization No Longer Works

Pre 2025 SEO assumed improving classic ranking from 5 to 3 was the leading visibility indicator. Post citation decoupling, that fails for 48 percent of queries. The inputs that drive AIO citation are structural and entity based, not rank based. Sites optimizing for AIO eligibility signals first now have two years of compounding citation accrual over competitors still chasing position 1.

5. Ranking Signals for AI Overview Citation

Ordered by leverage (high to low). Each tagged as Shared with classic SEO or AIO specific.

5.1 Information Gain (HIGH leverage, AIO specific)

Google patent US 11,995,114 B2 establishes the principle: documents adding novel information beyond what is widely available rank higher. AIO citation weights this heavily because the synthesis engine prefers sources contributing specific facts, original data, or first hand observations over sources rewording common knowledge. See framework-infogain.md for the ten categories.

Triage check: if removing your page would leave no specific knowledge unavailable elsewhere, it fails the Information Gain bar.

5.2 Server Rendered Substrate (HIGH leverage, AIO specific)

The first byte server response is the entire surface available for AIO parsing. Reading mode bots do not execute JavaScript. Client side hydrated pages are invisible. See framework-contentfirst.md for the substrate doctrine.

Triage: curl -A "GPTBot" https://example.com/page must return H1, lede, H2s, FAQ content, and schema JSON-LD. Missing any one means the page fails the substrate signal.

5.3 Schema Completeness (HIGH leverage, Shared)

Pages with valid, server rendered JSON-LD using the @graph pattern with @id cross references and sameAs links earn approximately 27 percent uplift in AIO visibility per Google Search Central 2024 internal data. Completeness matters more than count. A single Organization plus WebPage plus Article graph with all properties populated outperforms three sparse blocks. See framework-schema.md. AIO specific schema stack in Section 7 below.

5.4 Entity Salience (HIGH leverage, Shared)

AIO synthesis relies on entity recognition to match query entities to candidate pages. Pages that explicitly declare entities (business name, owner, location, services, named partners, geographic scope) in visible text and sameAs networks earn higher citation probability than pages that imply entities. See framework-entitysalience.md.

Triage check: a reader new to the site should be able to identify business name, owner, location, and topic in the first 200 words.

5.5 E-E-A-T Pillar Weighting (HIGH leverage, AIO specific, AI Mode emphasized)

AIO weights E-E-A-T more heavily than classic ranking because synthesis is editorial selection, not relevance ranking. Author bylines with verifiable credentials, reviewer credit for YMYL, and reputation infrastructure (Wikipedia, Wikidata, industry directories) shift citation probability significantly. AI Mode weighs this even higher because Personal Intelligence integration since January 22, 2026 means user trust signals influence response shape. See framework-eeat.md. Section 8 below covers AIO specific weighting.

5.6 Reading Mode Friendly HTML (MEDIUM leverage, AIO specific)

Heading hierarchy (one H1, logical H2 to H6), descriptive anchor text, semantic sectioning elements (<article>, <section>, <main>, <nav>, <aside>), and lists where appropriate signal extractability. kime.ai 2026 found 4.2x citation rate for HTML <table> over equivalent prose, 2.7x for <ol> for sequential content, 1.8x for <ul> over prose.

5.7 Freshness Signals (MEDIUM leverage, Shared, query type dependent)

For time sensitive queries (news, product release dates, current rates), AIO favors pages with recent dateModified and visible last updated stamps. Use <time datetime="2026-MM-DD"> for any date relevant to content. Schema includes datePublished and dateModified. For evergreen queries, freshness matters less but should be honest (no date manipulation, see framework-hcs.md Section 9.6).

5.8 Internal Link Density (MEDIUM leverage, Shared)

Princeton GEO study (SIGKDD 2024) found pages with at least three inbound internal links from topically related pages earn higher AIO citation rates than orphan or single linked pages. Internal linking signals topical centrality to the synthesis engine. See framework-internallinking.md.

5.9 Brand Mention Frequency (MEDIUM leverage, AIO specific)

AIO synthesis weights unlinked brand mentions on third party sites as a trust signal independent of backlinks. A business earning 50 unlinked mentions across industry publications and review platforms in a quarter is more likely to be cited than a business with 5 mentions, regardless of backlink count. See framework-digitalpr.md and framework-trustsignals.md.

5.10 YMYL Strictness Threshold (HIGH leverage for YMYL pages, AIO specific)

YMYL content faces a higher bar. Pages without credentialed authors or reviewers are systematically excluded from AIO citation on YMYL queries regardless of other signals. September 2025 SQRG update expanded YMYL to include elections and civic institutions. See framework-ymyl.md.

Signal triage summary

For a new client engagement, the high leverage signals are 5.1 Information Gain, 5.2 Server Rendered Substrate, 5.3 Schema Completeness, 5.4 Entity Salience, 5.5 E-E-A-T, and 5.10 YMYL Strictness (where applicable). These six move citation probability most. The medium leverage signals (5.6 Reading Mode HTML, 5.7 Freshness, 5.8 Internal Link Density, 5.9 Brand Mentions) compound the gains and stabilize against regeneration volatility.

6. The AI Overview Eligibility Page Pattern

Every priority page targeting AIO citation matches this pattern.

6.1 Lede Answer Block

Immediately under the H1, a 40 to 75 word direct answer to the page's target query. Citeable standalone, meaning a reader (or synthesis engine) extracting only the lede sees a complete, accurate, specific statement.

<header>
  <h1>How to Pay Quarterly Estimated Taxes in 2026</h1>
  <p class="lede">
    Federal quarterly estimated taxes for tax year 2026 are due April 15,
    June 15, September 15, and January 15, 2027. Pay through IRS Direct Pay
    (free), EFTPS (free, requires enrollment), or debit and credit card
    via IRS approved processors (fee applies). The safe harbor amount is
    100 percent of last year's total tax, or 110 percent if prior year
    AGI exceeded 150,000 dollars.
  </p>
</header>

That lede is 65 words. Four specific dates, three payment methods, the safe harbor rule. Citeable in isolation.

6.2 Dual Extraction Layer

Provide both prose answers and structured equivalents (lists, tables) on the same page. Prose wins AIO citations because the synthesis engine prefers natural language for answer body. Structured equivalents win featured snippets and rich results. Both coexist.

Example: a prose paragraph stating the 2026 quarterly deadlines, followed by an HTML <table> with quarter, period covered, and due date columns. The prose is the AIO surface. The table is the featured snippet surface.

6.3 Answer First Pattern Under Every H2

Every H2 opens with a 40 to 75 word direct answer paragraph addressing the H2 question. Supporting detail follows. This gives the synthesis engine an extractable answer for any sub query that matches an H2.

Anti pattern: opening an H2 with background context, history, or transition. The engine reads the first paragraph as the answer. Burying the answer in paragraph three reduces citation probability.

6.4 FAQPage Schema Mirroring Visible Q and A

Where the page includes FAQ content, visible <details> and <summary> pairs are mirrored exactly in FAQPage JSON-LD. Question name matches the summary text. Answer text matches the visible answer. Frase.io 2025 found pages with FAQPage schema matching visible content earn approximately 3.2 times the citation rate of pages with prose only FAQ.

6.5 Comparison Tables for Choice Queries

For comparison intent queries (X vs Y, best X for Y), use HTML <table> with <thead>, <tbody>, <th scope="col">. kime.ai 2026 documented 4.2x citation rate for HTML tables over equivalent prose for comparison data.

6.6 Numbered Processes for How To Queries

For procedural queries, use <ol><li> rather than prose or <ul>. 2.7x citation rate for ordered lists over prose for sequential content. HowTo schema in Section 7.4 layers on top.

6.7 Author Byline and Reviewer Credit for YMYL

Every content page carries a visible author byline at the top of the article, linked to an author page with credentials. For YMYL topics, a reviewer credit appears alongside identifying a credentialed expert who verified the content.

<header>
  <h1>Quarterly Estimated Taxes 2026: LLC and S-Corp Guide</h1>
  <div class="article-byline">
    <p>By <a href="/authors/amanda-emerdinger/" rel="author">Amanda Emerdinger</a>, Enrolled Agent</p>
    <p>Reviewed by Dr. Jane Smith, MD (for medical YMYL pages)</p>
    <time datetime="2026-04-15">Published April 15, 2026</time>
    <time datetime="2026-05-14">Updated May 14, 2026</time>
  </div>
</header>

6.8 dateModified Visible and in Schema

Both visible dateModified and schema dateModified must be present. Visible because users see it. Schema because the synthesis engine reads it for freshness calibration. Both reflect actual substantive updates, not date manipulation. See framework-hcs.md Section 9.6.

6.9 The Substrate Validation curl Test

Before considering a page AIO ready:

# H1 in first byte
curl -A "GPTBot" -s https://example.com/page | grep -o "<h1>[^<]*</h1>"

# Lede paragraph present
curl -A "ClaudeBot" -s https://example.com/page | grep -o '<p class="lede">[^<]\{100,\}'

# FAQPage schema server rendered
curl -A "Google-Extended" -s https://example.com/page | grep -c "FAQPage"

# Article schema in head
curl -A "GPTBot" -s https://example.com/page | grep -c "application/ld+json"

All four checks must return non zero. Failure on any one means primary content or schema is not in first byte and the page is not AIO eligible regardless of other signals. The full skeleton implementing all of 6.1 through 6.9 follows the structure in framework-contentfirst.md Section 6.2, with the additions specified in this framework's Section 7.

7. Schema and Structured Data Stack for AI Overviews

7.1 The Core Graph

Every page carries the core graph in document head, server rendered, in first byte: Organization with @id, name, url, logo, contactPoint, sameAs to LinkedIn, Wikidata, Wikipedia. WebSite with @id, url, publisher, potentialAction SearchAction. WebPage with @id, url, isPartOf, datePublished, dateModified. Person for the page author with @id, name, jobTitle, hasCredential, worksFor, sameAs to ORCID and professional registries. The @id cross references tie entities into a graph the synthesis engine can traverse. Full graph specification in framework-schema.md.

7.2 Article with Author and Reviewer

{
  "@type": "Article",
  "headline": "Title matching visible H1",
  "author": { "@id": "https://example.com/authors/jane-doe/#person" },
  "reviewedBy": { "@id": "https://example.com/authors/dr-smith/#person" },
  "publisher": { "@id": "https://example.com/#organization" },
  "datePublished": "2026-04-15",
  "dateModified": "2026-05-14",
  "mainEntityOfPage": "https://example.com/article-url/"
}

reviewedBy is critical for YMYL Article schema. The reviewer Person must exist as a separate Person node with credentials, license number where applicable, and license verification URL.

7.3 FAQPage Mirroring Visible Q and A

FAQPage Question name matches visible question text exactly. Answer text matches visible answer text exactly. Mismatches result in Google removing rich result eligibility and reduce AIO citation probability.

{
  "@type": "FAQPage",
  "mainEntity": [{
    "@type": "Question",
    "name": "When are quarterly estimated taxes due in 2026?",
    "acceptedAnswer": {
      "@type": "Answer",
      "text": "April 15, June 15, September 15, 2026, and January 15, 2027."
    }
  }]
}

7.4 HowTo for Procedural Content

{
  "@type": "HowTo",
  "name": "How to Pay Quarterly Estimated Taxes",
  "totalTime": "PT10M",
  "step": [{
    "@type": "HowToStep",
    "position": 1,
    "name": "Calculate the amount due",
    "text": "Estimate annual federal tax liability and divide by four."
  }]
}

Step name and text match visible H3 or list item text. position numbers match step order.

7.5 SpeakableSpecification, ProfilePage, and Companions

For voice extraction overlap, add SpeakableSpecification to Article schema pointing to the lede via cssSelector. For author bio pages at /authors/[slug]/, use ProfilePage schema with the author as mainEntity, populating jobTitle, hasCredential, worksFor, and sameAs. Full ProfilePage pattern in framework-eeat.md Section 4.2.

7.6 Schema Lives in Head, Server Rendered, in First Byte

Hard rule from framework-contentfirst.md Rule 4. Schema injected by GTM, by client side React, or by any JavaScript path is invisible to AIO parsing. Validation: Google Rich Results Test (zero errors), Schema.org Validator (syntactic correctness), curl -A "GPTBot" -s ... | grep -c "application/ld+json" (server rendered in response body), and @id reference resolution (every @id reference points to a defined node).

7.7 Common Schema Rejection Causes Specific to AIO

• FAQPage Question name not matching visible summary text
• HowTo step text not matching visible list items
• Article reviewedBy pointing to a Person without credentials or license verification (YMYL failure)
• @id references that do not resolve to defined nodes
• datePublished in the future
• dateModified earlier than datePublished
• Author Person without jobTitle or hasCredential (E-E-A-T failure for YMYL)
• Schema injected via JavaScript

8. E-E-A-T Weighting in AI Overview Selection

8.1 Why E-E-A-T Weights Heavier Here

Classic ranking uses many signals to determine relevance. AIO citation uses fewer signals to determine editorial credibility because the synthesis engine is making editorial selection, not ranking. Credibility signals weight disproportionately. A page from a credentialed expert with strong reputation network can earn citation on a query where a higher ranking page from an anonymous site does not.

This is a structural shift in optimization economics. Pre 2024, sites could earn classic ranking through link building and content volume alone. Post 2024, AIO citation requires verifiable expertise and reputation infrastructure that link building alone does not produce.

8.2 Author Authority Signals

Every content page carries an author byline at the top. The author Person node in the schema graph carries name, jobTitle relevant to the topic, hasCredential listing degrees and licenses, worksFor referencing the publishing Organization, and sameAs links to ORCID, LinkedIn, professional licensing boards, industry associations. The author has a dedicated /authors/[slug]/ page with full ProfilePage schema. Reading mode bots do not execute JavaScript so author byline and credentials must be in server rendered HTML.

8.3 Reviewer Credit for YMYL Topics

YMYL content (health, finance, legal, civic, safety, major life decisions) requires a separate credentialed reviewer in addition to the author. The reviewer is identified by name and credentials in a visible "Reviewed by" line near the byline, and as a separate Person node referenced from Article reviewedBy. YMYL pages without reviewer credit are systematically excluded from AIO citation regardless of other signals. See framework-ymyl.md for the credentialing matrix per category.

8.4 Citation Infrastructure

Inline outbound citations to authoritative sources boost AIO citation probability by approximately 30 percent per Princeton GEO study (SIGKDD 2024). The pattern: descriptive anchor text linking to .gov, .edu, peer reviewed journals, standards bodies, primary sources. The link itself is the citation. The destination is authoritative.

8.5 Reputation Research Surface

AIO synthesis weights reputation signals: brand search results, review platform ratings, earned media, professional licensing verification. See framework-sqrg.md Section 8 for the reputation research procedures Google's quality raters apply (the same signals the synthesis engine reads).

Minimum reputation infrastructure for AIO eligibility: Google Business Profile claimed and maintained, active presence on at least three relevant review platforms, press kit at /press/, comprehensive About page describing business, owner, and editorial process.

8.6 Brand Entity in the Knowledge Graph

For sites with sufficient notability, presence in the Google Knowledge Graph via Wikidata and Wikipedia is the strongest E-E-A-T signal available. See framework-knowledgegraph.md. Wikidata has a lower notability bar than Wikipedia and is the practical entry point for small businesses and solo practitioners. The Wikidata QID, once established, is referenced from Organization sameAs, completing the entity reconciliation chain AIO synthesis uses.

9. Citation Defense Strategies

9.1 The 70 Percent Regeneration Volatility Problem

AI Overview content changes approximately 70 percent of the time on identical query re runs. When the overview regenerates, approximately 45.5 percent of cited sources are replaced. AI Mode self overlap on three runs: 9.2 percent.

This is not a bug. It is the operating state of the surface. The synthesis engine samples from a candidate pool probabilistically. A page in the pool is not guaranteed citation on any regeneration even if it was cited on the previous one.

Defense posture: stop measuring citation as binary on a single check. Measure as probability across multiple samplings over time. A page cited on 60 percent of weekly samplings of the same query is winning even if it appears uncited on a given Tuesday.

9.2 Why Pages Get Dropped and Re Added

Sampling is influenced by Information Gain freshness (pages with new information in the last month are more likely to be selected on regeneration), schema integrity (edited pages are reevaluated), brand mention pace, competitor pages entering the candidate pool, and periodic query reinterpretation. Static unmaintained pages drift out over months. Pages refreshed substantively with ongoing brand mentions stay in.

9.3 Freshness Cadence

Refresh must be substantive, not date manipulation. dateModified update is legitimate when content actually changed. Update changelogs documenting what changed help signal substantiveness.

9.4 Information Gain Refresh

Refresh that adds Information Gain (new sub topics, recent data, edge cases, contrarian findings) is the highest leverage refresh. Date only refreshes without content substantiation are detectable and counter productive. Patterns that compound: adding a new H2 covering an edge case, adding specific data where prose was qualitative, adding first hand experience documentation, adding a comparison table where the original was prose.

9.5 Maintaining Schema and Entity Declarations

Content edits frequently break schema. Common patterns: FAQ question reworded in visible content but FAQPage schema name not updated, new H2 added but FAQPage schema not extended, author change with byline update but no schema author update, ordered list item added but HowToStep position numbers not renumbered.

After every content edit, re run Section 7.6 validation. Set up an automated check on every content publish to catch drift.

9.6 Cross Site Brand Mentions

Mention frequency stabilizes citation against volatility because mentions signal entity authority independent of any single page's signals. Pace targets for a small business or solo practitioner: 5 to 10 new unlinked brand mentions per quarter from credible third party sources, 1 to 3 earned media placements per quarter, ongoing review accumulation on GBP and at least two other platforms. See framework-digitalpr.md and framework-linkbuilding.md.

9.7 When NOT to Over Rotate

The most common failure mode. A page cited last week and uncited this week is normal volatility. Do not rewrite the page based on a single drop, add hedging that weakens Information Gain, move the page, or strip schema thinking it triggered a penalty.

Do wait two weeks and re sample multiple times, check if competitor pages entered the candidate pool, and if drop is sustained over 28 days then audit Section 13 criteria for specific failures. Stability across rolling four week windows is the right measurement frame, not snapshot citation on a given day.

10. Tracking and Measurement

10.1 GSC AI Overview Impressions

Google Search Console includes AIO impressions in the standard web search type since 2024. They are not separated. To infer AIO presence, compare impression count for a query against average position. A query with high impressions and position below the top 30 is likely earning impressions through AIO citation rather than classic ranking.

10.2 GSC Performance Workflow

For each priority query: open GSC Performance, filter by query, note impressions, clicks, CTR, average position. Manually run the query in Google search, observe whether AIO appears, check whether the page URL is cited. Document state in a tracking sheet. Monthly sweep of all priority queries produces a baseline citation map. Track delta over time.

10.3 Manual Sampling Cadence

Weekly for top 10 priority queries. Monthly for next 50. Quarterly for long tail. Protocol: incognito browser, target market location set, run query, screenshot AIO if present, note citations, run same query in AI Mode (different tab), note citations, note any sub queries exposed. Manual sampling catches what GSC does not surface directly: which sources are cited, what sub queries are running, which competitor pages are entering the pool.

10.4 Third Party Tools

Profound (AI brand visibility), Otterly (citation tracking across engines), Athena HQ (AI search analytics), BrightEdge AI Catalyst (enterprise tier), Semrush AI Toolkit (integrated). Tool comparison and selection criteria in framework-aicitations.md. For AIO specifically, any of these surfaces the citation events. Choice is more about budget and existing tool stack than AIO specific capability.

10.5 Server Log Analysis

Access logs reveal bot patterns on a self hosted origin:

# Google-Extended visits in last 7 days
sudo awk '/Google-Extended/' /var/log/nginx/example.com.access.log | wc -l

# All AI engine bots aggregated
sudo awk '/GPTBot|ClaudeBot|PerplexityBot|Google-Extended|OAI-SearchBot/' /var/log/nginx/example.com.access.log | wc -l

Sudden drops in Google-Extended or Googlebot frequency on a previously cited page can indicate de prioritization in the candidate pool. The nginx access log on a self hosted Debian or Bubbles origin gives full visibility.

10.6 Conversion Lift Measurement

The 23x conversion rate finding is sample dependent. Measure site specific lift: GA4 with default channel grouping treating Organic Search distinctly, segment by landing page, compare conversion rate on cited vs non cited pages over rolling 90 day windows. Lift will not be 23x for every site but will be substantial enough to justify optimization investment for any business with measurable organic conversions.

10.7 Metrics to Avoid

No such thing as AIO rank position. AIO position is not a valid metric. Single day citation snapshots misrepresent volatility. AIO impressions bundled with classic in GSC make standalone impression tracking impossible. Use binary citation tracking across a sample of priority queries with weekly cadence.

11. Multi Engine Awareness

11.1 AI Overview vs AI Mode Citation Overlap

13.7 percent citation overlap means optimization for one does not deliver the other. Treat them as related but distinct surfaces. AI Mode runs higher fan out (9 to 16 sub queries vs 8 to 12), has Personal Intelligence integration since January 22 2026 that shapes responses per user, and has lower self overlap (9.2 percent vs 30 percent). Sample AI Mode separately. Manual sampling protocol same as Section 10.3 but in the AI Mode tab.

11.2 AI Mode Specifics

75 million daily active users as of January 2026, over 1 billion queries per month. Sub queries exposed in UI for some prompts (use to reverse engineer fan out). Personal Intelligence means individual users see different answers including different citations. Optimization patterns functionally identical to AIO with one addition: broader topical coverage matters more because higher fan out creates more sub query surface area.

11.3 Bing Copilot Inline AI Summaries

Bing Copilot inline AI summaries are a parallel surface on Bing SERPs. Copilot synthesis prefers .gov, .edu, .org institutional sources (heavier than Google), exact match keyword presence (more literal than Google's semantic matching), and Bing Webmaster Tools verified properties. Pages optimized for AIO generally perform on Copilot with institutional source weighting being the main differentiator.

11.4 ChatGPT Search, Perplexity, Claude, Meta AI

Citation patterns diverge from AIO in ways that warrant separate treatment. The broader AI citation framework covers each engine specifically. See framework-aicitations.md. The scope handoff: this framework stops at the Google AI surface boundary.

11.5 Why AIO Optimization Does Not Guarantee Other Engines

Different engines weight signals differently. ChatGPT favors topical authority hubs and recently updated content. Perplexity favors primary sources, academic citations, recency. Claude favors analytical voice, balanced perspectives, detailed methodology. Bing Copilot favors institutional sources. Meta AI favors content surfaced via Facebook and Instagram engagement.

11.6 Cross Engine Tradeoffs

Optimization choices that lift one engine sometimes reduce performance on another. The cross engine tradeoff matrix lives in framework-multiengine-tradeoffs.md, scheduled for Phase 3. Until that exists, the operative rule: optimize for AIO first (largest US user base via Google), AI Mode second, then engine specific tuning per the aicitations framework.

12. Common Mistakes and Anti Patterns

12.1 Optimizing only for top 10 organic. Surfer December 2025 and Ahrefs February 2026 prove 62 percent of AIO citation opportunity is outside top 10. Fix: optimize for AIO eligibility signals first, classic ranking second.

12.2 Schema injected via JavaScript. Invisible to AIO parsing. Fix: move to server side template rendering. Verify with curl -A "GPTBot" -s ... | grep -c "application/ld+json".

12.3 Lede buried under intro paragraph. The first paragraph under H1 is treated as the answer surface. Fix: lede 40 to 75 words immediately under H1, citeable standalone, specific facts.

12.4 Generic answers without specific data. AIO synthesis preferentially extracts specific facts. Fix: replace generic claims with specific ones. "Quarterly taxes are due quarterly" becomes "Quarterly taxes for 2026 are due April 15, June 15, September 15, 2026, and January 15, 2027."

12.5 Author bylines without verifiable credentials. AIO citation drops for unverified authors, particularly YMYL. Fix: /authors/[slug]/ page with ProfilePage schema, jobTitle, hasCredential, sameAs to LinkedIn and professional registries.

12.6 YMYL content without reviewer credit. Systematically excluded from AIO citation. Fix: credentialed reviewer named in visible text and Article reviewedBy property.

12.7 Refusing to refresh because organic traffic is stable. Pages ranking classically but not earning AIO citation look healthy in GSC while losing share of the larger surface. Fix: track AIO citation as a separate metric, refresh to add Information Gain and structural extractability.

12.8 Treating regeneration volatility as fixable. Each drop is not a remediation event. Fix: measure across rolling four week windows. Sustained drop over 28 days is a remediation trigger.

12.9 Chasing AIO presence on queries the site has no authority on. Information Gain is non negotiable. Fix: maintain topical focus. AIO amplifies authority that already exists.

12.10 Over rotating after a single regeneration drop. Most common failure mode. Fix: do nothing for two weeks. Re sample. If sustained, apply Section 9.7 discipline.

13. AI Overview Audit Rubric

13.1 First 90 Days Subset (Per Page)

For a new client engagement, these five items move citation probability fastest. Audit and remediate before deeper rubric work.

A page passing these five is in the AIO candidate pool. A page failing any one is not, regardless of other work.

13.2 Per Page Audit Rubric (Full)

Score 15. World class AIO ready page: 13 or higher with zero F1 to F5 fails.

13.3 Site Wide Audit Rubric

Score 10. World class AIO ready site: 9 or higher with zero critical fails on S1, S2, S5, S9.

14. Maintenance Schedule and Report Templates

14.1 Maintenance Cadence

Weekly. Sample top 10 priority queries in Google search and AI Mode, record citation state. Review GSC AI Overview impression delta. Check server logs for Google-Extended and Googlebot frequency on priority pages. Refresh one priority page with Information Gain addition.

Monthly. Sample next 50 priority queries. Run schema validation sweep on priority pages. Review competitor pages entering the candidate pool. Refresh time sensitive content per Section 9.3.

Quarterly. Site wide Section 13 audit. Refresh evergreen content. Review brand mention pace against Section 9.6 targets. Calibrate against current AIO state data.

Annually. Full framework review against current AIO state. Refresh underlying frameworks (contentfirst, eeat, infogain, ymyl, schema) against current state. Strategic review of priority query selection.

14.2 Implementation Report Template

# AI Overview Framework Implementation Report

**Site**: {{BUSINESS_NAME}}
**Implementation Date**: {{TODAY}}

## Summary
- Priority pages audited / passing first 90 days subset / passing full rubric
- Site wide rubric score: X/10

## Baseline AIO Citation State
- Priority queries sampled, queries with site cited in AI Overview, queries with site cited in AI Mode, GSC AI Overview impression baseline (28 day)

## Substrate Work
- Pages migrated from client side to server rendered
- Schema blocks moved from JS injection to server rendered head
- curl test pass rate before vs after

## Schema, E-E-A-T, Information Gain Work
- Counts per category, what was added, what was removed

## Tracking Infrastructure
- GSC baseline captured, manual sampling cadence documented, third party tool selected, conversion lift measurement deployed

## Sign-Off

14.3 Audit Report Template

# AI Overview Framework Audit Report

**Site**: {{BUSINESS_NAME}}
**Audit Date**: {{TODAY}}

## Executive Summary
One paragraph assessment.

**Site wide score**: X/10
**Average priority page score**: X/15
**Current AIO citation rate** (priority queries, weekly sampled): X percent

## Findings by Section
Substrate, schema, E-E-A-T, Information Gain distribution

## Critical Failures
List with remediation

## First 90 Days Subset Findings
Per priority page table of F1 to F5 results

## Recommended Remediation Order
Critical, High, Medium

## Sign-Off

End of Framework Document

Document version: 1.0 Created: 2026-05-14 Maintained by: ThatDeveloperGuy

Google AI Overviews and Google AI Mode are the dominant visibility surface for 48 percent of US searches and over 70 percent of informational queries in Q1 2026. Citation decoupling means classic ranking position no longer gates access. The signals that drive citation are structural readability, entity authority, E-E-A-T credentials, and Information Gain. Sites that build on those signals win sustained citation across the volatility. Sites that ignore them lose share regardless of how strong their classic SEO position remains.

This framework consolidates the AIO specific operational guidance previously distributed across the contentfirst doctrine, the aicitations multi engine reference, and the SEO-Search-Appearance and SERP-Optimization manuals. Apply it after framework-contentfirst.md, in parallel with framework-eeat.md, framework-infogain.md, framework-schema.md, and framework-entitysalience.md.

Companions

• framework-contentfirst.md, substrate doctrine, the prerequisite for AIO eligibility
• framework-aicitations.md, broader AI citation across ChatGPT, Claude, Perplexity, Bing Copilot, Meta AI
• framework-hcs.md, Helpful Content System
• framework-infogain.md, Information Gain
• framework-eeat.md, E-E-A-T pillars
• framework-ymyl.md, YMYL classification and credentialing
• framework-sqrg.md, Search Quality Rater Guidelines
• framework-schema.md, the graph pattern and schema completeness
• framework-entitysalience.md, entity declarations and salience scoring
• framework-knowledgegraph.md, Knowledge Graph and Wikidata
• framework-multiengine-tradeoffs.md, cross engine optimization tradeoff matrix (scheduled Phase 3)
• SEO-Search-Appearance.md, multi engine surface map
• SERP-Optimization.md, feature targeting playbook
• 14 tier Engine Optimization Stack, Tier 3 AI Search Domination