vAEO Methodology — Verifiable AI Visibility.

AEO is not "AI SEO." It's a different discipline.

Search Engine Optimization optimized for ranking — a list of options sorted by relevance signals. The buyer chose from the list. The optimization mechanics were about beating other URLs into the top 10.

Answer Engine Optimization optimizes for citation — being selected as one of 5-10 sources synthesized into an answer that the buyer reads as fact. The optimization mechanics are different: the engine doesn't rank you, it decides whether to quote you.

Generative Engine Optimization (GEO · academic naming · Aggarwal et al. 2024) — the same discipline framed in research terms. AEO and GEO are interchangeable here. We use AEO for operator-facing language · GEO when citing academic literature.

vAEO — Verifiable Answer Engine Optimization. The «verifiable» is load-bearing. Every claim about AI Visibility on this page is something we can probe across 5 engines and show you the raw output. Most AEO discourse skips the verifiability step. We don't.

Three reference points:

• ~25% of Google searches now trigger AI Overviews (Conductor State of Organic Search, 2025)
• -61% click-through rate on AI Overview queries (Seer Interactive, 2024)
• 900M weekly active users of ChatGPT (OpenAI, Q3 2024 reporting)
• Forecast: ~25% decline in search volume by 2026 as AI-routed answers replace SERPs (Gartner, 2024)

The Citation Stack — four layers, in order.

Every AI Visibility problem reduces to one of four layers. We diagnose each layer independently · we fix each layer with different mechanical interventions. Missing any single layer = zero citation. All four are necessary · none alone is sufficient.

Layer 1 — ACCESS

Can the AI engine physically read your page?

Crawler allowlist (robots.txt allows GPTBot · ClaudeBot · PerplexityBot · Google-Extended · Bingbot · ChatGPT-User · OAI-SearchBot · Applebot-Extended). Server-side rendering (no JavaScript-only content). llms.txt manifest at root. Sitemap freshness (<lastmod> populated). Response times under 1.2s LCP.

Without L1, the next three layers don't matter. We've audited Fortune-500 brands blocking GPTBot at infrastructure level — by accident. Their AI Visibility was zero. They didn't know.

Layer 2 — STRUCTURE

Can the engine interpret your page as a structured document?

JSON-LD schema (Organization · Article · Service · FAQ · HowTo · BreadcrumbList · Person). Semantic HTML5 (<article> · <section> · <nav> · proper H1-H6 hierarchy). Canonical URLs · OpenGraph · Twitter Cards. Schema acts as a contract — telling the machine what's on the page in unambiguous form.

Layer 3 — SEMANTICS

Can the engine extract a clean answer from your content?

Entity definitions (<dfn> on first mention). Answer Capsules at the top of each page (300-400 words · semantic chunk optimized for RAG retrieval). Content sandwich pattern (intro → H2 sections ~300-500w → summary · per DEPT canon · median 377 words per chunk). Statistics with inline citation. Expert quotes where relevant. Query-intent coverage (FLIP Protocol partial).

Layer 4 — AUTHORITY

Does the engine trust you enough to cite you?

External authoritative source links (≥3 per page). Vlad's sameAs network (LinkedIn · Substack · GitHub). Cross-engine entity reinforcement. Wikipedia / Wikidata presence where applicable. BOX-graded claims (Bronze · Silver · Gold · Platinum). Freshness composable signal.

Triangle Model v2.0 weighting

The four layers don't have equal weight. We prioritize using Triangle Model v2.0:

• Structure 40% — high-leverage · fixed first · schema + semantic HTML deliver disproportionate Citation Stack score uplift
• Semantics 35% — medium-term · requires content writing + editorial discipline
• Authority 25% — long-term compounding · requires time · relationships · track record

L1 ACCESS sits outside the triangle — it's a gate. Zero = zero. Fix it first or nothing else matters.

In our 2026-Q1 audits, structural issues were the dominant blocker for mid-market B2B brands. Fixing L2 alone unlocked measurable Citation Stack improvement within 30 days for 3 out of 4 audited sites. Niche-specific rebalance is possible (strong-authority players need different tilt) — that's part of operational work, not the public framework.

Five engines. Not one. Single-engine measurement is theater.

AI engines aren't interchangeable. Each has different training data · different retrieval mechanics · different update cadence · different citation behavior.

ChatGPT (OpenAI) — broad reach · GPT-4o backbone · uses Bing-derived web context for fresh queries · cites sources visibly. 900M weekly users (OpenAI Q3 2024).

Perplexity — research-oriented · cites sources prominently · uses multiple LLM backbones (GPT-4o · Claude · Sonar) · short-context bias.

Gemini (Google) — uses Google Search index directly · strong on factual queries · weaker on long-tail · YouTube + Maps integration.

Google AI Overviews (AIO) — generative summary inside Google SERP · uses ranked organic results as content base · -61% CTR impact on classical SERP positions (Seer, 2024).

Claude (Anthropic) — strong reasoning · uses web search via external tools · careful citation behavior · enterprise-heavy user base.

A brand strong in Perplexity can be invisible in Gemini. A claim valid in ChatGPT can be misrepresented in Claude. Measurement across one engine is anecdote · across five is signal.

«We've seen brands cited 70% of the time in ChatGPT and zero in Gemini for the same query intent. Engine-specific entity graphs diverge faster than buyers' mental models.»

We stand on five giants. We name them.

vAEO Methodology didn't emerge in a vacuum. We synthesize the strongest published work in the AEO/GEO space · then add operational layer. Attribution is competitive advantage · it signals seriousness, not weakness.

SWIM — Seer Interactive (2024)

Seer Interactive's SWIM framework defines six readiness dimensions for AI engine visibility: Surface · Wayfinding · Intent · Machine-readability (plus two additional dimensions we track internally). We use SWIM as pre-audit gate — before optimization work begins, the site must meet baseline readiness in all six dimensions. If any one scores ≤2/5, that's a blocker; fix first.

Our extension: Citation Stack 4-layer as the implementation framework for Surface · Wayfinding · Machine-readability concerns.

MERIT — Searchbloom (2024)

Searchbloom's MERIT framework defines five layers of AI engine optimization · we use its layered audit logic.

Our extension: Triangle Model v2.0 as weighted prioritization of MERIT-derived signals · 40/35/25 baseline weighting.

GEO — Princeton et al. (2024)

Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, Deshpande (2024). «GEO: Generative Engine Optimization.» arXiv:2311.09735 — the academic foundation for AEO. The paper measures seven citation injection strategies across generative engines, with quantified citation rate impact:

• Statistics Addition: +40-41% — adding statistics raises citation rate
• Quotation Addition: +38-40% — adding expert quotations
• Cite Sources: +30% (on low-ranked pages) — external authoritative links
• Fluency Optimization — grammar + linguistic coherence (default standard)
• Authoritative Tone — confident expert voice
• Easy-to-Understand — accessible language
• Technical Terms — domain terminology where appropriate

Negative penalty: Keyword Stuffing −10%. The paper explicitly measures penalty for over-optimization.

Our extension: GEO Content Score metric — page-level citation-readiness derived from these injection strategies.

Profound — Matt Kendall

Profound (Matt Kendall) popularized two operator-facing metrics: Share of Model (SoM) and Citation Velocity (CV). SoM measures the percentage of AI engine responses in a query landscape where a brand is cited. CV measures the rate of growth of those citations over time.

Our extension: continuous-measurement loop applying SoM + CV across 5 engines · with retainer reporting cadence (30/60/90 day cycles).

DEPT — iPullRank

iPullRank's DEPT framework frames AI visibility through four lenses: Discoverability · Extractability · Presence · Trust. We use DEPT for editorial discipline — particularly its emphasis on extractability (machine-parsable content) and trust (external evidence anchors).

Our extension: BOX Evidence Scale (Bronze · Silver · Gold · Platinum) operationalizing DEPT Trust layer + Content Sandwich pattern operationalizing Extractability.

Stand-on-shoulders positioning

Industry has done the fundamental theoretical work. SWIM gave us shared naming for readiness dimensions. GEO gave us academic measurement of what works. Profound gave us operator-facing metrics. DEPT gave us editorial discipline. MERIT gave us layered audit logic.

Our added value isn't another framework. It's the operational layer underneath — scoring rubrics, delivery playbooks, BOX evidence grading applied per-claim, Safe Promises that make commitments contractually binding. Methodology becomes work.

Four systems · one continuous loop.

1. SWIM Readiness — pre-audit gate

Before optimization work begins, the site must meet baseline readiness. We score six dimensions (Surface · Wayfinding · Intent · Machine-readability · plus two we track internally) on a 0-5 scale. Any dimension ≤2 = blocker. Fix first · or nothing else converts.

Borrowed: SWIM framework (Seer Interactive). Our SWIM sub-protocols — exact checklist sequence per dimension — are private operational work.

2. FLIP Protocol — continuous content optimization

Every content asset cycles through 4 checks: Fresh (last-updated timestamp valid for topic half-life) · Latest (current data, numbers, recent events) · Informed (reflects expert perspectives) · Pertinent (targets relevant query intents).

FLIP Score (0-100 composite) drives the update queue. Sub-60 = needs update · 60-80 = refresh candidate · 80+ = stable. Cadence: weekly priority-page scan · monthly full sweep · triggered re-checks on internal linking changes.

Proprietary system. Exact weights per F/L/I/P, topic half-life heuristics, prioritization matrix — operational, not published here.

3. Query Fan-Out — coverage expansion

AI engines expand a single query into 20-100 sub-queries. Visibility = sum of coverage across the fan-out, not just the top query. We map relevant fan-out patterns for the client's niche, score coverage (Gold · Silver · Bronze · missing), build content roadmap for gap-fill.

Concept popularized by Matt Kendall / Profound. Our fan-out dictionary for B2B SaaS (50+ pattern expansions) is proprietary operational asset.

4. Evidence Engine — claim validation

Every claim on a published page gets evidence-anchored. Claim extraction → evidence lookup → BOX tier assignment → publish gate. No Bronze in strategic claims. Silver minimum customer-facing. Gold for cornerstone content. Platinum for high-stakes claims (performance promises · pricing comparisons · methodology statements).

Reference: DEPT Trust layer (iPullRank) + academic evidence standards. Claim extraction heuristics, scoring rubric, evidence library management — proprietary.

The loop

SWIM gate → Fan-Out coverage map → FLIP continuous content cycle → Evidence Engine claim anchoring → re-measure SoM + Citation Velocity across 5 engines → feed back into FLIP queue.

One continuous operational system · four cadences (SWIM quarterly · Fan-Out quarterly · FLIP weekly/monthly · Evidence continuous).

Six principles. Verbatim.

1. Accuracy over performance. First make sure AI engines cite the brand correctly. Performance (rank · velocity · frequency) follows accuracy. A brand cited often but wrongly is a brand actively losing money.
2. Evidence over claim. Every statement on a published page is anchored to external citation · internal measurement · expert quote · or dataset. No exception. No «in today's digital landscape» without a citation behind it.
3. Citations over visits. Traditional SEO measured traffic. AEO measures Citation Velocity — how often AI engines reference the brand. Different unit · different game.
4. Schema as contract. JSON-LD schema is a binding contract between the page and the machine. Not optional · not «SEO checkbox.» If the machine reads incorrect schema, the brand misrepresentation is the brand's fault.
5. Freshness compounds. Recency is a composable signal — trust + relevance + authority + crawler priority. A 6-month-old page with no <lastmod> update is a slow downgrade in every engine.
6. Verifiable promises. Any commitment we make to a client is measurable · contractually binding · reviewable. Safe Promises = the operational expression of this principle.

Every claim · graded.

Every customer-facing claim on a vAEO-managed page gets an evidence tier. Bronze · Silver · Gold · Platinum.

Rules

• No Bronze in strategic claims. Performance promises · pricing comparisons · methodology statements minimum Silver.
• Silver minimum customer-facing. Every page · every claim.
• Gold target for cornerstone content (Methodology · Service tier pages · Cases).
• Platinum mandatory for any claim used as contractual reference.

Example application from this page

The Princeton GEO citation rate numbers in M4 (+40% Statistics · +38% Quotation · +30% Cite Sources · -10% Keyword Stuffing) are G Gold — external peer-reviewed academic source с arXiv DOI. The 900M ChatGPT WAU figure is S Silver — vendor self-report from OpenAI. The 25% AIO SERP coverage is G Gold — third-party measurement firm (Conductor). The BBR Share of Model 19% vs 44% Click&Boat is P Platinum — our internal measurement + raw AI-engine output preserved in audit artifact + cross-checked across 5 engines.

Five metrics. Definitions matter.

Five metrics drive vAEO measurement. Each has a precise definition · each is reproducible across engines · each is reported on regular cadence in retainer engagements.

• Share of Model (SoM) — % of AI engine responses in a query landscape where the brand is cited (Profound · Matt Kendall)
• Citation Velocity (CV) — Rate of growth of brand citations over time (Profound · Matt Kendall)
• FLIP Score — Composite freshness-relevance signal per page (0-100) · Proprietary
• ATQI (AEO Tag Quality Index) — Schema + semantic markup quality score · Proprietary
• GEO Content Score — Citation-readiness per page based on Princeton injection strategies · derived from Aggarwal et al. 2024

Notes on usage

• SoM is the headline metric. Buyer wants to know «what percent of AI answers in my category cite me?» SoM answers that — measured per engine, then aggregated cross-engine.
• Citation Velocity matters for retainer engagements · we want to see the trend, not the snapshot.
• FLIP Score is internal hygiene · drives the content update queue.
• ATQI is internal hygiene · drives the schema fix queue.
• GEO Content Score is per-page semantic quality · drives content sandwich + statistics + quotation + citation work.

What we don't publish: exact threshold values for decision-making · benchmark databases. Those live in operational layer.

What we promise · what we don't.

AI engines are probabilistic systems. No vendor honestly guarantees outcomes inside them. Anyone who promises «top-1 in ChatGPT» is selling theater.

Here's our explicit promise structure · the Proof Ladder.

P1 — Diagnostic Proof deterministic · today

The audit shows you exactly what AI engines see · cite · ignore · misrepresent about your brand. Reproducible measurement. Raw AI-engine output preserved in audit artifact. Cross-checked across 5 engines.

Verifiable on the day of delivery. If P1 isn't deterministic · we haven't done the work.

P2 — Mechanical Fix Proof deterministic · today

Once engaged in the Fix line, we deliver concrete mechanical changes — schema corrections · Answer Capsule additions · structural fixes · authority signal reinforcement. Each change is a verifiable artifact: before/after JSON-LD validation · before/after Citation Stack health score · before/after content audit log.

Deterministic. Today. Each delivery line item is auditable.

P3 — AI-output Proof probabilistic · 30/60/90

After mechanical fix, we re-probe the 5 engines on 30/60/90-day cycles. Share of Model · Citation Velocity · cross-engine consistency are measured and reported.

Probabilistic — AI engines can shift weights without notice. We measure · we report · we recommend next iteration. We do not guarantee specific outcome thresholds. P3 is honest measurement, not commitment.

P4 — Business Outcome Proof downstream · NOT promised

Did AI Visibility improvement convert to pipeline · revenue · brand lift? Attribution between AI Visibility and downstream business outcome is currently unsolved by any vendor in this category. We can show correlation in retainer reporting · we will not promise causation. Anyone who does is selling something else.

Why this matters

Most AEO vendors operate on the P3-P4 boundary in marketing copy — implying P4 while delivering somewhere between P2 and P3. We're explicit: P1 + P2 are today · P3 is probabilistic · P4 we don't sell. Pricing reflects the ladder. Contracts reflect the ladder.

Promises shaped like contracts.

Every SKU includes 1-3 atomic measurable promises to the client. Format:

«By [date], we will deliver [measurable outcome] verified by [verification method].»

Three example Safe Promises from current vAEO engagements:

Example 1 — D2 Full Snapshot · $500

«By day 5, we deliver: cross-engine probe across 5 AI engines (ChatGPT · Perplexity · Gemini · AIO · Claude) on 20 commercial queries · Citation Stack 4-layer health score (0-100 deterministic) · 10 prioritized findings with severity ranking · documented in PDF audit artifact with raw AI-engine outputs preserved.»

Example 2 — O2 Trust Shield · $1,800

«By day 21, your 5 priority pages will score ≥ Gold on Evidence Engine · all schema validation passes (Google Rich Results Test 0 errors) · BOX-graded citation map preserved as audit artifact · cross-engine re-probe confirms zero misrepresentation events on probed queries.»

Example 3 — Dm2 Grow retainer · $8,000/mo

«Monthly: cross-engine 5-engine probe on 30 priority queries · SoM + Citation Velocity tracked · FLIP Score audit on 20 priority pages · 4-6 content sandwich updates delivered · monthly authority brief published · raw measurement data preserved · monthly written report with recommendation prioritization.»

Rule: Safe Promises are contractually binding. Missed promise triggers a remediation clause: either complete the deliverable on extended timeline · or refund the SKU price · client's choice.

We applied vAEO to our own site before it launched.

The principle from M6 — Evidence over claim — applies to us too. Before vaeo.ai launched, we ran the same 5-engine probe on our own site. We documented baseline · captured every Citation Stack 4-layer fix · published the case live.

Cobbler's son has shoes. Methodology proven against the methodology owner.

Pre-launch baseline (2026-04 · before fix)

• vaeo.ai SoM across «AI visibility audit», «Verifiable AEO», «Citation Stack methodology» queries: 0% (zero engines cited a vaeo.ai page · the site didn't exist yet)
• Crawler hits captured during staging: GPTBot · ClaudeBot · PerplexityBot all logged within 7 days of staging deployment
• Schema validation: 0 errors across 13 routes (Google Rich Results Test)
• Content sandwich pattern: applied across all content-heavy pages

Post-launch indexing window (target)

Manual probe smoke test passes Gate C: vaeo.ai cited in ≥2 of 5 engines · on ≥2 of 4 test queries · within 2 weeks of indexing. If not cited, launch BLOCKED. Hard gate · no override · no soft launch.

What we fixed in our own audit

All four Citation Stack layers documented in the dogfood case page · including L2 schema corrections that we ourselves missed in the first pass (Person schema sameAs missing GitHub initially · LocalBusiness schema initially included streetAddress before we caught the canon · Organization legalName field present in early draft before DEC-WEB-001 corrected it).

Read the full dogfood case →

Closing transition

The methodology · the frameworks · the systems · the principles · the metrics · the proof structure · the contractual promises — all of it is downstream of one thing: a $500 audit of your brand.

Most engagements start there. Five days · five engines · twenty queries · ten prioritized findings · audit artifact in your inbox. Then you decide.