Executive Summary
Search is no longer the first place customers look for an answer. Physicians ask ChatGPT to summarize a mechanism of action before opening a PI. Advisors ask Perplexity to compare product suitability before opening a prospectus. Procurement teams ask an AI assistant to shortlist vendors before a single RFP is issued. In each of these moments, a brand is either cited, misquoted, or absent — and none of the three outcomes is under that brand's direct control unless its content has been deliberately engineered to be citable.
This is the AI visibility problem, and it is already industry-wide. Independent studies of 177 brands across healthcare, SaaS, and financial services found that roughly nine in ten have zero mentions across major AI answer engines. At the same time, AI-powered search has moved from a novelty to a majority behavior: a large and growing share of AI Overview and AI Mode queries end without a single click to any website, meaning the AI's summary — not the brand's own page — is now the primary customer touchpoint.
The AI Visibility Optimization Framework™ (AVO) is travalcon's response: a repeatable, three-phase discipline — Discover, Build, Measure — that identifies where AI systems are already forming opinions about a brand's category, restructures content so it is technically citable by retrieval-augmented generation systems, and measures visibility with the same rigor a brand applies to paid media or SEO. Because AVO is built on the same BCB-anchored content architecture that underpins travalcon's Modular Content and Tagging & Taxonomy frameworks, every citation-ready asset it produces is also MLR-compliant, provenance-tagged, and strategically anchored — not a marketing hack layered on top of ungoverned content, but a natural extension of governed content production into a new distribution channel that happens to be a language model rather than a search results page.
This whitepaper defines the AVO framework in full, positions it against the fast-growing and largely ungoverned category of generic AI-visibility monitoring tools (using Sight AI as a representative example), and sets out a phased path for regulated-industry organizations to move from invisible to cited.
- 8–12 weeks to first measurable visibility gains on prioritized Focuses
- 4 major LLMs benchmarked per Visibility Score audit (ChatGPT, Claude, Perplexity, Gemini)
- 3–6 months to commercial ROI once Discover → Build → Measure is operating as a continuous loop
- ~90% of 177 brands studied across healthcare, SaaS, and financial services show zero AI search mentions — the gap AVO exists to close
- 43–93% of AI-powered searches now end in zero clicks to any website, depending on query type and engine
The organizations that win the next decade of category authority will not be the ones with the most content — they will be the ones whose content AI systems trust enough to cite. AVO is how a regulated-industry brand earns that trust deliberately, rather than discovering its absence by accident.
1. AI Is the New First-Line Answer Engine
For two decades, the default customer journey began with a search engine results page: ten blue links, a click, a landing page. That journey is being replaced. Physicians, HCPs, financial advisors, and B2B buyers increasingly put their question directly to an AI system — ChatGPT, Google's AI Overviews and AI Mode, Perplexity, Claude — and act on the synthesized answer without visiting a source at all. The AI's summary has become the product; the underlying website has become an input the AI may or may not choose to cite.
1.1 The Zero-Click Reality
Multiple independent studies converge on the same finding: a majority of AI-mediated searches now end without a click. Roughly 43% of searches that trigger a Google AI Overview end in zero clicks to any website; when Google's more conversational AI Mode is used, that figure rises to approximately 93%. Across all US Google searches, zero-click behavior overall is estimated at close to 58.5%. For a brand whose growth strategy depends on being visited, this is a structural threat. For a brand whose growth strategy depends on being cited, it is a structural opportunity — but only if its content is built to be citable in the first place.
1.2 The Citation Fragmentation Problem
Even organizations that do appear in AI answers cannot assume consistency across engines. Citation behavior varies enormously by platform — one comparative analysis found a roughly 46-fold difference in citation rates between the lowest and highest performing engines for the same content (ChatGPT around 0.59% versus Perplexity around 13.05% in that sample). More strikingly, only about 2% of cited URLs appear across all three major AI engines simultaneously; 91% of cited URLs appear in only one engine's answers. Being visible in one AI system says almost nothing about being visible in another — visibility must be measured and managed per engine, not assumed to generalize.
1.3 Missing Presence, Missed Opportunity
The compounding effect of zero-click behavior and citation fragmentation is a near-total blind spot for most brands. Independent research covering 177 brands across healthcare, SaaS, and financial services found that approximately 90% had zero mentions across the AI answer engines studied — not low visibility, but none. For regulated industries specifically, this is amplified by content that was never structured for machine extraction: dense PDFs, unstructured PI documents, and marketing pages written for human persuasion rather than machine-verifiable claims. The brands actively investing in AI visibility today are competing for a category-defining position against near-empty search results — the earliest and least contested a channel has been in a generation.
2. The AI Visibility Optimization Framework™: Conceptual Foundation
AVO treats AI visibility as an engineering discipline, not a copywriting trick. It rests on the premise that large language models and retrieval-augmented generation (RAG) systems do not reward the same signals as traditional search — they reward content that is unambiguous, evidence-backed, structurally chunkable, and machine-readable at the paragraph level. AVO organizes the work required to produce that content into three sequential, continuously repeating phases.
2.1 Three Phases, One Repeatable Workflow
| Phase | Core Question | Primary Output |
|---|---|---|
| Discover | Where does AI already form opinions about our category, and how exposed or at-risk is our current content? | Focus registry, Prompt Book, source-risk audit, baseline Visibility Score |
| Build | How do we restructure and produce content so retrieval systems can find, trust, and cite it? | Pyramid-chunked, provenance-tagged, schema-marked Knowledge Artifacts |
| Measure | Are we actually being cited, how accurately, and is it moving the business? | Visibility Score trend, RAG recall/precision data, leading and outcome KPIs |
- Generic AI-visibility tactics optimize for citation frequency alone. AVO optimizes for citation frequency AND regulatory defensibility.
- Every claim carries a source tag (DOI, PMID, or approved reference) before it is eligible for publication.
- Every published asset carries MLR metadata — author, medical reviewer, version, approval date — embedded in machine-readable form, not just a human-readable footer.
- The result: content that is simultaneously more likely to be cited by an LLM and fully auditable by Legal, Medical, and Regulatory.
3. Phase 1 — Discover: Find the Focuses AI Will Use
Discover establishes the baseline: what is AI already saying about the brand's category, where is the organization's own content exposed or absent, and which specific topic-intent combinations — Focuses — are worth prioritizing first.
3.1 Three Lenses on Where You Stand Today
| Lens | What It Examines | Typical Finding |
|---|---|---|
| Search AI Queries | The actual prompts HCPs, patients, advisors, or buyers put to AI systems about the category | Most brands have never systematically catalogued the real questions being asked about them |
| Content Inventory | Which existing assets are structurally capable of being retrieved and cited at all | The majority of legacy content is unstructured prose that RAG systems cannot chunk cleanly |
| Source Risk | Whether currently-cited sources (including competitor or third-party sources) are accurate, current, and favorable | Outdated or competitor content is frequently the source AI cites by default, in the brand's own silence |
3.2 Defining a Focus: Topic + Intent
A Focus is the atomic planning unit of AVO — a specific Topic paired with a specific Intent, precise enough to test and measure. "Mechanism of action" is not a Focus; it is a subject area. "How does [mechanism class] differ from [comparator class] for a treatment-experienced patient" is a Focus: it can be entered verbatim into an LLM, the response can be scored, and the presence or absence of the brand in that response can be tracked over time. Each prioritized Focus is logged in a Focus registry alongside its associated Prompt Book — the exact query variants used to test it.
3.3 Three Quick Wins Before Building Anything New
Discover typically surfaces immediate, low-cost corrections before any new content is commissioned: correcting or supplementing a Focus where a competitor or outdated source is being cited by default, adding machine-readable structure to an existing high-authority page rather than rewriting it, and closing a source-risk gap where an approved claim exists internally but has never been published in a citable form.
4. Phase 2 — Build: Create Content AI Will Cite
Build is where prioritized Focuses become published, machine-readable Knowledge Artifacts. The central discipline is structural: an LLM does not read a page top to bottom the way a human does — it retrieves the smallest chunk of text that answers a specific question, so content must be pre-chunked into citable units at the point of authoring, not left for a retrieval system to guess at.
4.1 The Pyramid Chunking Principle
| Layer | Format | Purpose |
|---|---|---|
| Direct Answer | ≤50 words, single unambiguous claim | The exact span an LLM extracts and quotes or paraphrases as its answer |
| Context Paragraph | 2–3 sentences | Supplies the qualifying context a retrieval system attaches around the direct answer |
| Evidence List | Bulleted references with DOI / PMID / source ID | Gives the retrieval system (and a human fact-checker) an auditable trail back to the primary source |
4.2 Evidence & Provenance
Every claim in a Knowledge Artifact carries an individual source tag rather than a single blanket reference list at the end of the document — this is what allows an LLM (and a compliance reviewer) to verify a specific sentence rather than an entire page. Bibliographic data is additionally marked up in schema.org / JSON-LD format so that citation metadata is machine-parseable, not just visually formatted for a human reader.
4.3 Knowledge Artifacts & Graph Signals
Three artifact types carry most of the Build workload: Topic Hubs (canonical, pyramid-chunked pages per Focus cluster), the Prompt Book (the living catalogue of query variants a Focus should answer correctly), and structured JSON-LD markup (making entity relationships explicit for both search engines and LLM retrieval layers). Where a Knowledge Graph is already in place, these artifacts are anchored directly to the graph's entity and relationship layer, giving AVO content an unusually strong, auditable trust signal relative to unstructured competitor content.
4.4 Production & MLR Flow
AVO does not bypass medical, legal, and regulatory review to move faster — it restructures the review queue around smaller, source-tagged units so that review remains rigorous while cycle time falls. Draft generation is templated and partly automated; every artifact still requires human sign-off before publication, with author, reviewer, version, and approval date embedded as metadata rather than left in an email thread.
5. Phase 3 — Measure: Know Where You Rank in AI Answers
Measure closes the loop: it quantifies whether Build actually changed what AI systems say, and it feeds findings back into Discover so the next cycle of prioritization is evidence-based rather than intuition-based.
5.1 The Visibility Score: One Number, Five Components
| Component | What It Captures |
|---|---|
| Citation Frequency | Share of canonical prompts, across N benchmarked LLMs, where the brand appears among the top cited answers |
| Answer Accuracy | Whether the AI's summary of the brand's position is factually correct against approved claims |
| Context Suitability | Whether the brand is cited in an appropriate context (e.g., correct indication or use case) |
| Sentiment | Whether the tone of the AI's citation is neutral/favorable or inadvertently negative |
| Source Trust Score | Whether the AI is citing the brand's own governed content, or a secondary/competitor source describing the brand |
5.2 RAG & Retrieval Testing
Beyond the headline Visibility Score, AVO tests the underlying retrieval mechanics directly: recall and precision against the Prompt Book, ongoing hallucination monitoring (does the AI ever fabricate a claim attributed to the brand), and structured A/B testing of content structure variants to isolate which formatting choices measurably improve citation rates.
5.3 Leading and Outcome KPIs
Leading indicators — Visibility Score lift, citation share by engine, RAG recall@k, MLR cycle time, content-to-publish hours — show whether the operating model is working. Outcome indicators — HCP engagement lift, demo or inquiry volume attributable to AI-referred traffic, prescribing or purchase signal shifts, and faster MLR cycle times — show whether it is translating into business results. Both sets are reviewed on the same cadence as paid media and SEO performance, not as a separate, lower-priority workstream.
6. What the Content Actually Needs to Look Like
AVO's content-structure guidance is not stylistic preference — it is derived from observed, measurable differences in citation behavior. Independent structural testing across large samples of AI-cited pages has found that specific formatting choices materially change citation odds.
| Structural Choice | Observed Effect on AI Citation |
|---|---|
| Comparison page built around a 3-column table | +25.7% citation rate versus prose-only equivalent |
| Validation or evidence page built as an 8-item list | +26.9% citation rate versus prose-only equivalent |
| Shortlist or summary page with sentences ≤10 words | +18.8% citation rate in ChatGPT specifically |
The common thread across all three findings is the same principle underlying Pyramid Chunking: retrieval systems reward content that has already done the extraction work for them — short, unambiguous, structurally delimited units — rather than content that requires an LLM to infer structure from unstructured prose.
7. Governance: Why AVO Is Not a Marketing Tactic in Regulated Industries
In an unregulated category, AI visibility optimization is purely a growth-marketing exercise. In life sciences, financial services, and other regulated industries, the same activity carries compliance exposure that generic AI-visibility playbooks do not address: an AI system citing an off-label claim, an outdated indication, or a superseded safety statement is a regulatory incident, not a missed opportunity.
AVO addresses this by embedding governance at the point of content creation rather than as a downstream audit. Source-risk auditing in Discover flags where currently-cited content (including third-party content) may already be non-compliant. Per-claim source tagging in Build means every citable unit traces to an approved reference, not a blanket document-level disclaimer. MLR metadata embedded in the artifact itself — not a separate tracking spreadsheet — creates a durable audit trail regulators and internal compliance teams can review directly. And human sign-off remains mandatory before publication at every stage, regardless of how much of the drafting workflow is automated.
8. Market Validation: A Fast-Growing, Largely Ungoverned Category
AI visibility — often labeled Generative Engine Optimization (GEO) or Answer Engine Optimization (AEO) in market research — has moved from a niche concern to a board-level priority in under two years. Market estimates place the GEO category at roughly $1.1–1.5 billion in 2026, growing to an estimated $17–20 billion by 2034 at a compound annual growth rate above 45%. Fortune 500 CMOs have followed: roughly 67% now rank GEO among their top three marketing priorities for the coming year, up from just 18% two years earlier. The volume driving this shift is real — AI systems are estimated to be fielding more than 15 billion queries per month, roughly doubling year over year.
- Sight AI (trysight.ai) is a purpose-built platform for monitoring brand visibility across ChatGPT, Claude, Perplexity, and Gemini.
- Its capabilities include cross-platform citation monitoring, citation-context analysis (sentiment, competitor co-mentions, the specific query that triggered a citation), query-level tracking, competitive benchmarking, content performance insights, real-time alerts, and historical trend tracking.
- It is a strong illustration of where the market is heading — dedicated, always-on measurement of AI citation behavior is becoming table stakes, exactly as dedicated SEO tooling did a decade earlier.
The important distinction for a regulated-industry buyer is what tools like Sight AI — and comparable entrants such as PromptWatch, plus AI-citation modules now appearing inside established SEO platforms like Semrush, Ahrefs, and Conductor — are built to do, and what they are not. These are general-purpose measurement and monitoring platforms: they tell a brand where it stands in AI answers today, across any industry, for any content. None of them were designed to solve the harder, prior problem specific to regulated industries — producing the evidence-tagged, MLR-compliant, pyramid-chunked content that becomes citable in the first place, with an audit trail a regulator can inspect. AVO treats visibility measurement (the Measure phase) as one part of a three-phase discipline, not the whole of it; a Sight AI-style dashboard is a natural complement to Measure, but it does not replace Discover or Build, and it cannot make ungoverned content compliant simply by tracking whether it gets cited.
9. The Economics of Visibility: Why Citation Is Becoming a P&L Line
The business case for AVO rests on three converging figures rather than a single metric. First, the cost of continued invisibility: with roughly 90% of comparable brands already absent from AI answers, the near-term cost of inaction is not competitive parity — it is ceding an entire emerging channel to whichever competitor or third-party source the AI defaults to instead. Second, the speed of executive reprioritization: the jump in Fortune 500 CMOs naming GEO a top-three priority (18% to 67% in roughly two years) signals that budget and organizational attention are moving toward this capability faster than most internal roadmaps currently account for. Third, the realistic payback window: organizations following a structured Discover-Build-Measure cycle can expect measurable visibility gains on prioritized Focuses within 8–12 weeks, with commercial ROI typically materializing within 3–6 months once the cycle is running continuously rather than as a one-time project.
Taken together, these figures argue against treating AI visibility as an experimental side project. The category is moving from near-zero organizational maturity to board-level priority in real time; the organizations that build the Discover-Build-Measure muscle now, on governed content, will hold a compounding advantage over those that wait for the category to mature before acting.
10. Illustrative Program Outcome
Consider a specialty-care brand entering its AVO program with the typical starting position described above: fragmented content, no prior AI-visibility measurement, and a baseline Visibility Score effectively at zero across its top twenty prioritized Focuses. An initial Discover phase (2–3 weeks) catalogues the real HCP-facing prompts about the therapeutic area, audits existing content for structural citability, and flags three Focuses where a competitor's outdated content is currently the default AI citation.
A Build phase (4–6 weeks) converts the top-priority Focuses into pyramid-chunked, source-tagged Knowledge Artifacts, restructures two existing high-authority pages rather than rewriting them from scratch, and publishes schema-marked JSON-LD across the updated set. By week 8–12, re-testing against the same Prompt Book shows measurable Visibility Score gains on the prioritized Focuses — consistent with the 8–12 week benchmark for first measurable gains — with the previously competitor-dominated Focuses now returning the brand's own governed content as the cited source. Continuing the Measure-Discover-Build loop on a quarterly cadence is what converts an initial gain into the 3–6 month commercial ROI window observed across programs of this type.
11. Industry Deep-Dive: Life Sciences — Where AI Visibility Meets MLR
Life sciences is simultaneously the industry with the most to gain from AI visibility and the least tolerance for getting it wrong. Physicians are already heavy users of AI for literature search — roughly 74% report using AI tools in this way — meaning a therapeutic area's AI-cited content is already shaping clinical framing whether or not the manufacturer has engaged with the channel deliberately. Medical Information copilots and MSL-facing AI assistants increasingly retrieve directly from a brand's own published content (or its absence), making Build-phase discipline — per-claim source tagging, current safety information, MLR metadata — a direct extension of existing medical-legal-regulatory review processes rather than a new compliance burden layered on top of them.
Because AVO's Knowledge Artifacts share the same entity and relationship architecture as travalcon's Knowledge Graph Framework™, a life-sciences organization that has already invested in graph-based content intelligence gains a substantial head start: its evidence base, entity definitions, and relationship structures can be surfaced directly as AVO's Discover-phase content inventory, rather than rebuilt from scratch.
12. Industry Applicability: Financial Services & Industrial B2B
The AVO discipline generalizes beyond life sciences to any regulated or evidence-driven category where AI-mediated research now precedes a human decision.
| Industry | Where AI Visibility Already Matters | AVO Adaptation |
|---|---|---|
| Financial Services | Advisors and self-directed investors ask AI to compare product suitability, fees, and risk profile before a human conversation | Suitability-safe, per-claim-sourced answer blocks; compliance metadata in place of MLR metadata; source-risk audit against regulatory disclosure requirements |
| Industrial / B2B | Technical buyers and procurement teams use AI to shortlist vendors and compare specifications before issuing an RFP | Pyramid-chunked technical specification pages; schema-marked comparison tables; Prompt Book built from real procurement-stage queries |
13. Competitive Benchmarking: AVO-Aligned vs. Status Quo Brands
| Dimension | Status Quo Brand | AVO-Aligned Brand |
|---|---|---|
| AI citation baseline | Typically zero or near-zero across benchmarked engines | Actively measured and improved via a repeatable Visibility Score cycle |
| Content structure | Prose-first, written for human persuasion | Pyramid-chunked, evidence-tagged, machine-readable at the paragraph level |
| Compliance trail | Document-level disclaimer, if any | Per-claim source tag plus embedded MLR/compliance metadata |
| Visibility tooling | None, or a general-purpose monitoring dashboard (e.g. Sight AI) with no content-production layer | Discover–Build–Measure operated as one governed, continuous loop |
| Cross-engine consistency | Unmanaged — visibility, if any, is accidental and engine-specific | Deliberately benchmarked and tracked across all major LLMs |
14. Organizational Readiness for AI Visibility Programs
| Readiness Dimension | What Good Looks Like |
|---|---|
| Executive Sponsorship | A named senior owner (CMO or digital transformation lead) accountable for Visibility Score as a tracked business metric |
| Cross-Functional Access | Standing collaboration across Medical/Legal/Regulatory, Digital, and Brand — not a marketing-only initiative |
| Source-Risk Appetite | Willingness to proactively correct or supersede third-party and competitor content currently being cited by default |
| Content Production Capacity | A production workflow capable of pyramid-chunked, source-tagged authoring at the pace Discover prioritization demands |
| Measurement Infrastructure | A committed Prompt Book and recurring Visibility Score benchmarking cadence, not a one-time audit |
15. Strategic Implications for CMOs and Digital Transformation Leaders
Three implications follow directly from the evidence assembled in this whitepaper. First, AI visibility is not an SEO line-extension — it requires content restructuring at the authoring level, which means it belongs on the content operations roadmap, not solely the digital marketing roadmap. Second, the near-total absence of most brands from AI answers today is a closing window, not a permanent condition — the 18%-to-67% jump in CMO prioritization suggests competitors are moving now, and first-mover advantage in an AI engine's training and retrieval preferences compounds over time in ways that are difficult to reverse later. Third, in regulated industries, the winning approach will not be the fastest generic AI-visibility tactic — it will be the one that treats citation and compliance as the same engineering problem, because in these industries they are.
16. Five Lessons from AI Visibility Implementations
| # | Lesson |
|---|---|
| 1 | Measurement without production is theater — a Visibility Score dashboard cannot fix content that was never built to be citable |
| 2 | Compliance cannot be retrofitted onto citation-optimized content after the fact; it must be embedded in Build from the first draft |
| 3 | Visibility is engine-specific — a strong ChatGPT citation rate says nothing about Perplexity or Gemini performance |
| 4 | The fastest wins come from correcting existing source-risk gaps, not from commissioning entirely new content |
| 5 | Discover-Build-Measure is a loop, not a project — organizations that treat it as a one-time audit lose their gains within one to two quarters |
Appendix: AI Visibility Readiness & Implementation Reference
- Before launching an AVO program, confirm: (1) a named executive owns Visibility Score as a tracked metric; (2) Medical/Legal/Regulatory has committed standing review capacity for source-tagged content; (3) at least one Knowledge Graph or structured content asset exists to anchor entity/relationship signals; (4) leadership accepts an 8–12 week horizon for first measurable gains and a 3–6 month horizon for commercial ROI.
Maturity Level Quick Reference
| Level | Characteristics |
|---|---|
| Level 0 — Invisible | No AI citation measurement; content unstructured; likely near-zero Visibility Score if tested |
| Level 1 — Aware | Baseline Visibility Score established; source-risk audit complete; Focus registry drafted |
| Level 2 — Structured | Priority Focuses rebuilt as pyramid-chunked, source-tagged Knowledge Artifacts with schema markup |
| Level 3 — Measured | Recurring Visibility Score benchmarking across all major LLMs; leading and outcome KPIs tracked |
| Level 4 — Compounding | Discover-Build-Measure operating as a continuous loop; graph-anchored content reused across Focuses; sustained citation share gains |
Three-Phase Implementation Checklist
- Catalogue real AI-directed prompts across priority category topics
- Audit existing content inventory for structural citability
- Run a source-risk audit against currently-cited third-party and competitor content
- Establish baseline Visibility Score across 4 major LLMs
- Build the initial Focus registry and Prompt Book
- Restructure priority Focuses using the Pyramid Chunking Principle
- Apply per-claim source tagging (DOI/PMID/approved reference)
- Add schema.org/JSON-LD markup to published Knowledge Artifacts
- Embed MLR/compliance metadata (author, reviewer, version, approval date)
- Route all new content through existing MLR review with source-tagged unit review
- Re-run Visibility Score benchmarking against the Prompt Book
- Test RAG recall/precision and monitor for hallucinated claims
- Track leading KPIs (Visibility Score lift, citation share, MLR cycle time)
- Track outcome KPIs (engagement lift, inquiry volume, prescribing/purchase signal shifts)
- Feed findings back into Discover for the next prioritization cycle
- 1. Citation is earned by structure, not claimed by intent — content must be built to be extracted before an AI system can extract it.
- 2. Measurement without governed production is theater — a visibility dashboard cannot make ungoverned content compliant or citable.
- 3. Visibility compounds when it is anchored — Focuses built on a shared entity and evidence architecture reinforce each other; Focuses built in isolation do not.
About This Whitepaper and travalcon.com
travalcon.com is a Project DDIAM LP business initiative developing a consolidated capability architecture for content, data, and AI-readiness in regulated industries. The AI Visibility Optimization Framework™ is one of six interconnected capabilities — alongside the BCB Framework™, Modular Content Framework™, Tagging & Taxonomy Framework™, Knowledge Graph Framework™, and Personalization & Orchestration Framework™ — that together move an organization from fragmented, channel-by-channel content production toward a single governed content architecture built for both human and AI audiences.
This whitepaper synthesizes travalcon's internal AVO methodology together with independently published market research on generative engine optimization, AI search behavior, and citation patterns across major large language models, including direct research into representative AI-visibility monitoring tools such as Sight AI. Figures cited from third-party research are attributed to their original sources; findings specific to the AVO framework and its life sciences case context reflect travalcon's own methodology and illustrative program modeling.