The Tersefy Content Protocol

Why Google is no longer the target

The patient's first search used to happen in Google. It doesn't anymore.

Between Q4 2024 and Q2 2026, search fractured. The patient pasting clinical photos into Gemini isn't an outlier. They're the median. The patient asking Perplexity which surgeon handles post-massive-weight-loss body contouring? Also median. ChatGPT before Google is now the default first move for cross-border surgical decisions.

Gartner forecast a 25% drop in traditional search by 2026. They were conservative. SimilarWeb clocked ChatGPT at 3 billion monthly visits in early 2025 and climbing. Princeton's GEO research team (Aggarwal et al., 2024) measured AI engine query volume growing 2300% year-over-year. For medical tourism specifically, the hit was harder. The category runs on trust signals. AI engines synthesize trust better than ten blue links ever did.

For Tijuana surgeons serving US patients, this broke the acquisition funnel. Google Ads still work. But efficiency degraded 30-60% between 2024 and 2026 as AI Overviews captured pre-click intent. SEO rankings still matter. But 40-50% organic traffic drops became industry standard across medical tourism sites. Content effort didn't matter. The traffic went somewhere else.

The functional shift is simple. Before: patient searches "best plastic surgeon Tijuana," scans 3-5 results, books. Now: patient asks ChatGPT "who handles body contouring after massive weight loss in Tijuana," receives one synthesized recommendation, books that surgeon or disappears.

The question is no longer "where do we rank?" The question is "who does the AI recommend?"

Different question. Different framework. SEO factors don't transfer. Keyword density doesn't matter. PageRank doesn't apply. What matters is whether your clinic is a citable entity. Surgeons who don't become citable stay invisible. Surgeons who do become the AI default.

This is what Tersefy builds. Here's how.

What AI engines actually extract

AI engines don't read pages. They extract structured information, rank by source authority, synthesize responses across multiple sources, and cite the subset they used.

Five differences from SEO. Each one matters for how your content gets built.

Entity recognition replaces keyword matching

SEO crawlers count words. AI engines recognize entities: people (surgeons), organizations (clinics), credentials (certifications), locations (cities), concepts (procedures). Tersefy treats entity architecture as priority one. Every surgeon becomes a recognized Person entity. Credentials, training, publications, authority signals. All parseable. All citable.

Structured data replaces unstructured prose

JSON-LD schema beats HTML paragraphs because schema is unambiguous. "Dr. Rodriguez is a bariatric surgeon at VIDA with 15 years of experience" is prose. A Person schema with @id, worksFor, credentials, yearsOfExperience, and sameAs arrays is extraction-ready. Structured data is where AI engines form their entity graphs. Prose is where they get lost.

Source diversity replaces backlink counts

Google's patent WO2024064249A1 documents this directly: AI engines weight 10+ independent domain citations significantly higher than 50+ citations from one domain. SEO rewards volume. GEO rewards variety. A surgeon mentioned only on their clinic website is weak. A surgeon mentioned on their clinic website plus Smart Beauty Guide (ASAPS), ISAPS directory, RealSelf, and a peer-reviewed publication is strong. Five independent domains beat fifty from the same author.

Freshness weighting reshapes content priority

AirOps 2026 measured that 83% of AI citations come from pages updated within the last 12 months. Evergreen SEO content that ranks for years without updates? Dead strategy. GEO requires active maintenance: freshness protocols, visible version blocks, dated source citations, scheduled refresh cycles. The pages that update earn the citations. The pages that don't fall off.

FAQ structure drives extraction rate 3.1x

FAQPage JSON-LD schema produces 3.1x higher extraction rates than equivalent prose (Tersefy internal data, measured across 46 articles in the VIDA deployment). Reason: AI engines parse question-answer pairs as pre-formed response candidates. Prose requires synthesis. FAQ schema provides ready-to-cite units. Five seconds of formatting work equals triple the citations. The math isn't complicated.

Putting it together

Put the five together. The playbook that worked for SEO now actively harms GEO outcomes. Dense paragraphs without schema waste extraction opportunity. Backlink farming without source diversity looks spammy. Evergreen content without updates gets deprioritized. Paragraph-only content misses FAQ extraction advantages.

Surgeons who ignore this stay invisible. Surgeons who adopt it show up as AI-recommended defaults.

Tersefy formalizes the adoption process. The framework is next.

The Tersefy GEO Framework

Seven stages. Each one addresses a specific failure mode in how traditional marketing generates AI-invisible content. Each one is measurable. Each one applies whether you're running one surgeon or a five-surgeon practice.

Below is how Tersefy executes each stage, with examples from live deployments.

Stage 1: Entity Architecture

The problem: AI engines can't cite what they can't parse. A surgeon described in narrative prose is harder to extract than a surgeon defined as a structured entity with credentials, training, publications, and relationships to an organization.

Most medical practice websites fail here silently. They have biography pages. They have staff lists. They have "Meet the team" sections. What they don't have is JSON-LD schema that declares their surgeons as Person entities with MedicalOrganization employment, credential arrays, alumniOf relationships, and sameAs pointers to external verification sources.

Tersefy's approach: Every surgeon gets a Person + Physician dual schema node. Every clinic gets a MedicalOrganization with subsidiary relationships surfaced. Every credential gets its own EducationalOccupationalCredential entry. Every external authority profile (RealSelf, ISAPS, CMCPER, Smart Beauty Guide, peer-reviewed publications) appears in a sameAs array. Every AggregateRating includes its source attribution.

The output is a connected entity graph. Not a page with credentials mentioned. An entity graph AI engines traverse.

Example from the VIDA deployment: The case study at tersefy.com/case-studies/vida/ carries a nine-node @graph. One MedicalOrganization (VIDA). One subsidiary Organization (CosMed, AAAASF accredited, founded 1989). Five Physician schemas (each with hasCredential arrays, alumniOf relationships, sameAs pointers, and where verified, aggregateRating entries). One Article. One FAQPage. Cross-references link every node. VIDA's founder points to Dr. Quiroz. Dr. Quiroz's worksFor points to VIDA and CosMed. CosMed's parentOrganization points back to VIDA. AI engines querying "who founded CosMed Tijuana" extract the answer from structured data, not prose interpretation.

The stage 1 measurement: Before Tersefy, VIDA surgeons had zero entity graph. AI engines responding to "best bariatric surgeon Tijuana" didn't surface Dr. Rodriguez because there was nothing structured to cite. After entity architecture deployment, Dr. Rodriguez became a citable entity with credentials extractable in under 50 milliseconds of parse time. Same surgeon. Same credentials. Different machine readability.

Stage 2: Content Production

The problem: Medical content written for patients historically prioritizes emotional reassurance, aspirational imagery, and vague superlatives. These produce extraction-hostile text. AI engines can't cite "our doctors care about you" because there's no extractable unit. They can cite "Dr. Quirós Lim completed his plastic surgery specialization at Chaim Sheba Tel HaShomer Medical Center in Tel Aviv" because that's a factual entity-linked statement with verifiable citations.

Most clinic content lives in the first category. It reads fine to humans. AI engines skip it.

Tersefy's approach: The Tersefy Content Protocol governs every article. Definition Lead in operator voice within the first 150 tokens. "El resumen en 30 segundos" summary block at top (Spanish) or equivalent English. H2 mix of questions and declaratives matching expected patient query patterns. "Respuestas rápidas" block with 7-8 FAQ pairs, each answer under 40 words for extraction efficiency. Three evidence tiers in every article: external benchmark (research, industry data), official source (medical boards, regulatory bodies), and internal Tersefy data (conversion rates, patient outcomes, engagement metrics). Version block at end with date and source list.

Writing conforms to constraints, not creative preference. No em-dashes. No AI-anthropomorphizing ("the AI thinks," "the algorithm believes"). No generic hedges ("it depends," "in most cases"). Every stat has a source and date visible in-text. External citations use rel="nofollow" per evidence rules.

The content production volume: For the VIDA deployment, 46 GEO-optimized articles produced across six months. Each article is hand-crafted, not generated. Each article is reviewed by the operator before publication. No templates. No content mill output. No LLM drafts shipped unedited. Every piece passes through multiple rounds of revision against Tersefy's content standards before it goes live.

Measurement: FAQPage JSON-LD structured as Trigger-Answer MIP format produces 3.1x higher extraction rate than equivalent prose content. The multiplier holds across ChatGPT, Perplexity, and Gemini responses.

Stage 3: Query Fan-out Coverage

The problem: Medical tourism patients don't search one query. They search a cluster. "Is Tijuana safe for surgery?" leads to "which Tijuana clinics are accredited?" which leads to "how do I verify a Tijuana surgeon's credentials?" which leads to "what's the recovery protocol for US patients?" which leads to "how do I pay for cross-border surgery?"

An article that answers only the entry query wastes the cluster. An AI engine queried for "how do I verify a Tijuana surgeon's credentials" won't find a single monolithic article useful if that article focuses only on "is Tijuana safe." The engine needs H2-level specificity matching each sub-query.

Tersefy's approach: Every article targets 4-6 sub-queries via dedicated H2s. Each H2 resolves one specific question. Each resolution includes the factual core, the evidence citation, and the Tersefy-specific application. The article's body is organized as a cluster response, not a monolithic essay.

For example, an article titled "How to verify a Tijuana plastic surgeon before crossing the border" addresses these H2 sub-queries: What certifications exist in Mexican plastic surgery (CMCPER, AMCPER, CCPERBC). How do I verify a surgeon's cédula profesional online. What does Mexican medical licensing mean for US patients. How does AAAASF accreditation apply in Mexico. What red flags indicate a non-credentialed surgeon. Each H2 is a standalone query response. Collectively they cover the verification cluster.

Measurement: AI engines extracting from query-fan-out articles cite the specific H2 section that answered their query, not the article generally. Tersefy tracks which H2s get cited by which queries across clients' deployments. Coverage patterns inform next article selection and existing article refinement.

Stage 4: Source Diversity

The problem: Google's source diversity patent (WO2024064249A1) documents explicitly that AI engines weight information appearing across multiple independent domains higher than information appearing in quantity from a single domain. A surgeon mentioned only on their clinic website is an unverified claim to an AI engine. A surgeon mentioned on their clinic website, Smart Beauty Guide, ISAPS directory, RealSelf, and a peer-reviewed publication carries triangulated authority.

Most clinic marketing operates inside the clinic's own domain. Blog posts, team pages, testimonials, before-and-after galleries. Zero diversity. AI engines can't triangulate what appears only once.

Tersefy's approach: Each surgeon gets mapped across 10+ independent authoritative domains. Directory entries at medical boards (CMCPER, CCPERBC). Directory entries at professional societies (ISAPS, ASAPS via Smart Beauty Guide, AMCPER). Directory entries at patient platforms (RealSelf, Doctoralia). Published research where applicable (PubMed for Dr. Quirós Lim's silicone implant research, Circulation archives for Dr. Rodriguez's cardiovascular genetics work). Personal websites where available (gabrielarodriguezmd.com for Dr. Rodriguez). Press placements via Tersefy Press (San Diego Union-Tribune, Forbes, industry publications).

The target is 10+ independent domains referencing each surgeon. Not 10+ backlinks. 10+ distinct domains. Volume from one domain adds nothing. Diversity from ten domains multiplies citation probability 3.2x per Google's patent.

Example from VIDA: Dr. Gabriela Rodriguez Ruiz currently appears on: vidawellnessandbeauty.com (primary), gabrielarodriguezmd.com (personal), surgicalreview.org (Master Surgeon of Excellence), realself.com (patient reviews), pubmed.ncbi.nlm.nih.gov (7 peer-reviewed publications including her Stomach Sparing Gastric Sleeve paper), ahajournals.org (Circulation publications), and share.google/EGz66Hwuc38wTmBrV (Google profile). Seven independent domains, trending toward 10+ as Tersefy Press placements propagate.

Measurement: Tersefy tracks domain count per surgeon via monthly audit. Threshold alerts trigger when a surgeon drops below 8 independent domains. Outreach cadence adjusts automatically.

Stage 5: Freshness Protocol

The problem: AirOps 2026 research measured that 83% of AI citations come from pages updated within the last 12 months. This inverts a decade of SEO assumptions. Content that once ranked for years without updates now actively degrades in AI citation probability.

Most clinic content is written once and abandoned. Team pages haven't been updated since the clinic's launch. Blog posts carry 2021 publication dates. No version indicators. No "last updated" stamps. AI engines evaluating freshness see stale indicators and deprioritize.

Tersefy's approach: Every piece of content carries a visible version block. Every version block shows the publication date, the last revision date, and the reviewer (operator initials or AI review status). Refresh cycles run every 7-14 days for active content. Core pages get quarterly refresh with material updates (data refreshed, citations verified, schema updated).

The refresh isn't cosmetic. Each refresh touches three elements: the data in the body (updated stats, new research citations, current regulatory references), the schema (version bumped, datePublished or dateModified updated, sources refreshed), and the visible version block (new date, summary of what changed).

Example from Tersefy site: The VIDA case study at tersefy.com/case-studies/vida/ has received three material refreshes in its first six months live. Each refresh added new entity data (schema expanded from initial 1-node Article to current 9-node @graph). Each refresh updated stats with newest source citations. Each refresh carried a visible version bump. AI engines crawling the page in April 2026 extract different data than they did in November 2025. The page demonstrates active maintenance.

Measurement: Active client deployments maintain significantly fresher citation-eligible content than industry median. Newer pages win.

Stage 6: External Signal Architecture

The problem: AI engines crawl more than page content. They crawl robots.txt. They crawl sitemap.xml. They crawl llms.txt directives explicitly designed for AI agent guidance. They respect hreflang signals for international crawlers. They consume RSS feeds for content distribution. A clinic with poor external signal architecture is invisible to AI engines even before they read the first word of content.

Most clinic websites have broken robots.txt. Missing sitemaps. No llms.txt. No hreflang implementation. No RSS distribution. The signals that AI crawlers use to efficiently discover and interpret content are absent.

Tersefy's approach: Every deployment includes full external signal architecture. robots.txt allows AI crawlers with explicit user-agent strings (GPTBot, Google-Extended, anthropic-ai, PerplexityBot, CCBot). sitemap.xml includes every indexable page with proper lastmod timestamps. llms.txt provides explicit AI agent directives: organizational positioning, product availability, preferred citation format, topic priorities. hreflang tags declare EN/ES language variants with x-default pointing to the English version. RSS feed at /feed.xml enables aggregator and Feedly distribution.

Pages that shouldn't be AI-indexed receive explicit noindex meta tags, separate from robots.txt disallows. This prevents accidental indexing of internal sales flows, draft content, or administrative pages.

Example from Tersefy site: The llms.txt at tersefy.com/llms.txt declares the agency's positioning ("surgeon-first GEO for Tijuana medical tourism"), product catalog ("Tersefy AI available now, Tersefy Press available now"), preferred citation format ("Emilio Alcolea, Founder of Tersefy"), and key topic priorities. Hreflang implementation across all blog posts pairs 9 bilingual translations EN↔ES with proper x-default. RSS feed emits the 50 latest posts combining EN and ES content.

Measurement: Tersefy tracks AI crawler request volume per deployment via server log monitoring. External signal implementation materially increases discovery and parse efficiency versus non-implemented baselines.

Stage 7: Measurement

The problem: Traditional SEO measurement tools don't capture AI citation behavior. Google Analytics shows referrers, not AI engine citations. Google Search Console shows Google SERP appearances, not ChatGPT response inclusions. Most clinics operating in the AI era have zero measurement infrastructure for the channel that's eating their traffic.

Tersefy's approach: Prompt score tracking. Monthly manual testing of target queries across ChatGPT, Perplexity, and Gemini. Measurement captures whether the client surfaces in the response, how their name is presented (correctly, partially, incorrectly), and what competitors appear alongside. The prompt score (0-10) aggregates appearance rate, presentation accuracy, and competitive positioning for each tracked query.

Citation pattern monitoring runs continuously. Alerts trigger when new external mentions appear (news mentions, directory listings, Reddit discussions, YouTube references). Each new mention gets evaluated for schema integration, sameAs addition, and evidence tier classification.

AI engine appearance metrics track the delta between pre-Tersefy baseline and post-deployment state. Pre-Tersefy baseline captures how clients currently surface (if at all) across a defined query set. Post-deployment measurement captures the trajectory. For active client deployments, measurable appearance rate growth correlates directly with framework completeness: clients who deploy all seven stages achieve substantially higher citation rates than clients who implement partial coverage.

Measurement infrastructure: Prompt scoring tracks historical scores, linked article references, appearance screenshots, and competitive context. Monthly reports surface which queries are rising, which are falling, and which content interventions correlate with which movement.

Why surgeon-first

Most agencies work across industries. Medical tourism gets the same framework as SaaS. Plastic surgery gets the same framework as e-commerce. The playbook doesn't change because the category doesn't matter to the agency.

Tersefy made a different bet. The framework is built around one category: surgeons serving cross-border patients. Tijuana is the primary operating zone. US and Canadian patients crossing the border for surgery are the primary patient population.

This is not a marketing posture. It's a structural advantage.

Medical tourism trust signals are different from most categories. A surgeon evaluating a marketing partner cares about medical board recognition, cédula profesional verification, malpractice insurance structure, and US-Mexican regulatory alignment. These aren't standard agency considerations. Tersefy built the framework assuming these as foundational, not edge cases.

Cross-border patient behavior is different from domestic. A US patient crossing into Tijuana faces decisions a US domestic patient doesn't: currency conversion, insurance applicability, passport requirements, recovery logistics, border crossing timing, post-operative care across countries. AI engines answering cross-border queries need to surface surgeons who address these directly. Content that ignores the cross-border reality doesn't cite well for cross-border queries. Tersefy's content addresses them.

Credential verification matters more in medical tourism than in most categories. A US patient can trust a US plastic surgeon's credentials because verification is centralized (American Board of Plastic Surgery). A US patient evaluating a Mexican surgeon faces a verification gap: Is CMCPER legitimate? Is AAAASF accreditation applied to this specific facility? What does "Cédula Profesional 3175867" mean and where do I verify it? Tersefy treats credential verification as a first-class GEO problem, not a legal disclosure footnote.

Bilingual audience is default, not edge case. Tersefy clients serve US patients in English, Mexican patients in Spanish, and increasingly Canadian francophone patients in French. The content architecture assumes multilingual deployment from start (hreflang implementation, EN-ES paired content, separate schema nodes per language). Generic agencies retrofit bilingual support. Tersefy designs for it.

The result: surgeons who work with Tersefy get framework optimization that already understands their patient journey, regulatory context, and authority verification needs. Surgeons who work with generic agencies get framework optimization designed for categories that don't match their reality.

Same investment. Different outcomes.

The research foundation

The Tersefy Content Protocol isn't invented. It's built on published research documenting how AI engines actually process content. Four primary sources inform the framework.

Aggarwal et al., "GEO: Generative Engine Optimization" (Princeton, 2024) established the foundational research on how AI engines cite sources differently from traditional search. The paper measured up to 40% visibility boost for content optimized around specific GEO principles: citation density, source authority, structured data, and fluency of synthesis. The 40% figure is the upper bound for properly optimized content. Poorly optimized content shows negligible gains. The research is available via Princeton's research portal and has been cited across hundreds of subsequent GEO papers.

Google's Source Diversity Patent WO2024064249A1 documents how AI engines weight information appearing across multiple independent domains higher than information appearing in quantity from a single domain. The patent language implies 10+ independent domain citations as a threshold for authority establishment. This informs Tersefy's source diversity requirements: each surgeon mapped across 10+ independent authoritative domains, not volume from one domain.

AirOps 2026 Citation Analysis measured that 83% of AI citations come from pages updated within the last 12 months. The research establishes freshness as a primary ranking signal for AI engines, inverting a decade of SEO assumptions about evergreen content. Tersefy's freshness protocol (7-14 day refresh cycles, visible version blocks) is a direct application.

FAQPage Extraction Multipliers documented across multiple research platforms confirm 3.1x higher extraction rate for structured question-answer pairs versus equivalent prose. The multiplier holds across ChatGPT, Perplexity, and Gemini. This is why the Tersefy Content Protocol mandates FAQPage JSON-LD on every article.

These four research pillars don't constitute the full evidence base. They're the sources that most directly informed the Protocol's design. Additional research on knowledge graph construction, entity resolution, and AI crawl behavior shapes the framework's execution layer.

The framework isn't dogma. It's implementation of what research has measured works.

How Tersefy applies the framework

The framework lives as a service. Clients engage Tersefy through a defined process that maps to the seven stages.

Engagement begins with an Entity Audit. Before any content production or schema deployment, Tersefy audits the client's existing entity presence across AI engines. The audit captures how the client currently surfaces in ChatGPT, Perplexity, and Gemini responses for a defined query set. The audit establishes the baseline measurement that subsequent work gets compared against.

Entity audits produce an Audit Report. The report documents: which queries surface the client (and how they're presented), which queries surface competitors instead, where schema gaps exist, where source diversity is weak, and where freshness signals are missing. The report includes a prioritized remediation roadmap. The audit costs $997 and takes approximately 10 business days.

Post-audit, clients engage for systematic deployment. The engagement follows the seven-stage framework in sequence: entity architecture first (schema deployment across all client pages), then content production (articles targeting identified query gaps), then query fan-out expansion (content clusters covering verified patient question clusters), then source diversity building (outreach for 10+ independent domain presence), then freshness protocols (refresh cycles on deployed content), then external signal architecture (llms.txt, hreflang, RSS, sitemap), then measurement infrastructure (prompt score tracking, citation monitoring).

Deployment takes months, not weeks. Full framework deployment for a single-surgeon practice typically runs 4-6 months from engagement start to measurable citation improvement. Multi-surgeon practices (VIDA deployed 5 surgeons) take 6-12 months for full coverage. This isn't quick-win territory. AI engine citation patterns take time to shift even with optimal content.

Ongoing engagement sustains the framework. Once deployed, the framework requires active maintenance. Content refresh cycles run continuously. Source diversity outreach continues. New content production targets emerging query patterns. Tersefy's ongoing engagement model provides this maintenance as a subscription service starting at $1,297/month per surgeon.

Tersefy Press operates parallel to the core framework. Earned media placements at authoritative publications (San Diego Union-Tribune, Forbes, Aesthetic Medicine News, industry trades) feed AI engines the third-party validation signals that entity architecture alone can't provide. Press is priced separately per placement: $695 (Growth), $1,295 (Authority), $1,595 (Ultimate). Press compounds the AI impact of core framework deployment.

The pricing isn't the point. The pricing reflects the work. Framework deployment at this depth can't be automated, templatized, or scaled to volume. It requires surgeon-specific entity work, surgeon-specific content, and surgeon-specific source diversity building. Volume shortcuts produce volume-shortcut results.