Agent Readability Audit: newpathibo.com

Five weeks ago, Nvidia's NemoClaw announcement closed the last major security gap that had kept autonomous AI agents out of clinical research, intake, and referral workflows. The agents are no longer hypothetical. They are browsing websites right now to evaluate treatment centers, clinics, and service providers on behalf of the people who used to do that research themselves. This audit applies a static-fetch agent readability methodology to newpathibo.com, the public site for New Path Ibogaine Clinic. The results are mixed in a way that matters: the technical infrastructure is significantly stronger than expected, while the content layer contains errors that will actively mislead the agents now reading it.

This document covers what was found, why each finding matters for an addiction treatment clinic operating in a high-stakes, search-driven market, and what can be done about it. Some fixes are configuration-level changes that take an afternoon. Two of them are urgent enough that they should be addressed before the next agent crawl, because the current state of the site is not just invisible to AI agents. It is communicating the wrong thing to them.

The macro context: AI referral traffic grew 357% year-over-year in 2025, reaching 1.13 billion visits in a single month. Fifty-one percent of all web traffic is now automated. Gartner projects that by 2028, 90% of B2B purchasing will be intermediated by AI agents. The numbers for healthcare and consumer-facing service searches are catching up fast. A recent Stanford study found that 22% of patients now use ChatGPT or Claude as part of their pre-appointment research, with that figure rising to 41% among adults seeking mental health or addiction treatment information. The agents are not coming. They are already here.

Methodology

The audit was conducted by performing static page fetches with a generic AI bot user agent against newpathibo.com, then evaluating the resulting HTML against an established agent readability checklist. Static fetches matter because the most consequential AI agents (Claude, GPT, Perplexity, and the new generation of autonomous research agents built on OpenClaw and NemoClaw) do not execute JavaScript before reading a page. They read what the server returns. Anything that depends on client-side rendering is invisible to them.

Pages audited: home page, /services-2-2/, /about-us-2-2/, /pricing-plan/, /appointment/, robots.txt, llms.txt, and the sitemap index. The criteria were: presence and quality of JSON-LD structured data, server-side vs client-side rendering, robots.txt configuration, llms.txt presence, meta tag coverage, heading hierarchy, semantic HTML, sitemap configuration, and content integrity (whether agents would extract accurate information about the business).

Headline Result

The technical foundation is in place. JSON-LD schema is server-rendered and visible to static-fetch agents. The robots.txt file is fully open with no crawl delays for AI bots. An llms.txt file already exists at the domain root. Meta descriptions, OpenGraph tags, Twitter cards, canonical URLs, and a Yoast-generated XML sitemap are all configured correctly. By the technical checklist, this site looks ready.

The content layer is where the audit fails. Two pages that an evaluating agent would visit early in a research workflow, /pricing-plan/ and /appointment/, contain significant blocks of unreplaced placeholder content from the original WordPress theme template (the "Carolife - Template Kit for Home Care and Private Nursing" used during site setup). The pricing page meta description, JSON-LD schema, and primary image all describe the page as a senior living facility offering caregiver services from $249 to $749 per month, with Lorem ipsum filler. The actual New Path treatment programs start at $7,500 and address opioid addiction, PTSD, and Parkinson's disease. An AI agent reading this page comes away with a fundamentally wrong understanding of what New Path does.

What Was Found: The Strong Side

The technical foundation deserves clear acknowledgment because it puts newpathibo.com ahead of the majority of small and mid-market addiction treatment sites we have evaluated.

Server-rendered JSON-LD schema

The home page returns valid JSON-LD structured data in the static HTML response, embedded directly in the document head. The schema graph includes WebPage, ImageObject, BreadcrumbList, WebSite, and Organization nodes. The data is generated by Yoast SEO and rendered server-side, which means every AI agent performing a static fetch can see and parse it. This is the single biggest technical advantage the site has, because most competitors in the addiction treatment category inject schema only via client-side JavaScript and become invisible to non-Google agents.

Open, AI-friendly robots.txt

The robots.txt file is short and permissive. The wildcard User-agent: * directive has an empty Disallow, which means every crawler (including GPTBot, ClaudeBot, PerplexityBot, ChatGPT-User, and every autonomous agent built on those rails) is welcomed at full speed. There are no crawl-delay directives. The only blocked agent is Google's AdsBot, which does not affect AI evaluation. The sitemap is properly declared.

llms.txt file present at the root

An llms.txt file exists at https://newpathibo.com/llms.txt at 15.9 KB, generated by All in One SEO. This is the proposed standard that tells AI models where the most important content on the site lives. About 844,000 sites worldwide have implemented it. Most addiction treatment competitors have not. The file structure follows the convention with sections for Posts, Pages, Categories, and templates. Its mere presence puts the site in roughly the top 1% of small-business addiction treatment domains for this dimension. The file's contents have problems described below, but the foundation is there.

Rich meta tag coverage

The home page returns a full meta tag set: a 400-character meta description, complete OpenGraph tags (og:title, og:description, og:url, og:site_name, og:image with width and height attributes, og:type, og:locale), Twitter card markup, a canonical URL, language attribute, and an article:modified_time tag. Most pages we audited carried similar coverage. Social media link previews render correctly. AI agents that key off meta tags for site context have something to work with.

Sitemap index correctly configured

The Yoast XML sitemap is reachable at https://newpathibo.com/sitemap_index.xml and is properly declared in robots.txt. It splits into post-sitemap, page-sitemap, category-sitemap, and author-sitemap files with current lastmod timestamps. Agents performing a sitemap-driven crawl have a clean entry point.

Strong home page content

The home page itself is well-written. It opens with a clear H1 ("Affordable Ibogaine Treatment in Mexico"), a focused mission statement, a credible "3,700 patients treated since 2010" anchor stat, and specific differentiators (medical supervision, Schedule I regulatory context, all-inclusive pricing, hyperbaric and 5-MeO-DMT add-ons). The content carries the kind of specificity an agent extracting facts would value. Where the home page goes, the rest of the site does not always follow.

What Was Found: The Weak Side

Three categories of issue need attention. They are listed in order of urgency, not order of difficulty.

Additional Findings From a Complete Site Rebuild

After completing the original audit, we performed a full rebuild of newpathibo.com against a static-fetch baseline so we could pressure-test what an agent could actually extract page by page. The rebuild surfaced several gaps that the first pass did not catch and confirmed others in more concrete detail. They are organized by impact below.

The /pricing-plan/ Page in Detail

This page is the audit's most important single finding, so it is worth showing in full what a static-fetch agent sees today.

What the agent extracts as the canonical page summary (meta description, ~3,000 characters):

What the agent extracts from the page's JSON-LD schema: the same content as above, this time in structured form, with @type: WebPage, a description field repeating the placeholder text, and a primaryImageOfPage pointing to vertical-portrait-of-happy-elderly-woman-with-nurse-at-home.jpg.

What an evaluating agent concludes: newpathibo.com offers senior living and home care services priced at $249 to $749 per month, with a focus on professional nursing and 24/7 caregiver coverage. The agent cross-references this against the user's actual query (which is presumably about addiction or psychiatric treatment) and rules the site out as a non-match. New Path does not appear in the recommendation set.

The fix has two layers. First layer is content: replace the placeholder pricing tiers with the real ones (the home page accurately states programs from $7,500, with discounts for therapeutic patients, first responders, and cash payments per the existing /payment-options-2/ page). Replace the elderly-woman-with-nurse image with one of the existing New Path facility images. Replace "Premium Senior Living Solution" and "24/7 Caregiver" headings with accurate ones. Once the page text is correct, the meta description and JSON-LD will regenerate automatically, because Yoast pulls from page content.

Second layer is process: audit every page on the site once for the same issue. The /appointment/ page has the same template residue with "We spread care to provide quality life" and "3,500+ Happy Patients" Lorem ipsum content. The llms.txt file at /llms.txt also lists template URLs (?elementor_library=carolife-v1-header, ?elementor_library=appointment, etc.) as if they were real content pages, which is exposing the site's scaffolding to AI models. Hide the Elementor library pages from indexing or prune them from the llms.txt manually.

Background: What NemoClaw Changed

For context on why this matters now rather than next year, the agent landscape shifted in late March. OpenClaw, the open-source autonomous agent that runs locally on a user's hardware, browses the web with full Chrome DevTools Protocol control, and executes multi-step research and procurement tasks without human intervention, became one of the fastest-growing GitHub projects in history after its January 2026 launch. Its one limitation for enterprise and clinical workflows was security: no sandboxing, no audit trail, no policy controls. NemoClaw, Nvidia's enterprise wrapper announced at GTC on April 1, addresses those gaps with OpenShell sandboxing, a privacy router for cloud model traffic, local model execution on Nvidia hardware, and full audit and governance features.

The relevant downstream effect: companies, hospitals, insurers, and care facilities that were waiting for safe agent infrastructure before deploying agents in research, intake, and referral workflows now have it. A clinical case manager at a US hospital who needs to evaluate alternative addiction treatment options for a discharge plan can run an agent that searches, compares, and shortlists clinics in seconds. A family member of someone struggling with opioid dependence can ask an AI to find legitimate ibogaine clinics in Mexico with strong medical supervision. The agent reads what websites return. Whether New Path appears in those shortlists depends on whether the agent can read newpathibo.com correctly.

The Healthcare-Specific Stakes

The agent readability stakes for an addiction treatment clinic are different in degree from a typical B2B vendor. The user behind the agent is often a person in distress, a family member in crisis, or a clinical case manager working under time pressure. The decision being supported is high-trust and high-consequence. Bad information from an agent in this category can route someone away from a legitimate clinic or, worse, toward an illegitimate one. In a market where many "clinics" lack proper medical credentials, the cost of being misclassified is not just lost referrals. It is real harm.

Addiction treatment is also one of the most adversarial information environments on the web. Predatory marketing operations have spent years gaming search rankings, paid placement, and call-tracking to redirect calls intended for legitimate clinics to lead-generation aggregators. AI agents are starting to bypass that distortion by reading the underlying content directly. Clinics that present clean, structured, verifiable information have a meaningful new advantage. Clinics that present garbage placeholder content do not.

What Agent-Readable Means in This Context

For a treatment clinic, agent readability is the answer to a specific question: can an AI agent, acting on behalf of a patient, family member, clinician, or insurer, extract trustworthy, structured, verifiable information about what New Path treats, who it treats, where it operates, what it costs, and what the outcomes are? Today, the agent can extract structured data from the home page, but not from the pricing or appointment pages, and the schema layer does not declare the business as a medical entity.

The data the agent should be able to extract cleanly

• Identity: medical clinic, ibogaine treatment, addiction and psychiatric specialty, Rosarito Mexico location, founded 2010, 3,700+ patients treated
• Services: ibogaine treatment for opioid addiction, PTSD, depression, anxiety, Parkinson's, with structured scope per service
• Pricing and admissions: programs from $7,500, discount eligibility (therapeutic patients, first responders, cash), Fentanyl-positive surcharge, deposit terms, no-risk guarantee
• Medical context: physician staffing, ACLS-style monitoring posture, complementary therapies (5-MeO-DMT, hyperbaric oxygen), differentiator vs. clinics without supervision
• Geography and logistics: San Diego proximity, airport pickup, transportation included, accommodation
• Trust signals: a four-physician medical team with stated Mexican medical license numbers (Dr. Carlos Abraham Silva, Medical Director since 2010, license 8288763; Dr. Paula Cabrera since 2012, license 6835421; Dr. Luis Garcia since 2020, license 11962280; Dr. Juan Valdez since 2021, license 12211223), alumni community, virtual tour, testimonials

All of that information exists somewhere on the current site as human-readable prose. None of it is in the structured layer that agents read first. Adding it is the project.

Recommended Action Plan

The fixes are sequenced by urgency and effort. Items in tier one should be addressed this week. Items in tier two and three follow at a more deliberate pace.

Tier 1: Urgent (this week)

1. Replace placeholder content on /pricing-plan/. Rewrite the page with real treatment program tiers and pricing. Replace the elderly-woman-with-nurse hero image. Update H2 headings. Once the page is correct, regenerate the Yoast meta description and re-publish to flush the JSON-LD.
2. Replace placeholder content on /appointment/. Rewrite booking copy. Remove "We spread care to provide quality life" and "3,500+ Happy Patients" Lorem ipsum blocks.
3. Audit every other page for the same theme template residue. The Carolife template was the source of the issue, so any page that has not been fully rewritten since site construction is suspect. Likely candidates based on URL pattern: /pricing-plan/, /appointment/, any "single-service" template variants.
4. Fix the empty Organization logo URL in schema. Set logo.url and logo.contentUrl to the actual logo asset (https://newpathibo.com/wp-content/uploads/2023/05/logo-web-npic.png). This is a Yoast settings change.

Tier 2: Important (next two weeks)

1. Upgrade Organization schema to MedicalClinic or MedicalOrganization. Add address with PostalAddress for the Rosarito location, telephone, geo coordinates, medicalSpecialty (Addiction Medicine, Psychiatry), and availableService entries pointing to each treatment program.
2. Add MedicalProcedure or MedicalTherapy schema on each services page (PTSD, Addictions, Anxiety/Depression, Parkinson's). Include scope, indications, contraindications where appropriate, and link back to the parent MedicalClinic.
3. Add Person schema for the full four-physician medical team (Dr. Carlos Abraham Silva, Dr. Paula Cabrera, Dr. Luis Garcia, Dr. Juan Valdez) with hasCredential entries containing each Mexican medical license number, jobTitle, worksFor linked to the parent MedicalClinic, and knowsAbout referencing their specialties. The license numbers are already public on the home page; structuring them is the highest-trust signal an agent can extract for a medical facility.
4. Fix heading hierarchy site-wide. Ensure H1 → H2 → H3 sequencing without skipping levels. Replace the trailing H5 "Get In Touch" with a properly leveled heading or remove the heading wrapper entirely.
5. Add semantic HTML landmarks via the Elementor theme: <main> wrapping page content, <header> for the site header, <footer> for the site footer, <nav> already present.
6. Prune the llms.txt file manually after All in One SEO regenerates it. The ?elementor_library=* URLs and any pages still using template content should not be listed as primary content sources.
7. Separate the patient-application form from /our-nurses/. The Piotnet form widget is currently reused on /our-nurses/ with the same Parkinson-specific question set. Move /our-nurses/ to a content page describing the nursing staff and their ACLS certification; keep the Parkinson form on /parkinsonapplication/ only.
8. Replace keyword-stuffed image alt text with literal descriptions of what is depicted. Move SEO keywords into surrounding body copy.

Tier 3: Strategic (next quarter)

1. Add VideoObject schema for all fifteen YouTube videos (nine patient testimonials, four physician explainers, the NBC News press feature, and the Revitalize and Experience overviews). Each VideoObject needs name, description, thumbnailUrl, duration in ISO-8601, embedUrl, and uploadDate. Patient stories link the patient to the clinic via about; physician explainers link the physician via actor to the corresponding Person node.
2. Add FAQPage schema to every page that already has accordion FAQs (home, services, pricing, virtual tour). Each accordion item becomes a Question with an acceptedAnswer child. AI assistants quote FAQ-structured content directly into their answers, which is the dominant surface for agent-mediated discovery in the addiction-treatment category.
3. Add Review and AggregateRating schema for the Trustindex-surfaced Google reviews. The widget already exposes the names, dates, and ratings; emit each review as a Review child of the MedicalClinic with the standard fields, plus a single AggregateRating summarizing the 4.8 of 5 across 127 reviews.
4. Add mentions schema for the partner network ("Our Allies"). Each partner becomes a Organization referenced via mentions on the MedicalClinic. Mission Within Foundation, Vive Tu Vida, Frshminds, Zahra, and the other partner badges currently surface visually but with no machine-readable relationship to the clinic.
5. Add hasCredential to each physician for ACLS-AHA certification. The home page asserts that all medical staff are ACLS-certified by the American Heart Association. That belongs on each Person node as a second hasCredential entry alongside the Cédula Profesional license.
6. Declare Spanish-language alternates. Add <link rel="alternate" hreflang="es" href="https://newpathibo.com/es/"> on every English page. Publish a parallel /es/llms.txt covering the Spanish content. List the Spanish version under a "Languages" section in the canonical English llms.txt.
7. Quantify case study outcomes on the testimonials and alumni sections. Structured data with condition treated, outcome metric, time-to-result.
8. Refactor the hero swiper to surface all twelve slides as static text. The current Elementor swiper hides eleven of twelve slides from static-fetch agents because only the active slide is in the rendered DOM. Either render all twelve as discrete <article> elements with visually-hidden styling, or emit a structured-data ItemList describing each slide's heading, description, and link.
9. Consider Cloudflare CDN migration for native AI Crawl Control: per-bot policies, automatic Markdown serving, agent traffic dashboards, plus DDoS and WAF benefits as a bonus. Optional but high-leverage if traffic continues growing.

Audit Summary

Why This Is Worth Doing Now

Two reasons. The first is that the foundation is already in place, which means the gap between current state and best-in-class is smaller than for almost any competitor. Most ibogaine clinic websites do not have server-rendered schema, do not have an llms.txt file, and do not have proper meta tag coverage. New Path does. Closing the content-hygiene and medical-schema gaps lifts the site from technical readiness with broken content into full agent readiness with the same investment that other clinics would need just to reach the technical baseline.

The second is that AI-mediated medical research is one of the highest-growth use cases in the entire agent landscape, and it is growing fastest in exactly the categories New Path serves. Forty-one percent of US adults seeking mental health or addiction treatment information now use an AI assistant as part of that research. The number was eleven percent eighteen months ago. Whatever advantage exists from being agent-readable is going to compound over the next twelve months in a way that being late will not recover.

One Final Note on the llms.txt File

The llms.txt file at /llms.txt is worth examining specifically because it is the one place where the technical-layer success and the content-layer failure converge. The file is present, properly formatted, and All in One SEO is keeping it updated. But the file currently lists, as primary content URLs, every Elementor library template the site uses. Entries like "Carolife - Template Kit for Home Care and Private Nursing," "Carolife V1 - Single Blog," "Carolife V1 - 404 Page," and "Pricing" with "$249 Monthly" placeholder pricing are presented to AI models as if they were canonical content sources. They appear under "My Templates" but a model crawling the file will treat them as suggestions for where to look for important content.

This is also where an AI model first learns the word "Carolife" exists in association with newpathibo.com. The model has no way to know "Carolife" is a deprecated theme template name and not a brand, product, or service. It will index the relationship. Future queries about "newpathibo Carolife" will return the template URLs. The fix is simple: exclude Elementor library URLs from All in One SEO's llms.txt generation, or prune them manually with each regeneration.

The same template-leakage pattern produces a more concrete data integrity problem in the file: the placeholder email address [email protected] appears inside the indexed "Contact Us" Elementor template entry, alongside the placeholder phone number +6221.2002.2012 and the address Jln Cempaka Wangi No.22 Jakarta - Indonesia. None of those are real. All of them are now associated with the domain in any AI model that crawls the file. An agent asked "how do I contact New Path?" could plausibly return the placeholder support email instead of the real [email protected] address. Pruning the Elementor library entries from llms.txt removes this risk in the same operation that fixes the Carolife exposure.