PublishedVerifiedEvery 6 weeksSources8 namedAuthored bySquareRank Team
Nutritionists · AI Citations · § 4.5.1 · WEDGE
Dietitian AI Citations on Health Queries
AI engines answering YMYL health queries1 need credentialed sources. The first-tier slot (Stanford, NIH, Harvard, Mayo) handles condition reviews but cannot answer "find an RD in Austin who specialises in PCOS and takes Aetna". The natural second-tier slot is credentialed practitioners — and Registered Dietitians, credentialed through the Commission on Dietetic Registration2, fit it precisely. Almost no RD site in 2026 is built to be cited from.
This is the playbook for closing that gap. Each section explains one lever — the tier-one source problem AI engines have, the actual citation patterns ChatGPT and Perplexity show, the credential-signal machinery (schema, byline, source manifest), the content shape that gets extracted, the crawler-access audit that must precede everything, and the measurement loop that confirms the work landed. The wedge is temporary; the practitioner sites that claim citation positions in the next 12-18 months will hold them.
§01The tiers
The tier-one source problem AI engines actually have
AI engines answering health queries face a structural ranking problem. The first-tier sources they can cite without controversy — Stanford Medicine, NIH, Harvard T.H. Chan School of Public Health, the Mayo Clinic, the CDC — publish excellent condition reviews and broad guidance, but they do not publish the locally-grounded, intent-shaped, individual-practitioner content the queries actually demand. 'Find a dietitian for PCOS in Austin who takes Aetna' is unanswerable from the first tier. The second tier the engines pull from is credentialed practitioners with citable site content, and Registered Dietitians — protected credential, national registry, named modalities — fit the slot for nutrition queries the way licensed pharmacists fit it for medication interaction queries.
The mechanism is straightforward once you see it. Google's published Search Quality Rater Guidelines1 tell raters to apply a higher E-E-A-T bar on YMYL health topics; the rater scores feed the machine-learning systems that shape ranking and AI Overviews citation. AI engines downstream of Google's index (and AI engines with their own ranking that mirror the underlying E-E-A-T patterns) propagate the same preference: credentialed authors with named expertise get cited over uncredentialed wellness bloggers, even when the underlying content is comparable in word count and surface quality. The trust signal is the wrapping, not the writing — and credentialed RDs already have the credential half of the wrapping that wellness sites cannot synthesize.
The second-tier opportunity becomes visible when you watch the actual citation patterns. Ask ChatGPT or Perplexity "best dietitian for IBS in [mid-sized U.S. city]" and the engines pull from a small set of source types: the Academy of Nutrition and Dietetics' Find a Nutrition Expert directory3, a few nutrition-specific directories (e.g. Find My Dietitian, Healthie), and individual RD practice sites when those sites carry credentialed bylines, named modalities, and schema that flags the practitioner as a Registered Dietitian. The slot that is genuinely contestable is the third category — the practitioner sites — and the practitioners holding it are the ones whose sites carry the credential signal in machine-readable form.
Why the second tier is the practical wedge
YMYL
Google's quality-rater frame for health and nutrition content — explicit higher E-E-A-T bar than non-YMYL.
What ChatGPT and Perplexity actually cite on dietitian queries
The empirical pattern across audits is consistent. On condition-led queries ('PCOS nutrition guide', 'IBS low-FODMAP overview') AI engines cite the first-tier health-authority sources almost exclusively. On practitioner-led queries ('dietitian for PCOS in Austin', 'RD who takes Aetna and does telehealth in Texas') the engines cite the Academy of Nutrition and Dietetics' Find a Nutrition Expert directory, a small set of nutrition-specific directories, and individual practitioner sites when those sites carry credentialed bylines and schema. On hybrid queries ('how do I find a dietitian for my PCOS' — half condition, half practitioner) the engines stitch citations from both tiers and the practitioner sites that get picked up are exclusively the ones with named credentials in machine-readable form.
OpenAI's published bot documentation4 distinguishes between GPTBot (training-side indexing), OAI-SearchBot (search index for ChatGPT Search), and ChatGPT-User (real-time fetch when a user's prompt triggers a web look-up). The practical implication for dietitians: real-time citation in ChatGPT depends on ChatGPT-User being able to reach the site at the moment of the query. If the Squarespace AI exclusion box6 is toggled on, ChatGPT-User cannot fetch — and the site is invisible to live citation regardless of how good the underlying content is.
Perplexity's documentation5 describes the same split — PerplexityBot for indexing, Perplexity-User for on-demand fetch — and Perplexity in particular favours content with explicit dates, named-source attribution, and clearly-bounded thesis sentences. The Perplexity citation card design (which shows the source publisher, the URL, and a snippet) rewards sites that can be quoted in a single self-contained passage, which maps directly onto the 134-167 word answer-block pattern this site uses on every page. The hub outlines the broader install; this page focuses on the specific levers that turn citation eligibility into actual citations.
§03Signals
The credential signal — schema, byline, and the source manifest
The credential signal AI engines read is a layered structure: a Person schema with jobTitle 'Registered Dietitian' and a hasCredential field pointing at the Commission on Dietetic Registration, a knowsAbout array listing the real modalities the practitioner specialises in, a visible byline on every page that mirrors the schema, and a source manifest of named external authorities cited inline throughout the content. Each layer is mechanical to ship and each adds a measurable lift in citation propensity. The full graph lives in Code Injection > Header so it covers every page, and the byline lives in the page template so authors do not have to remember to add it.
The Person schema is the single highest-leverage block. Schema.org's MedicalBusiness type7 inherits LocalBusiness and adds medicalSpecialty, and pairing it with a connected Person carrying credential and knowsAbout creates an entity graph the AI engines can read as "this practice is a Registered Dietitian's practice, run by this named person, with these named specialisations". The same content without the entity graph reads to a machine as "a website about nutrition" — which is what wellness blogs look like too.
JSON-LDPerson schema with credential + knowsAbout for an RD — the second half of the graph (first half on the hub)
<script type="application/ld+json">{"@context":"https://schema.org","@type":"Person","@id":"https://yourpractice.com/#practitioner","name":"Jordan Lee, RD, LD","jobTitle":"Registered Dietitian","hasCredential":{"@type":"EducationalOccupationalCredential","credentialCategory":"Professional Registration","recognizedBy":{"@type":"Organization","name":"Commission on Dietetic Registration"}},"knowsAbout": ["PCOS Nutrition Management","Medical Nutrition Therapy for IBS","Gestational Diabetes Counselling","Intuitive Eating"]}</script>
The visible byline is the second layer and is often skipped because it feels redundant against the About page. It is not redundant — the byline at the top of every article is what a human rater (and the AI engine's parser) sees first, and it is the field most directly aligned with the rater-guideline language about author expertise. "By Jordan Lee, RD, LD" at the top of every article reads as a credentialed author in two different parsing passes (the rater-guideline alignment and the machine-readable byline pattern); "Written by Jordan" reads as an unattributed wellness post.
The source manifest is the third layer and is the editorial-discipline lever. Every claim in the prose that could be checked against a named external authority gets cited inline ("According to the 2024 American Diabetes Association Standards of Care...", "The Academy of Nutrition and Dietetics' position paper on intuitive eating..."). The manifest gathers those citations into a numbered right-rail list with publisher, date, URL, and a one-line note. The combined effect — credentialed author plus visibly cited external authorities — is what AI engines read as "trustworthy on YMYL".
§04Content shape
The content shape that gets cited on dietitian health queries
The format that consistently gets cited has four properties. The first 200 words answer the primary query directly, in a self-contained passage that can be quoted without context. Every H2 section is a question or a near-question, with a 134-167 word bolded lead answering the heading and a longer expansion below it. Every quantitative or clinical claim names its source inline — not 'studies show' but 'the 2024 American Diabetes Association Standards of Care state'. And the named modalities a practitioner specialises in appear in the prose, the headings, the schema, and the byline area — repeated enough that an extractor reading the page concludes 'this practitioner specialises in PCOS' rather than 'this practitioner writes about many things including PCOS'.
The first-200-words rule is the most under-leveraged of the four. Search Engine Land's 2026 GEO research8 emphasises that AI engines extract from opening passages disproportionately — the model's retrieval pass weighs content near the top of the document more heavily because that is where authors put the thesis. The practical dietitian-site implication: the introduction to a PCOS nutrition article cannot be a soft "I'm so glad you're here, let's talk about PCOS"; it has to be "Polycystic Ovary Syndrome is an endocrine condition affecting an estimated 6-12% of women of reproductive age in the United States, and Medical Nutrition Therapy delivered by a Registered Dietitian is one of the first-line management approaches recommended by the Academy of Nutrition and Dietetics." The first version reads warmly; the second gets cited.
The named-modality repetition is the second high-leverage move. AI engines build entity graphs from a page's content: the more clearly a page signals "this is about X, by a credentialed practitioner of X" the more the engines slot it into the citation graph for X. Practitioners who specialise in PCOS but write blog posts on every nutrition topic dilute the signal; practitioners who concentrate the content on the named modality, repeat the modality in headings and byline, and carry the modality in the Person schema's knowsAbout array build a citation graph that holds for that modality specifically.
§05Crawlers
Crawler access: the 26-bot box you must leave off
The single most common reason a dietitian site is invisible to AI engines is the Squarespace AI exclusion box. Settings > Crawlers > AI Access ships unchecked by default — but a meaningful share of nutrition and health-adjacent sites have it toggled on, usually as a leftover from vague 2024-era advice about 'protect your content from AI'. The toggle does not protect content; it removes the site from citation eligibility across the 26 named bots Squarespace's exclusion controls, including the OpenAI, Anthropic, Perplexity, and Google-Extended crawlers. The audit is a single screenshot; the fix is a single click. Run it before anything else.
OpenAI's bot documentation4 is explicit that ChatGPT-User — the real-time fetch crawler — is the bot that determines whether a page can be live-cited when a user asks ChatGPT a question. If the Squarespace exclusion is toggled on, ChatGPT-User receives a 403 from the site's robots layer, the page is unreachable at the moment of the user's query, and the citation goes to whichever competitor's page is reachable. The same logic applies to Perplexity-User5 and the equivalent on-demand bots from Anthropic and Google. The exclusion is total across the 26-bot list, not selective by use case.
The 'protect your content' framing that drove some practitioners to toggle the box is itself confused. The exclusion box controls crawler access to the public site content — the same content human readers can already screenshot, copy, or quote. It does nothing about the underlying intellectual property; content licensing on AI training is a separate legal question that the Squarespace UI does not actually adjudicate. The practical reading: nutrition practitioners who want AI citation visibility leave the box unchecked, and any content-protection concern is handled at the policy and licensing layer, not at the crawler-access toggle.
§06Measurement
Measuring AI citation as a dietitian, without Search Console for AI
Google Search Console reports impressions and clicks for traditional search but does not yet expose AI Overviews citation data at the per-page level for most owners. ChatGPT, Perplexity, and Claude do not provide referral analytics in the way Google does. The practical 2026 measurement loop is therefore manual and panel-based: a curated list of 15-25 representative queries for the practice, run quarterly across ChatGPT, Perplexity, Google AI Overviews, and Gemini, with citation hits and source positions recorded in a tracking spreadsheet. This is the audit cadence we run on installs; the alternative is wishful thinking.
The query panel itself is the artefact that needs care. A good dietitian panel includes condition-led queries the practitioner targets ("PCOS nutrition management", "low-FODMAP for IBS"), local practitioner queries ("dietitian for PCOS in [city]", "RD who takes Aetna [state]"), credential-vs-uncredentialed queries ("difference between a nutritionist and a dietitian"), and program-led queries if the practice sells named programs ("6-week PCOS program", "RD-led intuitive eating cohort"). Twenty queries is enough to see signal; fifty is reliable; ten is too few to distinguish movement from noise. The same panel runs every quarter, the same prompts in the same order, the same engines.
The measurement output is not a single number; it is a citation map. Per query, per engine, the spreadsheet records whether the practice's site appeared in the citation list, what position it occupied, and what competing sources appeared above it. Movement quarter-over-quarter is the signal. The first quarter after an install is usually flat — entity recognition takes time to propagate — and the second-quarter delta is what we report to clients. The mechanism is slow but the lift, once it lands, is structural rather than algorithmic, which is why the second-tier wedge is worth the install effort in the first place.
