PublishedVerifiedEvery 6 weeksSources7 namedAuthored bySquareRank Team
Course creators · § 4.6.1 · How-to · The wedge
Course Discovery AI Queries
Prospective students ask ChatGPT, Claude, and Perplexity "what's the best course on X" in plain English1, and the model answers by listing two to five courses with brief descriptions and naming sources for further reading. The Squarespace course creator who shows up in that list has a structurally-named curriculum, a real instructor entity with credentials, a Course schema block on the sales page, and the AI crawler panel correctly set. The creator who does not show up has the wrong content shape, the wrong schema, or has the panel toggled on by mistake. All three are cheap to fix in one week.
This leaf is the methodology, the install, and the measurement loop for being cited in course-discovery answers. It covers how students actually phrase course-shopping queries on AI engines (long, conversational, outcome-driven), why named curricula cite while generic ones get skipped, the curriculum-shape pattern AI extracts from, the 5-step install end to end, the instructor Person schema with knowsAbout and hasCredential, the four pieces of citation content to ship first, and the manual measurement loop that confirms the work is landing.
§01Query shape
How students actually ask AI engines for a course
The course-shopping query shape on AI engines is fundamentally different from the keyword shape on Google. On Google, a student types 'best SaaS course' or 'newsletter growth course' — short, modifier-heavy, transactional. On ChatGPT and Claude, the same student types 'I'm a senior engineer thinking about leaving big tech to build a SaaS, what's a good course to learn product-market fit without quitting first' — long, narrative, situation-aware. The model parses the situation, identifies the help wanted, and answers with named courses and instructors. The course that shows up in that answer is the one whose content matches the conversational phrasing, not the one optimised for the keyword.
The pattern is consistent across the three engines doing most course-discovery routing. ChatGPT (~800 million weekly active users)1 tends to answer course queries with three to six recommendations, each annotated by audience fit and (sometimes) named instructor citations. Claude favours longer narrative responses with two to four named sources at the end and a structured comparison if the user asks. Perplexity returns the most-cited-source list inline, treating each cited course page as a top-line answer card the user can click directly. All three reward the same content shape — structurally-named curriculum, named instructor entity, defined audience, citation-grade source attribution.
The conversion math favours AI-routed course traffic when it lands. Students arriving through a conversational query have pre-qualified themselves through the model's clarifying back-and-forth — by the time they click through to the sales page, they know roughly what they want, what level they are at, what the engagement should look like. Visit volume is smaller than Google routes, but the close rate is materially higher. For high-ticket courses ($500-$5,000 price band) on Squarespace, this asymmetry is the most consequential shift in course discovery since the rise of organic YouTube as a top-of-funnel surface.
The course-AI routing math
800M
weekly ChatGPT users — many asking the conversational course-shopping queries Course schema can be cited for.
Why named courses cite well and generic ones do not
Three structural features of the online-course industry make AI engines treat named courses as good citation candidates more readily than they treat course directories or marketplaces. First, course content tends to be curriculum-and-outcome heavy, which matches the model's preferred extraction shape — clear modules with defined learning outcomes. Second, the instructor-led model means each course's authority anchors to a named human entity AI can disambiguate via sameAs links. Third, the conversational query pattern surfaces course-shaped answers more often than directory-shaped ones, because the student is asking for a recommendation, not a listing.
The third point compounds the first two. When a student asks "what's a good course on newsletter growth", the model is implicitly looking for a recommendation surface where it can cite an authority. A course marketplace like Udemy or Coursera is an aggregator, not an authority. A seven-figure brand's all-purpose course is a sales surface, not an authority. An individual creator with a structurally-named curriculum, an instructor Person entity, and Course schema reads, to the model, like the closest thing to an expert source for that specific learning intent. The model picks the closest expert. That is the wedge — and it favours Squarespace creators with a coherent personal brand over marketplaces with thin individual course pages.
The mechanism mirrors what Tiago Forte did with "Building a Second Brain", what Justin Welsh did with "The LinkedIn Operating System", and what Wes Bos did with his JavaScript courses at much larger scale. Each built a structurally-named curriculum that AI engines cite because the framework is concrete, defensible, and tied to a known instructor. The fractal applies at small-creator scale — a "4-Week Newsletter Engine" or a "Validation Sprint" with a published curriculum, an instructor Person schema entity, and Course JSON-LD works the same way for a 200-student course as the larger curricula work for the named creators. The threshold is structurally-named and schema-complete, not famous.
§03The shape
The curriculum-shape pattern AI engines extract from
Every Squarespace course sales page that wants to be cited needs a curriculum block structured for extraction. The block lives high on the page, follows a defined pattern, and ships with corresponding Course schema in the Code Injection header. Four to eight modules, each titled structurally rather than aspirationally ('Module 3 — Building the validation landing page' not 'Module 3 — The breakthrough chapter'), each with a one-sentence learning outcome. The block reads in 134-167 words in total and gives the model a complete, self-contained answer it can lift into a citation card without further inference.
The structural template is consistent across high-performing course pages. H2 is "Curriculum" or "What you'll learn". The intro paragraph defines what the course teaches and who it teaches in 30-50 words — short enough for a citation card. Then the module list: each module name, each learning outcome, one short line. Below that, a paragraph naming the format (self-paced, cohort, hybrid), the duration, the prerequisites, and the format of the deliverable (recorded video, live sessions, written assignments). The structure is mundane on purpose. The model's extraction pattern is mundane — give it the answer in the shape it expects, and it cites cleanly.
The naming choice is where most creators stall. Module names should describe the action or output ('Build the validation landing page', 'Ship the first 100 subscribers', 'Audit the cohort feedback loop') rather than the theme ('The build phase', 'Growth mindset', 'Reflection week'). Action names are easier for the model to map onto the student's stated goal in their conversational query. A student who said "I need to learn how to validate a SaaS idea" maps directly onto a module named "Build the validation landing page" and only indirectly onto a module named "The build phase". The first one cites. The second one gets skipped.
§04The install
The 5-step install, in order
The install runs in five sequential steps and skipping any of them produces incomplete citation. Crawler access first, because the model has to be able to read the sales page. Curriculum naming second, because the structurally-named modules are the extractable surface. Course schema third, because the entity context is what makes the citation defensible. Instructor entity fourth, because the model needs a real human to attribute the curriculum to. Manual tracking fifth, because AI-citation traffic mostly arrives without referrer data and direct query logging is the only honest measurement.
Step 1 — Crawler access. Open Settings > Crawlers in Squarespace. Confirm the AI exclusion toggle is unchecked3. Default state is correct; the trap is the creator who toggled it on after 2024-era "protect your course IP" advice. Verify in a private window that yoursite.com/robots.txt does not disallow GPTBot. ChatGPT-User does not strictly follow robots.txt anyway2 — its requests are user-initiated — but OAI-SearchBot does, and it decides whether you appear in ChatGPT Search source cards.
Step 2 — Curriculum naming. Audit the module names against the action-versus-theme test. Action names ship. Theme names get rewritten. The audit takes 30 minutes for an 8-module curriculum.
Step 3 — Course schema. Inject the Course JSON-LD block on the sales page via Page Settings > Code Injection > Header. Cover provider, instructor (nested Person), hasCourseInstance with courseMode and courseWorkload, and offers with price and priceCurrency. Sample block in the vertical-hub Course schema section.
Step 4 — Instructor entity. On /founder/ or the about page, inject Person JSON-LD with knowsAbout, hasCredential, and sameAs. Detailed code in the next section.
Step 5 — Manual tracking. Build a tracking spreadsheet with 10-15 conversational queries that map to your course audience and topic. Run weekly across ChatGPT, Claude, and Perplexity. Log appearances. The pattern is covered in detail in the dark-traffic leaf.
§05Entity wiring
Instructor Person schema — the entity-recognition layer
Person schema on the instructor's founder or about page is the entity-recognition handle AI engines use to confidently attribute a curriculum to a real human. For course creators specifically, two additional fields earn their keep beyond the standard pattern. hasCredential signals certifications, degrees, or recognised qualifications — material for any course in a credentialed domain (finance, legal, healthcare, design). alumniOf signals institutional affiliation, useful when the instructor's authority is anchored to a recognised university or notable employer. Both fields ship as optional but consistently lift citation likelihood in audits we run on course pages.
The Person spec5 ships knowsAbout as an array of strings, hasCredential as an EducationalOccupationalCredential, and alumniOf as an Organization. Strings work fine on Squarespace's Code Injection surface and read cleanly in JSON-LD validators. The pattern that performs best is five to ten specific topical tags in knowsAbout (each one matching a query intent), one to three hasCredential blocks for relevant credentials, and sameAs links to LinkedIn, podcast appearances, published articles, and any conference talks. The more reachable verification surfaces, the more confidently the model attributes the curriculum to the named instructor.
JSON-LDInstructor Person schema with credentials — Page Settings > Code Injection > Header on /founder/
<script type="application/ld+json">{"@context":"https://schema.org","@type":"Person","name":"Your Full Name","url":"https://yoursite.com/founder/","jobTitle":"Course Instructor","knowsAbout": ["SaaS validation for senior engineers","Pre-launch landing pages","Validating product-market fit without leaving employment","The Validation Sprint (proprietary curriculum)"],"hasCredential": {"@type":"EducationalOccupationalCredential","credentialCategory":"degree","name":"BS Computer Science"},"sameAs": ["https://www.linkedin.com/in/your-handle","https://podcast-with-you-as-guest.com/episode-link"]}</script>
The knowsAbout tags should mirror the conversational queries students would type into ChatGPT. The two installs reinforce each other — Course schema on the sales page gives the model the curriculum, Person schema with knowsAbout on the founder page gives the model the human authority. Code Injection is locked to Business plan and above; on Personal plan the workaround is the in-body author bio (less effective but still readable to most AI parsers).
§06The content
The four pieces of citation content to ship first
Four content slots, shipped in order, cover the majority of conversational query intents for a Squarespace course. The sales page with Course schema (already covered). One audience-deep-dive article per primary student archetype. One outcome page per defined transformation. One curriculum-explainer page that justifies the module structure in editorial form. Together those four pieces give the model enough surface to cite the course across the full conversational query universe, not just the curriculum query — which is critical because most discovery questions are about the audience or the outcome, not the curriculum name.
The audience deep-dive. One blog post per primary student archetype. Title is the archetype phrasing ("A course for senior engineers thinking about leaving big tech", "A course for solo founders post-product-market-fit"). The post describes the archetype, the typical challenges, references the curriculum modules that map onto each challenge, and links to the sales page. 1,000-1,500 words. Two named sources minimum. Ranks for the long-tail query AND cites for the conversational version.
The outcome page. One page per defined transformation ("From SaaS idea to validated landing page in 30 days", "From 0 to 100 newsletter subscribers in 4 weeks"). Describes what the course produces, with structural details (number of modules, weekly time commitment, deliverables) and a real student quote where available. 800-1,200 words. Half methodology demonstration, half conversion surface.
The curriculum-explainer page. A long-form editorial walkthrough of why the modules are sequenced the way they are sequenced. This page is the citation hook for queries like "what should I learn first when validating a SaaS" — the model cites the explainer page because the explainer page makes the structural argument, not because it sells the course. The sales page is downstream. The explainer page is upstream of the buying decision. This is the leverage point most course creators under-ship and the reason their sales pages get traffic but no qualified conversation.
§07Measurement
Measuring whether AI engines are actually citing the course
AI-citation measurement is harder than Google measurement for course creators specifically because the routing pattern is conversational and the student decision cycle is long — a learner might first encounter a course in a ChatGPT answer, return three weeks later via a Google search for the brand name, and finally buy after seeing a creator's newsletter. Direct attribution to AI citation is rare. The honest 2026 measurement stack is a manual query log run weekly, a GA4 custom channel grouping for the tagged subset, and a creator-side discovery survey that asks new students how they heard about the course. None of the three alone is sufficient. Triangulation is.
The manual query log is the floor. Pick 10-15 conversational queries that map to the course audience and topic — "what's a good course on SaaS validation for senior engineers", "I want to start a newsletter, what course should I take", "best courses for solo founders on product-market fit". Run each one through ChatGPT, Claude, and Perplexity on the same morning weekly. Log whether the course name, the instructor, or the sales page surfaces. Screenshot when you appear. Quarterly review aggregates the trend and informs which audience deep-dives or outcome pages need additional editorial coverage.
The GA4 layer captures the visible fraction. Create a custom channel grouping called AI Search using a regex that matches chatgpt.com|chat.openai.com|perplexity.ai|claude.ai. OpenAI started tagging some citation links with utm_source=chatgpt.com in 2024 and extended it to More sources in mid-2025; conversational inline links remain untagged. Expect GA4 to undercount actual AI-routed course traffic by an order of magnitude. The creator-side survey closes the gap — a single "how did you hear about this course" field on the checkout form, optional, captures the qualitative signal the analytics layer cannot. The full GA4 setup pattern is covered in the dark-traffic leaf.
