How RTO websites can improve visibility in ChatGPT Search and Google AI Overviews

A prospective student might ask: “Which Certificate III courses can I study part-time near Parramatta, what will the full cost be, how long will it take, and will I need to organise my own placement?”

That question combines location, delivery mode, fees, duration and placement. An answer may appear through ChatGPT Search, a Google AI Overview, Google AI Mode or an ordinary search result. These systems may use several sources, including one other than the RTO’s website.

An RTO cannot guarantee that either platform will cite, link to or recommend it. OpenAI publishes crawler-access guidance, not a source-ranking formula. Google says its AI features use existing Search systems and require no separate AI-optimisation method. Selection remains at the platform’s discretion. OpenAI’s ChatGPT Search guidance and Google’s AI-features guidance explain those limits.

What an RTO can improve is the quality of its source material: accurate course information, useful first-party detail, fewer contradictions, sound technical access and disciplined maintenance. These improvements also help students arriving through conventional search, referrals or direct navigation.

1. What AI-search visibility means for an RTO

ChatGPT Search

ChatGPT Search can search the web and include citations or source links. It may rewrite a question into several searches, retrieve different sources and synthesise a response. OpenAI cautions that search results and citations can still be incomplete, outdated or incorrect.

For an RTO, visibility may mean a course page is cited for a comparison, a location page supports a delivery query, or a user follows a referral link. A citation is not an endorsement, regulatory approval or persistent ranking. It is a source choice for one response at one time.

Google AI Overviews and AI Mode

Google AI Overviews can present an AI-generated summary within search results. AI Mode provides a conversational search experience. Both may break a question into related searches across subtopics and data sources, a process Google calls “query fan-out”.

Google says these features are grounded in its Search index and established ranking and quality systems. A page generally needs to be indexed and snippet-eligible, but eligibility does not ensure selection. Beyond ordinary Search eligibility and inclusion through the Search Console control where available, Google does not require special AI markup, schema or an AI text file. Its generative-AI optimisation guide also rejects content chunking and AI-specific rewriting as requirements.

Practical visibility therefore includes possible source inclusion, referral traffic and useful student actions such as reading fees, preparing an application or making an enquiry. It is not a public “trust score”, endorsement or stable citation position.

2. Begin with student research questions, not AI-search tricks

Content planning should begin with the decisions prospective students need to make:

• Availability: Is the qualification offered now, and when is the next planned intake?
• Mode and location: Is training face-to-face, online, workplace-based or blended, and where does each component occur?
• Duration and workload: What are the planned program length, attendance pattern and independent-study expectations?
• Fees and funding: What is the total fee, what else may cost money, and what eligibility must be confirmed?
• Entry and suitability: Are there formal prerequisites or provider pre-enrolment assessments?
• Placement and external requirements: Who arranges placement, and do screening, licensing, equipment or travel conditions apply?
• Support: What learner support exists and how is it requested?
• Pathways and comparisons: How does the course differ from related options?
• Enrolment preparation: Which documents, checks and equipment are needed?

One compound question may depend on a canonical course page and supporting pages about fees, funding, placement, support or locations. The course page should answer essential decision questions and link to deeper information with descriptive anchors. Supporting pages should link back where course context matters.

This does not justify a separate page for every query variation. Near-duplicate pages for “online”, “part-time”, “near me” and other wording variants can fragment signals and create maintenance problems. Google also says content does not need to be broken into tiny fragments or rewritten specifically for AI systems.

Clear headings, direct opening sentences, meaningful lists and well-labelled tables help readers scan. They are sound editorial practices, not a known citation formula. Preserve the qualifications and context a prospective student needs rather than manufacturing “AI-sized chunks”.

3. Make each course page a reliable first-party source

Training.gov.au is the authoritative National Register for nationally recognised training. It records training products, RTO registration and scope, but not normally a provider’s timetable, fees, support, delivery sequence or next intake.

An RTO page should distinguish:

1. The national training product: its code, title, status and nationally defined requirements.
2. The provider’s offering: where, when and how it is delivered, what learners pay, which pre-enrolment processes apply and what practical commitments are involved.

ASQA’s Information Practice Guide calls for clear, accurate and current pre-enrolment information, with examples including duration, mode, location, fees, support, equipment and placement. It does not prescribe that every item appear on every public page; each RTO must determine how it meets the Standards and other applicable obligations.

Where applicable and verified, a canonical course page will commonly make these points easy to establish:

• code, title, RTO name and registration identity;
• current delivery modes and locations;
• planned commencement arrangements;
• duration, schedule, attendance and independent workload;
• tuition fees, additional costs, payment options and funding conditions;
• formal prerequisites;
• provider suitability or pre-training processes, separated from prerequisites;
• placement, licensing, screening, technology, equipment or travel requirements;
• learner-support arrangements;
• carefully qualified pathways or occupational outcomes;
• the next step for eligibility checks, information or application.

Not every policy must be reproduced in full. A concise visible summary can link to current fees, refunds, support, complaints or funding information. Use descriptive link labels and preserve enough course context to avoid ambiguity; a direct destination is more useful than “learn more” leading to a general document library.

Treat the canonical page as the maintained overview of the current offer, not merely a campaign landing page. Use it consistently from navigation, search, campaigns and supporting content so students and retrieval systems encounter the same offer.

Match claims to the evidence. “Planned duration is nine months” is clearer than “finish fast”. “Funding eligibility is assessed before enrolment” avoids implying every applicant receives a subsidy. Do not promise employment, licensing or migration outcomes the course cannot guarantee.

The forthcoming guide to what the 2025 Standards for RTOs mean for an RTO website can provide the broader compliance context without turning a course page into a regulatory checklist.

Worked example: fragmented information versus a canonical page

This example is hypothetical. “Harbour Skills Institute” and all course details are invented.

The second version is not better because it repeats keywords. It resolves contradictions, identifies the actual offer and tells the student what to verify next. That also gives a search system a clearer first-party source.

4. Publish information that training.gov.au cannot provide

Paraphrasing the national training product adds limited provider-specific value. The RTO website is more useful when it explains what studying with that provider is likely to involve.

Useful first-party information may cover:

• how face-to-face, online, workplace and practical components fit together;
• the normal weekly study pattern;
• how practical skills are taught and assessed;
• how placement is prepared, arranged, monitored and supported;
• facilities, simulated environments, software, tools or equipment;
• how support is requested and what happens next;
• preparation problems such as missing checks, unsuitable devices or unrealistic availability;
• situations in which the delivery model may not suit a learner;
• practical differences between related qualifications or formats;
• recurring questions identified by admissions, trainers and support staff.

The detail must remain accurate and qualified. Do not describe a typical timetable as fixed when intakes differ, imply two qualifications are interchangeable or present one provider’s placement process as an industry-wide rule.

Operational staff often hold useful knowledge that has never reached the website. A content review can convert recurring questions into approved public guidance, with a named owner, evidence source and review trigger.

Distinctive information need not be novel. It must be genuinely attributable to the RTO’s offer and useful for a decision. A precise account of the weekly timetable, placement process or required software is usually more valuable than another generic qualification summary.

5. Remove contradictory, stale and hard-to-maintain information

Contradictions commonly arise from duplicate campaign pages, superseded qualifications, inconsistent location pages, old fee schedules, indexed PDFs, copied partner or agent content, and facts manually repeated across templates.

ASQA’s Information and Transparency Practice Guide emphasises accurate, current and non-misleading marketing. It identifies changes to training products, scope, training or assessment strategies, delivery modes and locations as review triggers, and addresses correct identification of the RTO and third parties.

Identify one canonical page for every current offering. Then:

• redirect a true duplicate or replaced page;
• retain an archive only for a genuine user, contractual or recordkeeping need;
• remove or restrict pages that should no longer be public;
• use `noindex` where a necessary public page should not appear in search;
• update internal links to the final current URL.

A canonical tag may help Google interpret duplicate URLs, but it does not correct wrong information for a reader.

Where possible, store high-risk facts once. A controlled fee, location or delivery-mode field is easier to govern than copies across pages, cards, brochures and campaigns. Give each material fact an owner, evidence source and review triggers covering HTML, PDFs, forms, feeds, partner content and links.

Calendar reviews are insufficient for facts that can change between checks. A fee approval, scope amendment, replaced product, new location, revised training and assessment strategy or funding change should trigger a task covering every dependent surface, including controlled third-party pages.

Where an agent or partner publishes course information, define ownership and update timing. The RTO may not control every external page, but it can control distributed source material, monitor important listings and stop linking to obsolete copies.

PDFs are not categorically unreadable: Google can index them, and other systems may process them. The problems are mobile usability, inconsistent updates and old versions remaining indexed. Publish an accessible HTML summary of essential decision information and retain a PDF where the download serves a separate purpose.

Links to training.gov.au are useful for student verification of the product, registration and scope. No official evidence shows that such a link creates an AI citation or special ranking signal.

6. Confirm technical eligibility for ChatGPT Search and Google

Quality information cannot be selected if a platform cannot reliably access it. Technical eligibility is necessary, not sufficient.

Some of the following checks will require access to Search Console, hosting, CDN or server-log settings.

ChatGPT Search access

OpenAI identifies OAI-SearchBot for surfacing websites in ChatGPT search results and GPTBot for potential model training. Their controls are independent. Blocking GPTBot does not require blocking OAI-SearchBot; allowing OAI-SearchBot does not imply consent to GPTBot access. OpenAI publishes current user agents and IP ranges in its crawler documentation.

Check:

• whether `robots.txt` allows OAI-SearchBot on intended public pages;
• whether the CDN, firewall, hosting security or bot manager blocks current published crawler IPs;
• whether pages and essential resources return successful public responses without login or challenge;
• whether important information is available without an inaccessible interaction;
• whether server logs show crawler requests, denials or errors.

A permissive robots file cannot overcome a challenge, firewall rule or failing origin. Test outside the administrator’s session, because caching or allowlisted staff IPs can hide a public failure. Recheck OpenAI’s published IP ranges rather than trusting a static list indefinitely.

Crawler access does not require exposing student portals, learning systems or unpublished documents. Make approved public information available while preserving normal controls elsewhere.

OpenAI’s publisher guidance says ChatGPT referral links automatically include `utmsource=chatgpt.com`. Preserve that parameter through redirects and classify it consistently in analytics.

Google crawl, index and display eligibility

Google says a page needs to be indexed and snippet-eligible for consideration in AI Overviews or AI Mode, without any guarantee of selection.

Where the Search generative AI control is available in Search Console, confirm that the property is set to include the site. Google says inclusion is the default, but a property can be excluded or inherit an exclusion from a parent property. This control affects AI Overviews, AI Mode and generative features in Discover; it is separate from Google-Extended training controls.

The foundation is ordinary technical SEO:

• permit Googlebot and return successful responses without accidental `noindex`;
• provide meaningful text that Google can render;
• select consistent canonical URLs and handle duplicates;
• use crawlable internal links with descriptive anchors;
• include current canonical URLs in XML sitemaps;
• control snippet directives deliberately;
• test JavaScript-rendered content where critical facts depend on it;
• maintain secure, usable mobile pages and reasonable performance.

Use Search Console to inspect indexing, selected canonicals, crawl issues, sitemaps, the Search generative AI control where available and search performance. It cannot explain every AI source-selection decision, but it can expose basic eligibility failures.

7. Structured data is a supporting layer, not a citation engine

Structured data describes entities and relationships already represented on a page. It can support eligible Google features and improve template consistency. It must match visible content, and valid markup does not guarantee a rich result.

It is not a documented AI citation mechanism. Google requires no special structured data for AI Overviews or AI Mode, while OpenAI publishes no schema-based citation formula.

Two former Google features matter when assessing older RTO advice:

• FAQ rich results: Google stopped showing them from 7 May 2026 and removed the documentation in June 2026, according to its Search documentation update log. `FAQPage` is not a route to the former presentation and was never documented as an AI citation mechanism.
• Course information rich results: Google began phasing these out in June 2025 and later removed Search Console and Rich Results Test support, as explained in its retirement notice. Google said ordinary rankings were unaffected.

Schema.org still defines `Course` and `CourseInstance`; the latter can distinguish an occurrence by mode, location or schedule. Their existence does not mean Google offers a corresponding rich result or that an AI system will cite the page.

Optional markup may include `Organization`, `BreadcrumbList`, `Article` or `BlogPosting`, and `Course` or `CourseInstance` where these accurately reflect visible content and serve an implementation purpose. Do not add `sameAs`, author data, training.gov.au links or a large schema graph on the assumption that it will create citations. Validate syntax and remove stale or unsupported values.

8. Measure what is observable

Separate technical health, platform appearances, referrals, student actions and content governance.

Use Search Console, URL Inspection, sitemap reports and server logs to confirm that important pages are crawlable, indexed and revisited. Track canonical mismatches, exclusions, blocked resources and legacy URLs still receiving traffic. OpenAI crawler requests in logs show access, not answer inclusion.

Google began rolling its generative-AI performance report out to a subset of properties in June 2026, and it may still be unavailable where a property is not included in the rollout or has insufficient impressions.

The report covers appearances in AI Overviews and AI Mode and can be segmented by page, country, date and device. It reports impressions rather than prompts or a dedicated generative-feature click metric. Treat impressions as evidence of appearances, not a complete account of influence or referral value.

For ChatGPT, report sessions carrying `utm_source=chatgpt.com`, their landing pages and downstream actions. Check that redirects retain the parameter and disclose gaps caused by consent settings or analytics blocking.

Useful outcome measures include course-page visits, progression to fees or placement information, enquiries, application starts and assisted conversions. Review them by landing page and course rather than combining all “AI traffic”. A few visits to a high-consideration course may matter more than many impressions producing no action.

Attribution remains incomplete. A student may read an AI answer, return through branded search and enquire; another may obtain the answer without clicking. Treat referral and conversion data as observable parts of the journey, not a complete measure of influence.

Governance measures include the percentage of active pages reviewed, unresolved conflicts, the age of high-risk facts and completion of reviews after material changes. These measures are often more actionable than fluctuating citation counts because the RTO can directly control them.

Manual prompt tests can reveal factual errors, omitted context or unexpected sources. They are directional observations, not stable rankings: results vary with wording, location, personalisation, model and index changes. Record the exact prompt, date, account conditions and observed sources.

9. A prioritised RTO AI-search readiness audit

Apply the following sequence:

1. Check crawler access and eligibility. Test Google crawl access, index status, snippet eligibility and the Search generative AI control where available, then test OAI-SearchBot access through robots, CDN, firewall and server responses.
2. Choose canonical current course pages. Map duplicates, campaign pages, superseded products, location variants and PDFs; redirect, archive or restrict them deliberately.
3. Verify material facts. Confirm identity, scope, mode, location, timing, fees, funding, entry, placement, support and next steps against approved evidence.
4. Remove contradictions. Compare canonical pages with cards, forms, brochures, location pages and third-party content.
5. Expose essential information in accessible HTML. Use sensible headings, descriptive links and usable mobile presentation; retain PDFs for separate purposes.
6. Add provider-specific information. Explain study patterns, practical delivery, placement, facilities, support, preparation issues and suitability limits.
7. Strengthen internal linking. Connect courses to relevant fees, funding, support, location, placement and application guidance.
8. Validate structured data. Keep only accurate, visible and operationally useful markup.
9. Assign owners and review triggers. Trigger reviews after changes to products, scope, strategy, fees, funding, mode, location, third parties or regulation.
10. Measure observable results. Monitor crawl and index health, available Google AI impressions, ChatGPT referrals, behaviour, enquiries and content-review completion.

Conclusion: become a better source by becoming a better website

The defensible way to improve an RTO website’s prospects in ChatGPT Search and Google’s AI features is not to pursue a hidden citation formula. Publish a clear account of what the RTO offers, make essential information accessible, remove conflicting versions and keep material facts controlled.

Prioritise canonical course pages, useful first-party information, clean crawler access and disciplined maintenance—not a new schema plugin or AI file. These improvements make the website easier for prospective students to assess and more eligible to serve as a source, without promising a citation the RTO cannot control.

Frequently asked questions

Can an RTO guarantee that ChatGPT Search or Google AI Overviews will cite its website?

No. OpenAI and Google provide eligibility and access guidance, but neither publishes a process that guarantees a particular source will be cited, linked or recommended. Results can vary by question, location, timing, index state and platform changes. An RTO can improve the accuracy, usefulness, accessibility and consistency of its pages, then measure observable appearances and referrals. It should not market that work as guaranteed placement.

Does FAQPage schema improve an RTO’s AI-search visibility?

There is no official evidence that `FAQPage` schema improves selection in ChatGPT Search, Google AI Overviews or AI Mode. Google stopped showing its FAQ rich result from 7 May 2026. Questions and answers can still be useful when they address genuine student concerns, but they should be written for readers and maintained as visible content. Adding the schema does not turn them into a citation mechanism.

Does an RTO website need an llms.txt file?

No current official guidance makes `llms.txt` a requirement for visibility in these products. Google explicitly says Search does not use it and requires no special AI file. OpenAI’s published guidance focuses on allowing OAI-SearchBot and its current crawler IP ranges. An RTO may maintain `llms.txt` for another system that uses it, but it should not replace robots controls, crawlable pages, internal links, sitemaps or accurate content.

Should important course information be published in HTML as well as PDF documents?

Usually, yes. PDFs can be indexed and may be processed by search and AI systems, so the reason is not that they are categorically unreadable. A visible HTML summary is easier to navigate, link to, update, render on mobile and keep aligned with the canonical course page. Keep a PDF where a downloadable or formally issued document is useful, and manage old versions so they do not remain discoverable after replacement.

How can an RTO measure visits from ChatGPT Search and Google’s AI features?

Track ChatGPT sessions carrying `utm_source=chatgpt.com`, their landing pages and subsequent actions. For Google, use Search Console’s generative-AI report where available; its rollout is limited and it principally reports impressions rather than prompts or a dedicated feature-click metric. Combine those signals with ordinary Search Console data, server logs, enquiries, application starts and assisted conversions. Manual prompt checks can supplement the evidence but should be labelled as dated observations, not stable rankings.