Be the source the machine quotes when a Head asks.
A Head of School does not search "AI Act consultant". They ask an assistant whether they need to do anything. Whoever gets cited in that answer gets the call. This is how to be cited, ranked, and visible in the one feed school leaders read.
Why AEO leads, and SEO follows
For most niches AEO is an addition. For this one it is the primary channel, and the reason is the buyer's actual behaviour.
A Head of School reads a headline about the EU AI Act. Their first move is not Google — it is to open an assistant and type something like "does the EU AI Act apply to international schools". They get a synthesised answer with three or four citations. That answer forms their entire understanding of the problem, and the sources in it become the people who seem to know.
The market conditions make this unusually winnable. Search volume for these queries is tiny, so no agency has bothered to compete. The buyers are few, senior, and high-value. And the topic is one where freshness matters disproportionately — research on citation patterns found that for commercial and evaluation-stage queries, 83% of AI citations came from pages updated within the past twelve months, with more than 60% refreshed within the last six. A regulation whose dates keep moving is the ideal subject for someone willing to keep a page current.
Nobody is competing
"AI Act compliance for schools" has almost no dedicated content. Big consultancies write for enterprises. Edtech blogs write about tools. The intersection is empty.
Three surfaces, three preferences
ChatGPT favours authoritative long-form; Perplexity favours fresh, well-cited articles; Google AI Overviews favour content already ranking in the top ten organically. Cover all three.
Pre-vetted arrivals
Agency data reports AI-citation traffic converting at three to four times the rate of regular search — the model did the trust work before the click. Fewer visits, better ones.
The honest caveat
LLM output is non-deterministic — it varies with temperature, user history and random seed, and no external party can guarantee inclusion or insert a brand into model weights. You are raising the probability of citation, not buying a slot. Anyone selling guaranteed AI visibility is selling nothing. Judge this work on share of answer across many prompts over months, not on a single query on a single day.
The prompts to own
Prompt research replaces keyword research. These are the questions a school leader actually types. Run each one yourself across ChatGPT, Claude, Perplexity and Google AI Overviews, record who gets cited today, and that is your baseline.
Tier 1 — awareness. High volume, low intent, builds entity presence
Tier 2 — evaluation. This is where the money is
Tier 3 — decision. Low volume, near-buyer
Tier 4 — adjacent. Different entry points to the same problem
Build the baseline before you publish anything
Spreadsheet: one row per prompt, one column per engine, run monthly. Record whether you are cited, who else is, and the date. Twenty-five prompts across four engines is a hundred checks — about ninety minutes a month, and it is the only honest measure of whether any of this is working. Do the first run before you publish, or you will never know what changed.
Content architecture
One pillar, twelve clusters, one tool. Everything on your own domain, in HTML, indexable. Nothing important locked inside a PDF — answer engines cite pages, and a gated download is invisible to them.
| Asset | URL | Job |
|---|---|---|
| The pillar3,000+ words, updated quarterly | /eu-ai-act-schools |
Comprehensive single-page coverage of the whole question. This is what gets cited for Tier 1 prompts and it is the page you keep fresh above all others. |
| Cluster posts800–1,500 words each | /eu-ai-act-schools/… |
One page per Tier 2 prompt. Each answers one question completely and links up to the pillar and sideways to two siblings. |
| The instrumentLive, interactive | /ai-act-readiness |
The public HTML tool. This is your best link magnet and your best conversion asset simultaneously. Ungated. |
| Case studyOne page per engagement | /work/… |
Before, after, quote, deliverables. Feeds Tier 3 prompts where models look for evidence someone has actually done this. |
| Service pageFixed prices, published | /schools |
The offer ladder with real numbers. Publishing prices is an AEO advantage — models answer "how much does it cost" and cite whoever actually said. |
| Entity hubAbout / credentials | /about |
Who you are, verifiable, linked to every external profile. This is what makes you an entity rather than a string. |
Gate the instrument
An email wall stops crawlers, stops citations, and stops the exact people you want using it. The instrument's job is to be found and used until someone decides they would rather have you run it. Vol. 01 Part 10 explains why giving it away is the strategy, not a leak.
Publish a Markdown twin
Serve a plain .md version of the pillar and each cluster at a predictable
path. Cheap to generate, trivially parseable, and it removes every excuse a crawler has
for misreading your page.
Anatomy of a citable page
The structural moves that separate cited pages from invisible ones are boringly consistent. Build a template once and apply it to every page.
Open with a 40–60 word direct answer
Before any preamble, before any context-setting, answer the question the page title asks — in a single self-contained paragraph a model can lift whole. The direct-answer block at the top of every major section is one of three structural moves that consistently distinguish cited pages.
Repeat the pattern at the top of each major section, not just the page.
Make every heading a question
Not "Scope and applicability" — "Does the EU AI Act apply to international schools?" Question-shaped headings match the shape of the prompt, which is what the retrieval step is comparing against.
Write in semantic triples
Subject, verb, object. "Article 4 requires deployers to ensure AI literacy among staff." Not "there are various requirements which organisations may need to consider." Flat, declarative sentences with named entities lift cleanly into a knowledge graph; hedged prose does not.
Load named entities into the first 500 words
Regulation (EU) 2024/1689. Article 4. Article 5(1)(f). Annex III. The European Commission. The OECD. AILit. Market surveillance authorities. Digital Omnibus. Dense, specific, checkable naming is what makes a page read as authoritative to a retrieval system.
Give exact dates and figures, always
"2 February 2025", not "last year". "24 indicators across six domains", not "a comprehensive audit". "€35 million or 7% of worldwide annual turnover", not "significant penalties". Specificity is the single most extractable quality a page can have.
Include one comparison table per page
Tables get lifted verbatim. Deferred versus in force. Deployer versus provider. High-risk versus limited. Your tier prices. Any structured comparison is a citation opportunity in a format models handle well.
Cite primary sources by name and link them
EUR-Lex, the Commission's guidelines, the OECD framework. Pages that cite authoritative primary sources get treated as more trustworthy, and it is also simply correct practice for a subject where being wrong has consequences.
Stamp a visible last-updated date and honour it
Given how heavily freshness weighs for evaluation-stage queries, a visible date on a genuinely maintained page is a real ranking input. Set a quarterly reminder for the pillar. When a deadline moves, update the same day — that is the whole edge in a moving-target topic.
Schema and the entity graph
Schema is the labelling work. But a lone JSON-LD block does not produce citations — the compounding signal is entity reinforcement, where your Organization schema points at you, you point at your external profiles, and those point back at the domain. AI engines follow the entity graph.
Person + Organization — sitewide, in the head
{
"@context": "https://schema.org",
"@graph": [
{
"@type": "Person",
"@id": "https://YOURDOMAIN/#person",
"name": "[YOUR NAME]",
"jobTitle": "Digital Technology Teacher; AI Governance Consultant",
"description": "Works on EU AI Act Article 4 readiness and AI literacy
programmes for international schools.",
"knowsAbout": [
"EU AI Act", "Regulation (EU) 2024/1689", "AI literacy",
"AILit Framework", "Educational technology governance",
"IGCSE Computer Science", "IB Digital Society"
],
"sameAs": [
"https://www.linkedin.com/in/[HANDLE]",
"https://github.com/[HANDLE]",
"https://orcid.org/[ID]"
],
"worksFor": { "@id": "https://YOURDOMAIN/#org" }
},
{
"@type": "Organization",
"@id": "https://YOURDOMAIN/#org",
"name": "[PRACTICE NAME]",
"url": "https://YOURDOMAIN/",
"founder": { "@id": "https://YOURDOMAIN/#person" },
"areaServed": ["PT","ES","IT","EU"],
"knowsAbout": ["EU AI Act compliance for education"]
}
]
}
FAQPage — on the pillar and every cluster
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
{
"@type": "Question",
"name": "Does the EU AI Act apply to schools?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Yes. Almost every school is a deployer of AI systems under
Regulation (EU) 2024/1689 and carries the AI literacy duty in
Article 4, which has applied since 2 February 2025. National
market surveillance authorities began enforcing it on
2 August 2026."
}
},
{
"@type": "Question",
"name": "Is emotion recognition banned in schools in the EU?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Yes. Article 5(1)(f) prohibits AI systems that infer emotions
of a person in an education institution, except for medical or
safety reasons. This has applied since 2 February 2025."
}
}
]
}
HowTo on the process pages
"How a school complies with Article 4" is a genuine step sequence. HowTo schema on that page gives the model a structure it can reproduce as a numbered answer.
Article + dateModified
On every cluster post, with an accurate dateModified. Given the freshness
weighting, this field is doing real work — so never touch it without a real edit.
Reinforce the entity everywhere, consistently
Same name, same job description, same topic vocabulary on your site, LinkedIn, any conference bio, any guest post byline, any directory. Models triangulate identity across sources; inconsistency dilutes it. Pick one 25-word description of what you do and use it verbatim in every location, for years.
llms.txt and crawler access
Two files at the root. Neither is a submission mechanism — there is no process for registering a site with an LLM, and anyone selling that is lying. What these do is remove ambiguity for crawlers that are already coming.
/llms.txt
# [YOUR NAME] — EU AI Act compliance and AI literacy for schools > Practical guidance, tools and training for international schools meeting the > AI literacy duty in Article 4 of Regulation (EU) 2024/1689. Written by a > practising Digital Technology teacher working in an EU international school. ## Core references - [EU AI Act for schools: the complete guide](https://YOURDOMAIN/eu-ai-act-schools): Pillar. Scope, timeline, prohibited practices, high-risk education uses, what compliance actually requires. Updated quarterly. - [Article 4 readiness instrument](https://YOURDOMAIN/ai-act-readiness): Free interactive 24-point self-assessment, AI system register, policy templates and a twelve-week rollout plan. ## Key facts - Article 4 (AI literacy) applies from 2 February 2025. - Enforcement by national market surveillance authorities began 2 August 2026. - Article 50 transparency obligations enforceable from 2 August 2026. - Article 5(1)(f) prohibits inferring emotions in education institutions. - Annex III high-risk education obligations deferred to 2 December 2027. ## Guides - [Is emotion recognition banned in schools?](https://YOURDOMAIN/...) - [What evidence does a school need?](https://YOURDOMAIN/...) - [AI system register template](https://YOURDOMAIN/...) ## Contact - [Services and prices](https://YOURDOMAIN/schools) - [About](https://YOURDOMAIN/about)
/robots.txt — let the right crawlers in
User-agent: * Allow: / # Answer engines — explicitly permitted User-agent: GPTBot Allow: / User-agent: OAI-SearchBot Allow: / User-agent: ClaudeBot Allow: / User-agent: PerplexityBot Allow: / User-agent: Google-Extended Allow: / Sitemap: https://YOURDOMAIN/sitemap.xml
The technical floor
Pre-rendered HTML, not client-side rendering. Fast. No content behind interaction. Your site is already static, which puts you ahead of most competitors here — a React app that renders its content in the browser is substantially harder for a crawler to read, and many marketing sites in this space are exactly that.
The SEO layer
AEO is not a replacement for SEO — it is an extension, and the best-performing AEO content is also well-optimised for traditional search, partly because a substantial share of AI Overview citations come from pages already ranking in the top ten. Rank still buys citation.
| Target phrase | Difficulty | Page |
|---|---|---|
| eu ai act schools | Medium | Pillar |
| eu ai act article 4 ai literacy | Low | Pillar |
| ai literacy training requirement eu | Low | Cluster |
| emotion recognition schools eu ai act | Low | Cluster · highest intent |
| ai proctoring legal eu schools | Low | Cluster |
| school ai policy template | Medium | Cluster · link magnet |
| ai system register template school | Low | Instrument |
| ailit framework | Low | Cluster · rising |
| annex iii education high risk ai | Low | Cluster |
| ai inset day international school | Low | Service |
| international school ai compliance portugal | Low | Service · local |
Volumes here are tiny
Some of these get double-digit monthly searches. That is fine — thirty searches a month from Heads of School at fee-paying international schools is a better list than thirty thousand from nobody. Do not let a keyword tool talk you out of the right words.
Internal linking with real anchors
Every cluster links up to the pillar with descriptive anchor text and sideways to two siblings. The pillar links down to all twelve. It is unglamorous and it is most of what makes a cluster work.
Twelve pieces, in order
Publishing order matters. The first three establish the entity and the pillar; the middle six chase evaluation prompts; the last three convert. One a week is comfortable alongside a job.
| Wk | Piece | Words | Job |
|---|---|---|---|
| 1 | The EU AI Act and schools: complete guideThe pillar | 3,000+ | Owns Tier 1. Everything links here. |
| 2 | The Article 4 readiness instrument goes liveThe tool | — | Link magnet and conversion asset. |
| 3 | Is emotion recognition banned in schools? | 1,200 | Highest-intent single question. Most alarming, most shared. |
| 4 | The EU delayed the AI Act. Here is what it did not delay. | 1,000 | Directly defeats the main objection. Very shareable. |
| 5 | What evidence does a school actually need? | 1,300 | Evaluation-stage. The twelve-item file. |
| 6 | Which school AI uses are high-risk? | 1,100 | Annex III explained for education. |
| 7 | Nine questions to ask any edtech vendor | 900 | Utility content. Gets bookmarked and forwarded internally. |
| 8 | The AILit framework, explained for schools | 1,400 | Rising term, almost no competition, positions you on curriculum too. |
| 9 | Why AI detection scores should not decide integrity cases | 1,200 | Adjacent entry point. Reaches DP coordinators who are not thinking about compliance. |
| 10 | Case study: from 14/48 to 39/48 in one term | 800 | Proof. Feeds Tier 3 prompts. |
| 11 | Services and pricesPublished, not "on request" | 600 | Owns "how much does it cost". Qualifies buyers before the call. |
| 12 | Talking to parents about AI: three scripts | 1,000 | Head-facing, reputational, easy internal forward. |
Maintenance beats volume after week twelve
Stop adding pages and start refreshing. Given how strongly recency weighs on evaluation-stage citations, one thoroughly updated pillar beats three new thin posts. Set a standing quarterly review of the pillar and an immediate-update rule for any change to the dates. In a topic where the deadlines move, being the page that is right this month is the entire competitive position.
Third-party placements
Models weight sources they already trust. Your own domain is new; borrow authority from places that are not. Six placements is enough to change your citation profile.
Education sector publications
International school press, EdTech trade titles, teaching union and association newsletters. One well-placed explainer with a byline and a link outperforms twenty of your own posts for entity authority.
Association guidance pages
Associations of international schools publish member guidance and are short of contributors on this topic. Offer to write it for free. Their domain has authority yours will not have for years.
Podcasts for school leaders
Show notes carry links, transcripts get crawled, and the format suits a topic that needs explaining rather than demonstrating. Pitch the delay angle — it makes a good hook.
Conference sessions with published slides
Publish the deck and a written summary on your own domain afterwards, and ask the organiser to link it. Two placements from one talk.
Your PhD and lecturing footprint
Academic and institutional pages carry weight and persist. Make sure every bio uses the same 25-word description and links the same domain.
Wikipedia and open references
Do not edit entries about yourself. Do improve the accuracy of AI Act and AI literacy articles with properly sourced material — it is legitimate contribution, and those pages are heavily drawn on.
LinkedIn — a thirty-day campaign
One platform. School leaders are on LinkedIn and effectively nowhere else professionally. Posting to four networks badly is worse than posting to one well, and you do not have the hours for four.
Three a week
Tuesday, Wednesday, Thursday, morning CET. Sustainable for a term. Missing a week is worse than posting twice.
5 : 3 : 2
Five parts useful, three parts opinion, two parts proof. Never more than one in ten posts asking for anything.
Text and carousels
Text posts and document carousels outperform links. Put the link in the first comment and say so in the post.
Reply to every comment within an hour
Engagement in the first hour drives reach, and school leaders comment thoughtfully. Each reply is a conversation with a prospect happening in public where their peers can see it.
Links in the post body
Put them in the first comment. Also avoid engagement bait, "thoughts?" endings, and posting the same thing three ways. This audience is unusually allergic to marketing tone.
The repurposing engine
You cannot generate original material three times a week alongside four jobs. You do not need to. Everything you have already built is source material.
| One asset you already have | Becomes |
|---|---|
| One instrument section | A cluster post, a LinkedIn carousel, a newsletter section, and an FAQ schema entry. Four outputs, one source. |
| One classroom activity | A LinkedIn post for teachers, a slide in a conference talk, a section in a parent-facing piece. |
| One objection from Vol. 01 | A post, a cluster page, and an FAQ answer. The six objections are six pieces of content that are guaranteed to land, because real buyers said them. |
| One delivery day | A behind-the-scenes post, a case study, three "what surprised me" observations, and a list of real questions that becomes next quarter's content plan. |
| One regulatory change | An update to the pillar, a post, an email to every past client, and a fresh dateModified that does real AEO work. |
| One scenario card | A poll post. "AI-drafted report comments — red, amber or green?" The contested cards were designed to split a room; they split a comment section the same way. |
The weekly hour
One hour, once a week: take one existing asset, produce the week's three posts from it, schedule them, and note anything that needs to change in the pillar. That is the entire ongoing marketing commitment once the twelve pieces are published. Anything more elaborate will not survive a term of teaching.
Measurement
Rank is the wrong metric. Grade this on Google position and you will conclude it is failing while citations climb in ChatGPT, Perplexity and Gemini. Track share of answer instead.
| Metric | Cadence | How, and what good looks like at six months |
|---|---|---|
| Share of answerThe primary metric | Monthly | 25 prompts × 4 engines. Count where you are cited. Cited in a quarter of them by month six is a strong position in a niche this narrow. |
| Competitor citations | Monthly | Who else appears. Early on it will be EUR-Lex, the Commission, and big consultancies. Displacing a consultancy on a school-specific prompt is the win. |
| Referral traffic from AI surfaces | Monthly | Small numbers, high quality. Watch time-on-page and the instrument completion rate rather than volume. |
| Instrument completions | Weekly | The clearest intent signal you have. Someone who works through 24 questions is a live prospect even if they never fill in a form. |
| Organic rank on the eleven phrases | Monthly | Secondary, but it feeds AI Overviews, so it still matters. |
| Discovery calls heldThe only metric that pays | Weekly | Everything above is a leading indicator of this. Four a week by week nine (Vol. 01 Part 13). |
Be your own customer first
Before any tooling, query the engines yourself and note who gets cited. Manual checking works fine at twenty-five prompts and it teaches you what the answers actually say, which no dashboard will. Only automate once the manual version is genuinely too slow.
What not to do
Four failure modes and two outright scams.
"Guaranteed inclusion in ChatGPT"
Impossible. No external party has administrative access to insert a brand into model weights or guarantee an output. Anyone offering this is taking money for nothing.
"Submit your site to AI search engines"
There is no submission process. Models discover businesses through crawling and indexing of public data; there is no paid AI listing programme.
Treating AEO as separate from SEO
They compound. A page that ranks organically is substantially more likely to be cited in an AI Overview. Do both or neither works well.
Publish and forget
Content age is tracked and this topic changes. An unmaintained pillar decays fast, and in a moving-target subject a stale page is worse than none because it can be wrong.
Optimising the homepage
Citations come from deep pages answering specific prompts, not the front door. Apply the answer-block discipline to every cluster page.
Keyword stuffing instead of entities
Repeating "AI Act compliance schools" does nothing. Naming Article 4, Annex III, the AILit framework and the Digital Omnibus, accurately, does everything.
The one rule underneath all of it
Everything in this volume works because the underlying material is genuinely correct, genuinely specific, and genuinely maintained. There is no version of this that succeeds with thin content and clever markup — the markup only helps a model find something worth quoting. Get the substance right and the technique compounds. Get it wrong and the technique accelerates a bad reputation in a sector where everyone talks to everyone.