GEO/AEO in Brussels — Be Cited Where EU Decision-Makers Search

#GEO/AEO in Brussels — Be Cited Where EU Decision-Makers Search

Nehos optimises AI citability in ChatGPT, Perplexity & Google AIO for Brussels: EU institutions, sector federations, Belgian FinTech. EN/FR/NL.

#The briefing note now has a first draft nobody commissioned

Watch how a file actually gets prepared in this city today. A parliamentary assistant with four hours before a committee meeting asks an answer engine to explain the state of play on a dossier. A journalist checking which industry bodies opposed a measure asks the same way. A corporate EU affairs director joining a new portfolio asks for the landscape of stakeholders before asking a colleague. In every case a synthesis arrives first, with a handful of organisations named in it, and the human work starts from that synthesis rather than from a blank page.

That first synthesis is where framing is set, and it is decided by a shortlist. Answer engines do not produce a ranked list of twenty results; they name two or three sources and move on. Being the fourth most authoritative voice on a file used to mean a smaller share of attention. In a generated answer it frequently means no mention at all. For organisations whose entire function is to be present in a debate, that is a structural change, and it is not solved by the search visibility work of the last decade.

The uncomfortable part for Brussels specifically is that the machinery of influence here was built for a different retrieval system. Position papers are written to be defensible in a meeting, not extractable by software. Reports are published as PDFs because that is what a board approves. The most quotable version of your organisation's view often sits behind a trade-press paywall, which makes it excellent for reputation among the two thousand people who subscribe and invisible to a model that cannot retrieve it. None of this was a mistake — it is simply optimised for a channel that is no longer the only one.

#Two mechanisms that get confused constantly

Almost every conversation we have about generative visibility conflates two entirely different things, and the distinction determines what work is worth doing.

The first mechanism is what a model absorbed during training. It is slow, opaque, impossible to audit and impossible to update on demand. Nobody can sell you a reliable intervention on it, and anyone who claims to is selling you a story.

The second is live retrieval. When an answer engine grounds a response — ChatGPT with search enabled, Perplexity, Google's AI Overviews, Copilot — it fetches documents at query time from an index and writes from what it fetched. This mechanism is observable, testable, and responsive to work you actually control: whether your content is crawlable, whether it is structured well enough to be extracted, whether it is dated, whether the claim a model needs exists as a self-contained statement somewhere it can reach.

Almost all defensible GEO work targets the second mechanism. Which leads to a check we run in the first week of every engagement and which fails more often than you would expect: the crawler controls. Organisations block GPTBot, PerplexityBot, ClaudeBot or Google-Extended in robots.txt — sometimes as a deliberate policy on training use, more often because a supplier added a blanket rule years ago — and then commission a project to increase their presence in AI answers. Those two positions are not compatible, and the trade-off deserves a documented decision at leadership level rather than a default set by someone's WordPress plugin.

#Where Brussels content loses citability

The PDF wall. A ninety-page report published only as a PDF is difficult for retrieval systems to parse cleanly and easy for them to skip. The fix is not to abandon the PDF, which serves its own purpose, but to publish a properly structured HTML version with headings, a substantive summary, and the key figures in text rather than in an image of a chart.

The undated page. Policy questions are time-sensitive and grounding systems discount content whose currency they cannot establish. A position page with no visible publication or revision date competes badly against a think tank note published last week, regardless of quality.

The claim buried in prose. Models extract statements, not arguments. If your organisation's position on a proposal exists only as the conclusion of six paragraphs of context, there is nothing clean to lift. Stating the position in a single sentence near the top, then arguing it below, costs nothing rhetorically and changes extractability completely.

The unattributed number. A figure with no source, no methodology and no date is exactly the kind of content a grounding system is designed to avoid repeating. A figure with a named source, a year and a link is the kind it will happily quote — with your name attached.

Entity ambiguity. Federations and institutes in this city have long formal names, French and Dutch variants, an acronym everyone actually uses, and often a similarly named counterpart in another Member State. If a model cannot resolve which entity is which, it will attribute your work to someone else. Consistent naming, an explicit relationship between the acronym and the legal name, structured organisation markup, and coherent presence in the reference sources used for entity disambiguation all reduce that risk.

Access strategy working against you. Members-only publication makes sense for some material and is self-defeating for the material you most want quoted. Deciding deliberately which tier of content is open is a communications decision with direct visibility consequences.

#How we run a GEO programme here

Baseline measurement first. We build a query set that reflects how your audiences actually ask — not brand queries, but the questions a policy officer, a journalist or a corporate affairs lead would type about the files you work on. We run them across the major answer engines in English, French and Dutch, repeatedly, and record who gets cited, in what role, and how accurately you are described when you appear. Generated answers vary between runs, so a single sample is worthless; we treat it as sampling, with repetition, and report a rate rather than an anecdote.

Publication architecture. We restructure how your outputs reach the web: HTML-first for anything intended to influence, clean heading hierarchies, summaries that stand alone, figures in text, stable URLs, visible dates, and named authors with real credentials attached. Structured data does the disambiguation work — organisation identity, publication metadata, question-and-answer blocks where the format genuinely fits.

Claim-level rewriting. We work through your position papers and briefings with your policy team, restructuring them so each key claim is self-contained, sourced and dated. The substance does not change — that is your business, not ours. The form becomes liftable.

Third-party presence. Your own site is not the only source models retrieve. Public consultation responses, joint statements, open-access research, standards bodies, statistical portals and the trade press all feed grounded answers, and their relative retrievability differs enormously. Open-access publication of a submission you were going to file anyway is often the single highest-leverage action available.

Trilingual coverage, honestly prioritised. Authority does not transfer across languages. An organisation cited constantly in English on an EU file can be entirely absent from Dutch-language answers about the Belgian dimension of the same file — and Dutch is usually the least contested space, which makes it the cheapest ground to take. French sits in between. We measure the three separately because they behave as three different markets.

The Belgian financial angle is a different game. For a fintech or a regulated financial player, the valuable queries are not policy questions but selection questions — who to use for a given payment flow, which provider handles a specific compliance requirement, how a Belgian rule applies in practice. Answer engines are noticeably more conservative on financial topics and lean on sources that carry visible institutional weight: supervisory publications, sector federation material, established professional media. The work there is less about volume of content and more about producing precise, verifiable, well-sourced explanations of the specific regulatory and operational questions your prospects actually ask, and about keeping every claim within what a supervised entity is permitted to say.

#What the first quarter looks like

Weeks one to three are measurement and diagnosis: the query set is built with your team, the baseline is sampled across engines and languages, the technical retrievability audit runs, and the crawler-control question gets a decision from someone empowered to make it. You end that phase with a number you did not have and usually with two or three surprises.

Weeks four to eight are structural: publication architecture, entity consistency, structured data, and the reworking of the highest-value existing documents. This is where most of the durable gain comes from, and it is unglamorous editorial and technical work rather than anything resembling a growth hack.

From week nine onward the programme becomes a rhythm: each new publication ships in the citable format by default, the query set is re-sampled on a fixed cadence, and the measurement feeds back into what gets published next. Answer engines change their behaviour continuously, so the deliverable is a maintained capability inside your organisation, not a report that ages.

#Measuring something real

GEO cannot be measured in ranking positions, and any report that shows you a rank has misunderstood the medium. We instrument four things from the outset: citation rate, the share of sampled queries where you appear at all; share of voice, your presence relative to the named alternatives in your space; representation accuracy, whether what the model says about you is correct — a wrong attribution is worse than absence; and sentiment and framing, how your role is characterised. Each is tracked per engine and per language, because they move independently.

Two cautions we give every client. Generated answers are volatile week to week, so short-horizon comparisons produce noise that looks like signal. And there is no ranking guarantee available in this discipline from anyone — the systems are not built to offer one. What is available is a defensible, measured programme of making your organisation the most retrievable and most citable source on the questions you care about.

#What we do not do

We do not generate content at volume to flood a topic. Grounding systems are not counting pages, and a corpus of synthetic material damages the source authority you are trying to build.

We do not buy or seed citations. Beyond being ineffective at any scale that matters, in a city where the Transparency Register exists and influence practices are scrutinised, a manufactured citation network is a reputational risk far exceeding any visibility gain.

We do not build a strategy around llms.txt or any similar convention. It is a proposal, not an adopted standard with confirmed uptake from the major engines, and treating it as the centrepiece of a programme substitutes a file for the actual work.

And we do not optimise for a single engine. Their retrieval behaviour and their user bases differ, and a programme tuned to one of them is a bet on a market share that shifts every quarter.

#Why Nehos in Brussels

• On the ground — Securex building, Cours Saint Michel 30A, 1040 Brussels. We sit with your policy and communications teams rather than emailing recommendations.
• We know the institutional register — the difference between a position paper, a consultation response and an expert report, what each is worth to a retrieval system, and which one you should be publishing openly.
• Trilingual measurement — English, French and Dutch tracked as three distinct markets, with Belgian usage handled properly.
• 47 specialists in AI, data and cloud; 200+ projects in production since 2014. We build the measurement infrastructure ourselves rather than reselling a dashboard.
• Method over manipulation — editorial restructuring, technical retrievability and honest measurement. No content farming, no purchased mentions.
• Free audit — 30 minutes to establish where you currently stand and whether your own robots.txt is working against you.

#Complementary services in Brussels

• our national GEO/AEO service
• the Nehos Brussels office
• our AI agency practice in Brussels
• AI agents for Brussels public affairs teams
• Next.js platforms for institutions and Belgian FinTech
• our client success stories

Related areas we cover: generative engine optimisation Brussels, ChatGPT citability Brussels, AI Overviews Brussels, answer engine optimisation Belgium, LLM visibility EU institutions.

The FAQ below answers the questions our Brussels clients ask most often on this topic.

Get in touch with our Brussels team for an initial conversation with no strings attached: we assess the potential of your project together and give you a costed estimate of the expected ROI.