GEO for Nonprofits: How a Charity Gets Found by AI

Why a charity is nearly invisible to AI

When someone asks an AI assistant who works on a particular cause in a given area, the model doesn't browse the web like a search engine: it composes the response from what it finds already structured and verifiable. Most nonprofits and voluntary organisations aren't. Their mission lives in annual reports, impact PDFs, brochures, and social posts — formats a model struggles to read and almost never cites.

The result is twofold. Sometimes the organisation is simply omitted, as if it didn't exist. Other times the model fills the gaps with guesses: activities it doesn't carry out, wrong data, or confusion with another similarly-named group. For an organisation that lives on trust, that's the worst possible starting point.

There's a structural reason for this, not a failure. In the voluntary sector, communications are almost always the last budget line, websites are often managed by volunteers, and the dominant culture is the document — the PDF, the brochure, the press release. These formats are designed for people, not machines. So even a serious, active organisation that produces a great deal ends up data-poor in the eyes of an AI. The gap between what it does and what a model can see is enormous — and that gap is exactly the problem to close.

A concrete example

A youth charity runs after-school support and educational activities in a local borough. I ask an AI assistant who works on educational support for young people in that area: if the charity exists only through a Facebook page and a few local news mentions, the response ignores it — or cites larger, more structured organisations, even if they're less relevant or further away.

Then I try asking directly about the charity, by name. The model, finding no clear data, fills in: it assigns a wrong location, confuses its activities with those of a similar-sounding group, or invents a founding date. Nothing malicious — just the normal behaviour of a system filling gaps. The problem is that whoever reads the response can't tell data from guesswork.

What changes when a funder or programme officer asks AI

Networks, foundations, and grant-making bodies increasingly use AI tools to map the sector: who's active on a theme, in which area, with what results. A donor looking for causes to support does the same. Appearing correctly in that mapping means being considered for opportunities you'd otherwise never hear about.

It's not about appearing larger than you are. It's the opposite: it's allowing whoever is looking for exactly your activity, in your area, to find you instead of nobody. Invisibility doesn't protect you — it quietly excludes you.

The questions that matter look like this: who works on employment inclusion in this city, which organisations run programmes for young people in this region, who has experience in community regeneration in this area. A programme officer or assessor might put these to an AI assistant to get a quick picture before opening a call for applications. The response shapes who gets contacted — and who stays off the radar.

The problem isn't how much content you produce, it's how it's structured

Many organisations publish more than people realise: impact reports, annual reviews, accounts. The problem isn't quantity, it's form. A PDF report is transparency that no machine truly reads: it exists, but remains inaccessible, and for an AI system it might as well not be there.

Structuring that data doesn't mean writing new content. It means making it queryable: declaring explicitly what the organisation does, where, with whom, and with what results — in a format machines read without having to interpret. It's curatorial work before it's technical. And it applies to nonprofits just as much as to anyone else who wants to be cited by an AI system.

What data to expose, in practice

The practical question is: what needs to be made readable? For a nonprofit, the data that matters is what identifies it and qualifies it unambiguously.

• Exact legal name and legal form: charity, CIO, community interest company, voluntary organisation, foundation, or benefit corporation. These are distinct entities, and declaring it prevents a model from treating the organisation like any commercial business.
• Formal identifiers: charity registration number, company number where applicable, and any regulatory registrations. These are the anchors that disambiguate the organisation from similarly-named groups and tie it to a verifiable existence.
• Areas and sectors of activity, linked where possible to shared vocabularies, and the territory where the organisation genuinely operates.
• People and governance, activities and projects with their dates, and any grants, partnerships, or recognitions that evidence the work.

Translated into Schema.org markup (NGO type), typing the organisation as Organization or NGO, these data points stop being text to interpret and become declared facts. That's the difference between a model that guesses and one that reads.

Schema.org for nonprofits: the types that matter

There's a technical level worth making explicit, because this is where many generic interventions fail. Schema.org, the vocabulary that search engines and AI systems use to recognise entities, has specific types and properties for nonprofit organisations — and using them well changes how a charity is interpreted.

The base type is Organization, but for a nonprofit it matters to go further. The nonprofitStatus property lets you declare the non-profit nature; founder and foundingDate fix origin and founders; areaServed defines the territory served; knowsAbout qualifies the areas of activity; funder and sponsor make supporters explicit; member and memberOf describe membership of networks and coalitions. Press coverage and recognition connect via subjectOf, which links the organisation to articles about it.

This isn't completeness for its own sake. Every verifiable property populated is one more anchor that reduces ambiguity: the less room you leave for interpretation, the less the model invents. Choosing which fields to populate, and with which sources, is exactly the work a plugin cannot do.

Three interventions that matter

Three moves make the difference, in order.

• Structure the identity. Mission, activities, and governance declared with Schema.org markup, typing the organisation as Organization or NGO, with JSON-LD on the official website. This is the foundation that makes the organisation readable.
• Link to verifiable references. Anchor the entity to existing entries for the themes, networks, funders, and territories it's connected to. These links are what distinguish you from similarly-named groups and give meaning to what you do.
• Make transparency queryable. Transform impact reports and activity data from static documents into structured, citable information.

Three mistakes I see often

Three recurring assumptions waste time and visibility.

• Thinking social media presence is enough. An active profile doesn't make the organisation readable: social content is hard for a model to interpret and rarely becomes a citable source. It's motion, not structure.
• Relying on a generic SEO plugin. An automated tool adds standard markup, but it doesn't know what distinguishes a CIO from a foundation, or which data of a voluntary organisation deserves to be highlighted. The markup comes out; the meaning doesn't.
• Rushing to create a Wikidata entry. Without verifiable notability, the entry gets deleted — and the attempt backfires. Build the anchor to existing references first; a dedicated entry, if warranted, comes after and only when the conditions exist.

Wikidata, open registers, and the myth of notability

The Wikidata question comes up because it's the open reference most cited by AI systems. The mistake is assuming you need your own entry. In most cases the value lies elsewhere: in connecting the organisation's entity to entries that already exist. The local authority it works in, the cause it addresses, the network it's part of, the funder that supports it — all verifiable nodes to anchor yourself to. Those links tell a model how to place you.

Alongside Wikidata, official registers serve as authority sources: in the UK, Charity Commission registration is a proof of existence and legal form worth making explicit and linkable; in other anglophone contexts, similar registers apply. The more official the anchors, the more the identity holds.

There's a widespread misconception: to be readable by AI, you need to be notable enough for a Wikidata entry. That's not true. A dedicated entry makes sense only when an organisation has verifiable notability — press coverage, awards. For all others, the work is connecting the entity to existing open references, without forcing anything that would get removed.

It's precisely small organisations that gain most from this work. They have less online presence, so they're the ones AI systems struggle most to see. Making them readable closes exactly that gap — and it can be done with limited resources. The value lies in the curatorial work, not the technical budget.

How to measure the result

This kind of work needs measuring — otherwise it remains a promise. The method is simple and repeatable, done with the same tools your audience uses.

Before intervening, I take a baseline: I put a set of relevant questions to the main AI systems and note the responses — by organisation name, by area of activity, by territory. For example: what does a certain charity do, who works on a cause in a given place, which organisations have run a type of project. This is the starting point.

After the intervention I repeat the same questions and compare. Is the organisation mentioned where it was absent? Is the data correct? Has confusion with similarly-named groups disappeared? Are the activities cited the real ones, not invented ones? It's a verifiable comparison, worth repeating over time, because models update and the picture needs refreshing.

A realistic path for organisations with limited resources

The voluntary sector works with tight budgets, and the work needs to be sized accordingly. The good news is that the value lies in the decisions, not the technical spend — and those can be made in stages.

A sensible path starts with the essentials: structuring the organisation's entity on the official website, with correct identifying data and legal form. That's the foundation, and it alone eliminates most of the confusion. The next step is linking: connecting the entity to open references that already exist, without creating anything new. Then one flagship piece of content — usually the impact report or activities page — is structured from document to readable information. Finally, monitor once or twice a year.

Each step makes sense on its own and prepares the next. There's no need to do everything at once; what matters is starting from the foundation and not skipping the basics.

Why it makes sense to move now

A fair question is why to act now rather than in a few years once everything is clearer. The answer is that the advantage is built precisely in the early phase.

Searches via AI assistants are growing quickly, including among programme officers, journalists, and networks mapping the sector. But the voluntary sector is almost entirely still unreadable to these systems — which means the space is open. Whoever structures their entity now becomes, for the model, one of the few clear reference points on a cause and a territory. And anchors to verifiable sources don't wear out: they accumulate and strengthen over time. Moving early means being the already-clean entity when everyone else starts catching up.

It's not a race against the clock — it's compounding interest. The earlier you start, the longer that work works for you.

Where to start

Three concrete steps, in order:

• Visibility audit. Ask ChatGPT, Perplexity, and Gemini the questions a donor or funder would ask about your organisation. Note what appears, what's wrong, and what's missing.
• Structure the identity. Legal name, legal form (charity, CIO, voluntary organisation, foundation, benefit corporation), and core activities in Schema.org markup on the official website. This is the foundation that makes the organisation readable.
• Link and monitor. Anchor the entity to existing open references (charity register, partner organisations, topics of activity on Wikidata) and repeat the baseline review every six months.