AI visibility (AEO) for organizations.
Answer engine optimization. When someone asks an assistant who does what you do, there is an answer. You are in it or you are not — in English and in Spanish.
Who should we fund for education equity work in North Carolina?
Three names come back: a statewide education trust, a university-affiliated college access program, and the regional office of a national nonprofit. All three publish outcome data and have been quoted in state education coverage.
Your organization was not among them.Illustrative. The organizations left out do the same work — they are just harder for a model to confirm.
The question gets asked whether or not you are in the room.
Search used to hand people a list and let them choose. Assistants hand people one answer, already narrowed. Being on page one of a list you were never shown is not a position you can recover from.
These are not hypothetical queries. They are the shape of what people now type instead of opening a search page and clicking through five tabs.
A program officer, on first-pass diligence
We are finalizing a shortlist — who is doing the strongest work on this in the Southeast?
Before anyone picks up a phone, and before you know you were considered.
A mother, at eleven at night, on a phone
¿Dónde puedo encontrar ayuda legal para inmigrantes cerca de mí?
In the language she thinks in. She will take the first answer she gets.
A reporter on deadline
Who could speak to this for a story I am writing?
Whoever gets named gets quoted, and being quoted is what gets you named next time.
Three things decide whether you are in the answer.
Most of what is sold as AI optimization is aimed at the wrong thing. You cannot influence what a model learned during training on any timeline that matters. What you can influence is retrieval: when the assistant runs a search before answering, whether it finds you, and whether what it finds is clear enough to repeat.
That comes down to three jobs. They get harder and more valuable as you go down.
Can it tell you are one organization?
Most organizations are several entities as far as a model is concerned: a legal name, a program name, an old acronym, three URLs, a directory listing with a dead address. Resolving that into one identity, with one canonical address and machine-readable markup pointing at it, is the floor. It is also the fastest thing to fix.
Can it lift a fact from your site?
Models summarize. Copy written to sound inspiring summarizes to nothing, because there is nothing in it to carry. Who you serve, where, how many, since when, in what languages, at what cost. Stated plainly, in text, on a page. Vague language is not a style problem here; it is the reason you are not quotable.
Does anyone else confirm it?
This is the one that decides most answers, and the one almost nobody sells, because it is real work rather than a setting. A model weighs what other sources say about you far more heavily than what you say about yourself. Funder pages, coalition listings, Candid and 990 data, local press, partner sites, conference programs. Your own website is the weakest available evidence about your own organization.
The hardest of the three, and the reason the other two are not enough on their ownAsk it in Spanish and the answer gets thinner.
Where can immigrant families get legal help in eastern North Carolina?
A full paragraph naming several organizations, with counties served, what each one handles, and two links.
¿Dónde pueden recibir ayuda legal las familias inmigrantes en el este de Carolina del Norte?
A shorter, hedged answer. One national hotline. A suggestion to search locally.
The organization did not change between those two answers. The evidence available in Spanish did. There is far less Spanish-language material for a model to retrieve, far fewer Spanish pages that state anything plainly, and almost nobody has done this work in Spanish yet. For an organization whose people ask in Spanish, that gap is the whole problem — and it is the least contested ground in this field.
Measure it, fix it, measure it again.
We start with a baseline. A set of questions your funders, your press, and the people you serve are actually asking, run against the assistants that answer them, in English and Spanish. Where you appear, where you do not, what is said about you, what is wrong, and who gets named instead.
Then we fix what the baseline found. The identity layer and the plain-statement layer are work we do on your site directly. The corroboration layer is slower and needs your team: it is a plan for where you should appear that is not your own website, and how to get there.
Then we measure again. Model answers drift week to week. A single audit is a photograph of something that moves. Quarterly re-measurement is what makes this a position rather than a project.
Two honest notes, because this field is full of people who will not say them. Nobody can guarantee placement in a model’s answer; anyone promising that is selling something they cannot deliver. And llms.txt, which you will see recommended everywhere, is almost never fetched — we will put one on your site because it costs nothing, but it is not the work and we will not bill it as such.
Who should we fund for education equity work in North Carolina?
Four organizations, including yours — with the counties you serve, the number of students you reached last year, and the fact that you work in English and Spanish.
The same question, after the work. The difference is what a model could confirm.
What does AI say about your organization?
Get your free AEO check in 24 hours.
What comes back
- Your real answers, from the assistants people actually use
- English and Spanish, side by side
- Who gets named instead of you, and why
- The three or four fixes that would move it most