llms.txt, Schema Markup, and AI Crawlers: A Technical Checklist for Nonprofit Websites

To make your site legible to AI, it helps to understand the two very different jobs happening behind the scenes. The first is crawling, which is when automated programs, called crawlers or bots, visit web pages and copy their content so it can be used later. The second is answering, which is when a tool like ChatGPT or Perplexity assembles a response to a user's question, sometimes drawing on what was crawled earlier and sometimes fetching a live page in the moment.

There are broadly three ways an AI system might encounter your content. Some crawlers gather large amounts of text to help train future models. Some fetch a page in real time because a user asked a question the assistant wants to answer with current information. And some answer engines lean heavily on traditional search indexes, layering an AI summary on top of results a conventional search engine already ranked. Each pathway rewards the same underlying quality: content that is easy for a machine to locate, parse, and trust.

The critical thing to grasp is that these systems do not see your website the way a person does. They do not admire your photography or your color palette. They read the underlying code and text. If your key facts, your mission, your services, your contact details, live only inside an image, a video, or a script that loads content after the fact, a crawler may never register them. Everything in this checklist exists to move the information that matters out of the decorative layer and into a form machines can reliably read.

Schema markup, also called structured data, is the single most powerful technical step most nonprofit sites are missing. Here is the idea in plain terms. When a person reads your homepage, they understand from context that Helping Hands is a charity, that a phone number is a phone number, and that a paragraph is describing your mission. A machine does not have that context automatically. Schema markup is a small, invisible layer of labeled information you add to your pages that spells out, in a vocabulary machines already understand, exactly what each piece of information means.

The shared vocabulary comes from Schema.org, a project supported by the major search companies. It is usually added to a page in a format called JSON-LD, a block of labeled data that sits quietly in the page code without changing how the page looks to visitors. A web developer can add it in an afternoon, and many website platforms and plugins can generate it for you. The payoff is that AI systems and search engines can read your identity directly rather than guessing at it.

A minimal example for a nonprofit's organization details is short and readable. The block below labels the organization as a nonprofit, states its name and mission, and provides its official identifier and contact point. This is the kind of information you want machines to have exactly right.

You do not need every schema type, only the ones that match the content you actually publish. The cards below cover the types most relevant to nonprofits. Add the ones that fit, and skip the rest. Getting a few correct is far better than sprinkling many that are inaccurate, because incorrect structured data erodes the trust these signals are meant to build.

The files above sit on top of a foundation, and the foundation matters more than any single trick. If the underlying page is messy, slow, or inconsistent, no amount of markup will fully compensate. Four foundational habits do most of the work, and none of them are exotic. They are simply the marks of a well-built website that happens to be exactly what machines need too.

The first is clean, semantic HTML. Semantic simply means using the right structural element for the right job: a real heading marked as a heading, a real list marked as a list, one clear main heading per page followed by an orderly hierarchy of subheadings. When your page has a logical outline that a machine can follow, it can extract your key points reliably. When everything is a visually styled block with no structural meaning, machines have to guess, and they often guess wrong. Alongside this, put your most important facts, your mission, services, contact details, and tax identifier, into plain readable text rather than locking them inside images or videos.

The second is speed and accessibility. Pages that load quickly and work on modest devices are easier for crawlers to process, and the same practices that make a site usable for people with disabilities, descriptive text for images, clear link labels, sensible reading order, also make it far more legible to machines. Accessibility and machine readability are two sides of one coin, which means the work you do for one audience quietly serves the other.

The third is consistency across the wider web. AI systems build confidence in a fact when they see it stated the same way in several places. If your organization's name, address, phone number, and mission match across your own site, your listing on charity databases, your social profiles, and any directories you appear in, machines treat that information as more trustworthy. Contradictions, an old address here, a slightly different name there, create doubt that can push you out of an answer.