How to structure a coalition position page so AI assistants cite it

If a coalition wants AI assistants to cite its position pages, the structure matters more than the prose. State your central claim and the single strongest piece of evidence behind it in the first 150 to 200 words, then make one claim per section, each carried by a sourced figure, under plain-language headings that match how the question is actually asked. Add a visible references layer, keep the whole page as clean server-rendered HTML rather than burying the substance in a PDF, and name the coalition, its members and the issue identically everywhere. An engine weighs the top of the page most heavily and lifts the cleanest passage that answers the question; a page built this way gives it one to lift. Here is how each part works, and why.

What question is the engine actually answering?

Before you structure a page, be clear about what the engine is doing when it reaches it. It is not reading the way a member or a supporter reads. It has been asked a specific question — “what does the coalition argue about this rule?”, “how many jobs does this sector support?”, “who opposes this tariff?” — and it is scanning for the cleanest passage that answers it well enough to quote. Your page is one candidate among many. It will be judged on how directly it answers, how trustworthy it looks, and how easily the relevant passage can be isolated and lifted.

That reframes the job. A position page is not a brochure for the cause; it is a candidate source for a question. Every structural decision below follows from writing for that moment.

Why put the answer first?

Lead with the answer because the engine weighs the opening of the page most heavily, and a buried answer reads as irrelevant. Put the claim and the evidence in the first 150 to 200 words — the position you hold and the single best figure that supports it — before any history, context or call to action.

This runs against instinct. Advocacy writing often builds to its point, setting the scene before stating the position. An engine does the opposite of a patient reader: it samples the top of the page, and if your stance is not there it frequently concludes the page is off-topic and moves to a source that gets to the point. The supporters scrolling for the full argument lose nothing — the context still follows. But the engine, and the staffer skimming on a phone outside a hearing room, both get the answer immediately. Answer-first is not dumbing-down; it is putting the conclusion where the reader who matters most will actually see it.

One claim per section, each with a sourced figure

After the opening, give every section a single claim and a single piece of evidence to carry it. Not five arguments bundled into a dense block, but one point, supported by one sourced figure, under its own heading.

The reason is mechanical. An engine isolating a passage to cite finds it far easier to lift a clean “X supports Y jobs, according to Z” than to disentangle a paragraph making five interwoven arguments at once. A figure with a named source — a government dataset, a regulator’s own impact assessment, an independent study — does double duty: it gives the engine something concrete to quote and it gives the claim authority the engine can stand behind. A bare assertion is easy to ignore; a sourced figure is hard to argue with. This is the same discipline good advocacy writing already uses: make a point, support it, move on.

Headings that match how the question is asked

Write headings as the plain-language questions people and engines actually use, not as internal slogans. “How many jobs does the sector support?” earns its place where “A thriving industry, a stronger nation” does not.

Engines map a query to the part of a page most likely to answer it, and clear question-shaped headings make that mapping obvious. They also help a human skim to the answer they came for. The test is simple: would a journalist, a congressional staffer or a regulator type something close to this heading into a search box or an assistant? If yes, the heading is doing its job. If it is a tagline, it is decorative — it may read well on a banner, but it gives neither reader nor engine a way in.

A references layer the engine can see

Give the page a visible references layer: the sources behind your figures, cited in the page itself, where both a reader and an engine can see them. This is the difference between a claim and a sourced claim, and engines treat the two very differently.

References do two things. They let an engine verify that a figure is genuinely supported, which raises the page’s standing as something safe to quote. And they connect your page to the wider web of authoritative material on the issue — government filings, regulators’ assessments, independent research, established reference sources and the knowledge graph engines lean on. A coalition that cites the serious sources, and is in turn cited by them, becomes part of the citable record on its issue rather than an isolated assertion floating on its own. Authority is not something you claim about yourself; it is something the surrounding sources confer, and a references layer is how you earn it.

Machine-readable structure, never locked in a PDF

Keep the substance in clean, server-rendered HTML — the headings, figures and references in the page itself — and never as evidence locked inside a PDF.

A position published only as a PDF is harder for engines to parse, slower to load and easy to overlook. Static-first architecture, with pages served quickly from the edge, gives an engine readable HTML with a clear heading structure and visible text it can lift directly. The formal filing still has its place: offer it as a download alongside the web page, for the people who want the document of record. But the argument, the figures and the sources should live as a normal web page first. If the only home for your evidence is a PDF, you have hidden it from precisely the systems you want to cite you.

Name everything the same way, everywhere

Decide on one name for the coalition, one for each member organisation and one for the issue, and use them consistently across every page, filing and reference source. This is the entity footprint, and engines build their model of who you are from that repetition.

Inconsistency dilutes the signal. If the coalition is one name on the homepage, a slight variant in a filing and a third version in the newsroom, an engine has to work harder to connect a citation to you — and may not connect it at all. Worse, it can attribute your argument to a near-namesake or split your authority across what looks like several entities. Pick the canonical name for each thing, use it everywhere, and you give the engine a single, coherent entity to attach trust and citations to.

A short before and after

Take a coalition arguing against a proposed tariff.

Before. The page opens with two paragraphs on the coalition’s founding and its values. The case against the tariff arrives in the fourth paragraph, woven into a single dense block that also covers history, fairness and competitiveness. The jobs figure is mentioned in passing, unsourced. The supporting study is a PDF linked at the foot of the page. The headings read “Our Story” and “Standing Together.”

After. The page opens: “The proposed tariff would put 42,000 jobs at risk across the sector, according to the regulator’s own impact assessment. The coalition opposes it on those grounds.” Each section that follows makes one claim — jobs, input costs, downstream prices — each with a sourced figure cited in a references block on the page. The headings read “How many jobs are at risk?” and “What would the tariff do to prices?” The full study is still downloadable, but its key findings live in the page as readable text.

Same facts, same position. The second version is the one an engine can read, trust and quote — and the one a staffer can act on in thirty seconds.

The pattern generalises. Whatever the issue, structure the page as a candidate answer: claim and evidence up front, one sourced point per section, question-shaped headings, visible references, clean HTML, consistent naming. Do that and your position page stops being a place that merely states your view and becomes the source an engine reaches for when the question is asked.

This is the companion to web design for Washington DC trade associations and coalitions and how to get your company mentioned in ChatGPT. For the mechanics underneath, see how LLMs decide what to say about brands, and for how we build and run these sites, Managed AI-Ready Websites.