Why AI Cites Original Numbers Over Explainer Pages

There is a page on your firm's website explaining what an H-1B is. There are several thousand others. When a prospective client opens ChatGPT and asks how the cap-exempt route works, the assistant does not pick the best-written of those thousand. It picks the one that gave it something specific enough to quote.

That distinction has quietly become the whole game. Research published through 2025 and into 2026 converges on the same finding: AI assistants reward pages that introduce a fact, not pages that restate one. Zyppy's citation analysis found ChatGPT cites only around 15% of the pages it actually retrieves — the other 85% are read and discarded. Controlled GEO benchmark testing found that adding statistics to a page lifted its visibility in AI answers by roughly 31%, and adding direct quotations by roughly 41%. Those are the two largest measured levers in the literature, and both describe the same thing: give the model a sentence worth lifting.

For a small immigration practice, this is better news than it sounds.

Ranking is no longer the gate

The old objection to any visibility conversation was that a four-attorney firm cannot outrank Boundless, Nolo, or a national firm with a content team. In AI answers, that constraint has largely dissolved.

Ahrefs found that only about 10% of URLs cited by ChatGPT sit in Google's top 10 results. seoClarity found that 25% of top-cited URLs have zero Google organic visibility at all — pages that never won a search result are winning citations. And Profound's analysis of roughly 730,000 conversations found the top 10 domains account for just 12% of all citations. There is no oligopoly to break into. The long tail is where most citations live.

What replaced ranking as the gate is quotability. A model retrieving twenty pages on cap-exempt H-1B petitions needs one that contains a claim it can attribute. Nineteen of those pages say the same thing in different words. The twentieth says: "Across 214 cap-exempt petitions we filed for nonprofit research employers between January 2024 and March 2026, 71% cleared without an RFE." That is the sentence that gets carried into the answer, with your firm's name attached to it.

Your practice already generates the data

The instinct is to assume original research means a survey team and a budget. It doesn't. It means counting something you already do and publishing the count with a date on it.

Three examples that any firm with a case management system can produce in an afternoon:

Outcome rates from your own docket. How many of your last 50 RFE responses were approved? What was your average time from filing to decision for adjustment of status cases in your service center last year? These are numbers no explainer page can produce, because they require having done the work.

A fee benchmark for your market. What does a family petition actually cost in your city, across the firms a client would realistically call? Publish the range, name the sample, date it. Prospective clients search this constantly, and almost nobody publishes a defensible figure.

What your intake actually tells you. Across your last 100 consultations, what did people ask about first? What did they get wrong about priority dates or work authorization? A counted answer to that question is more useful — and far more citable — than a general anxiety-management post.

None of these require you to be a researcher. They require you to be a practitioner who writes down what the practice already knows.

Where the number goes matters as much as the number

Zyppy's positional analysis found that 44.2% of ChatGPT's citations come from the first 30% of a page, 31.1% from the middle, and 24.7% from the final third. A finding buried in your conclusion is worth roughly half a finding stated in your opening.

So the structural rules are unglamorous and specific. Put the figure in the first two sentences. Attach a date and a sample size — "across 214 petitions filed between January 2024 and March 2026" is retrievable in a way that "in our experience" is not. Write one self-contained sentence that carries the claim without needing the paragraph around it, because that is the unit the model actually extracts. Then explain.

The same research found pages with FAQ schema and inline citations are weighted meaningfully higher in source selection. That is a half-hour task for whoever maintains your site, and it compounds across every page you have already published.

What to do this week

1. Pick one number you can defend. Pull it from your case management system, your intake log, or your billing records. Outcome rate, timeline, fee range, or intake pattern — one is enough to start.
2. Establish the denominator before you publish. "71% of 214 petitions" is citable. "Most of our petitions" is not. If the sample is small, say so; a stated n=40 is more credible than an unstated one.
3. Write a 600-word page around it, figure first. Lead sentence carries the number, the date range, and the sample. The rest of the page explains method and what it means for the reader.
4. Confirm your site is fetchable. Allow `OAI-SearchBot` in robots.txt — OpenAI's publisher documentation names this as the single official visibility lever, and Cloudflare has blocked AI crawlers by default since July 2025. Server-render the page; crawlers largely do not execute JavaScript.
5. Add FAQ schema and inline source links. Low effort, measurable weighting effect.
6. Re-measure in 30 days. Citation mixes shift fast — Reddit's share of ChatGPT responses swung from roughly 60% to 10% inside a few weeks in late 2025. Treat any snapshot, including this one, as perishable.

The wider shift

For twenty years, legal marketing rewarded whoever explained the law most thoroughly. That produced an enormous volume of nearly identical content, and AI assistants now treat that volume as interchangeable — because it is. The scarce thing is no longer explanation. It is evidence.

That inverts the usual advantage. A national content operation can out-publish you on explainer pages forever. It cannot publish your docket. The 214 petitions you filed, the 100 consultations you took, the fee range in your city — those exist in exactly one place, and it is your firm.

Publish one of them, with a date and a denominator, and you stop competing to be the tenth restatement of the law. You become the source the answer names.

Clientory helps immigration law firms become visible to the families and individuals who need them most — turning AI search into consultations, and consultations into clients.

See where your firm stands: get your AI Visibility Score at CLIENTORY.ORG