ChatGPT Wanted Your Numbers. It Couldn't Read Them. So It Cited Someone Else's.
Two weeks ago we told you ChatGPT routes citations through at least four hidden retrieval pipelines, and 11.6% of repeat prompts flip their primary source between runs. That research has a second finding that matters more for what is actually on your website: even when ChatGPT wants to cite you first, it cannot always read what you have published. When it cannot, it moves on to whoever it can read.
Source: Search Engine Land: ChatGPT citations change when hidden search pipelines switch
Researchers tracking ChatGPT's network traffic, not just its outputs, caught the model doing this in real time on pricing pages. Presented with a company's own site, ChatGPT reasoned in its retrieval logs that the pricing was not showing up directly, possibly because it was loaded with JavaScript, and cited a third-party aggregator instead.
The company had the right answer sitting on its own page. The model simply could not parse it, so it borrowed someone else's version of the same fact.
Your Site, Unreadable by Design
Here is the mechanism in plain terms. AI crawlers, including ChatGPT, Claude, and Perplexity, fetch raw HTML. They do not execute JavaScript the way a browser does. If your fee information, intake steps, or practice-area details load after the page renders, behind a script, a tab, an accordion, or a "click to expand" element, the crawler sees an empty shell where that content should be.
It is not that your content is weak. It is that the content never arrives.
For most consumer businesses, this costs them a G2 citation instead of their own pricing page. For an immigration firm, it costs you the answer to the exact questions that turn a search into a consultation:
- What does an H-1B petition run at your firm?
- What is your intake process for a family petition?
- Do you handle RFE responses, and how fast?
If those answers sit behind a script, ChatGPT will still answer the question. It will just answer it using a directory listing, a legal-marketplace aggregator, or a competitor's page that happens to render in plain HTML. The searcher gets an answer. It is just not yours.
The researchers documented this exact swap on two SaaS pricing pages, Profound and Peec. In both cases, ChatGPT's reasoning trace showed the same diagnosis: the pricing did not show up directly, possibly because it was hidden with JavaScript. The workaround was the same: cite G2 instead, a third-party review aggregator with none of the nuance of the company's actual pricing page.
The company was not beaten on content. It was beaten on parseability. Its numbers were correct, current, and published. They just lived in a part of the page the crawler could not see.
The Pattern Favors Your Own Page, When It Can
This is the part worth sitting with. The research does not show ChatGPT preferring third-party sources on principle. It shows the opposite: given a choice, the model reaches for the firm's own page first, because a direct answer from the source is the model's default preference.
The fallback to an aggregator only happens when the firm's own content is technically unreachable. That means the fix is not a content strategy. You likely already have the fee ranges, the case types, and the intake steps written down somewhere on your site. The fix is making sure a script-free crawler can actually reach them.
Most small-firm websites were not built with this in mind, because until recently nobody needed to think about what a non-human, non-JavaScript-executing visitor could see. WordPress themes, page builders, and templated legal-marketing sites frequently load exactly the content that matters most, including pricing tables, FAQ accordions, and tabbed practice-area breakdowns, via JavaScript after the initial page load.
It looks identical to a client browsing on Chrome. It is invisible to the systems now deciding who gets cited.
Check This Before Friday
You do not need a developer audit to find out where you stand. Do this instead.
- View your own practice-area pages with JavaScript disabled. In Chrome, open DevTools with Cmd+Option+I, go to the Command Menu with Cmd+Shift+P, and run "Disable JavaScript." Reload the page. Anything missing, including fee ranges, intake steps, and FAQ answers, is invisible to ChatGPT and Claude right now.
- Check the raw HTML. Ask a technical contact to run: `curl -s yourfirm.com/practice-areas/h1b | grep -i "fee\|cost\|\$"`. If the number you quote clients does not show up in the raw HTML, an AI crawler is unlikely to see it.
- Move your highest-value facts out of interactive elements. If your fee ranges or intake timeline sit inside a tab, accordion, or "read more" toggle, duplicate that text as plain, visible HTML on the page, even if it is styled to be less prominent. Crawlers read the markup, not the click.
- Prioritize the pages tied to your highest-intent searches first. Start with H-1B sponsorship costs, asylum consultation process, naturalization timeline, and RFE response services. These are the pages most likely to get a citation-worthy question, and the ones costing you the most when they render as blank to a crawler.
- Re-test after you fix it. Ask ChatGPT the same question you would expect a prospective client to ask, and see whether it now names your firm and cites your page instead of a directory.
None of this requires a redesign or a new content calendar. It requires finding out which of your existing pages are already answering the right question, just in a format only humans can see.
The Bigger Picture
Combined with what we found two weeks ago, that ChatGPT's citation sources shift between hidden pipelines 11.6% of the time, the picture is consistent: visibility in AI search is not decided once. It is decided every time a pipeline runs your practice area against a page it can actually parse.
A firm that fixes the JavaScript problem removes one entire failure mode from that equation, not because the content changed, but because the content became reachable. That is a smaller fix than most firms assume, and most competitors have not checked whether they need it.
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.