Most ChatGPT Answers Never Check the Web. Here's What That Means for Your Firm.

On October 12, Ahrefs published one of the largest looks yet at how ChatGPT actually builds its answers. Louise Linehan's team studied 66.6 million responses from free, logged-out ChatGPT sessions in August and September 2026 — the default experience a prospective client gets when they open the app without signing in. Only 14.4% of those responses ran a web search at all. When ChatGPT did search, it averaged about 1.6 lookups.

Put plainly: roughly one answer in seven checked the web. The other six were written from what the model had already absorbed.

For an immigration practice, that splits AI visibility into two separate problems. Most firms are only thinking about one of them.

What the Ahrefs Data Shows

A few findings from the study, as reported by Antonio Fernandez at Relevant Audience on October 13, are worth getting precisely right:

• Search is the exception, not the default. In free, logged-out sessions, 85.6% of responses answered without a live lookup. Ahrefs notes that paid and reasoning models search more often, so this is the floor, not the whole picture.
• When it searches, it searches narrowly. About 1.6 queries per searching response. That's a quick check, not a research project.
• Specific, factual questions pull in official sources. In a separate test of 20,000 US prompts, factual questions — pricing, features, company details — produced answers where 84.4% cited at least one official source, and 63.4% of citations went to official sites.
• ChatGPT is leaning toward original sources. Across US answers from March to September, citations shifted toward government bodies, regulators, and organizations' own pages, and away from sites that summarize them. Reddit fell from about one answer in five to about one in 43. Wikipedia fell from roughly one in seven to one in 35.

The study has limits Ahrefs states openly: one model, one period, broad prompts rather than "best attorney near me" questions, and citation data from Ahrefs' own Brand Radar product. Treat the exact figures as directional. The pattern is still hard to ignore.

Two Visibility Problems, Not One

When a family asks ChatGPT about their situation, one of two things happens.

The model answers from memory. This is most of the time. The answer reflects whatever the model learned about immigration law, about your city, and possibly about your firm, from the material it was trained on. A question like "what does an immigration attorney do in an adjustment of status case" or "is it worth hiring a lawyer for a marriage-based green card" will usually get an answer drawn from that stored picture. No page on your site is fetched. Nothing you published last month is read.

The model runs a live lookup. This happens less often, and mostly when the question is specific and factual: a fee, a phone number, whether a firm handles removal defense, which languages a firm consults in. Here, the answer depends on what ChatGPT finds in one or two searches — and the data suggests it favors the official source, which for your practice means your own website.

These are different problems with different fixes. The first is about the long-run picture of your firm and your practice area that exists across the web — the reputation the model has already absorbed. The second is about whether your own pages state clear, current facts that a quick lookup can pick up and cite.

Why This Matters for Small Firms

Large firms and national legal directories have years of coverage working for them in the first scenario. They show up in the material models learn from simply because they have been written about so often.

A three-attorney practice handling family petitions, DACA renewals, and naturalization usually doesn't have that footprint. That's the honest part. The encouraging part is the second scenario, where the playing field is more level. When a prospective client asks a specific question — "does [firm name] offer consultations in Spanish?" or "how much does [firm name] charge for an I-130?" — the lookup goes looking for an authoritative answer, and the study suggests your own site is where it wants to find one.

That's also where many firms come up short. Fees are "call for details." Languages aren't listed. The consultation page doesn't say whether consultations are virtual, how long they run, or what they cost. A live lookup that finds nothing concrete has nothing to cite.

How to Test Which Problem You Have

You can see both scenarios for yourself in about twenty minutes. Use a private browser window and stay logged out of ChatGPT, so you see the default experience a new visitor gets.

1. Ask a broad question about your practice area and city. For example: "What should I know about hiring an immigration lawyer for an asylum case in Houston?" Note whether the answer shows sources. If it doesn't, you're seeing the model's stored picture. Does your firm appear? Does the answer describe your practice area accurately?
2. Ask the same question and request a web search. Add "search the web" to the prompt, or use the search option in the interface. Compare the two answers. Which firms appear now? Which sites are cited?
3. Ask three specific questions about your own firm. Your consultation fee, the languages you consult in, and whether you handle a specific case type, such as consular processing or RFE responses. These are the questions most likely to trigger a lookup. Check whether the answers are correct and whether they cite your site.
4. Write down where each answer came from. If the broad answers ignore you but the specific ones are accurate, your long-run footprint is the gap. If the specific answers are wrong or vague, your own pages are the gap — and that one is faster to fix.
5. Fix the factual gaps first. Add clear fee ranges, consultation format and length, languages, and the case types you take to your consultation and practice area pages. Put them in plain text, not only in images or PDFs.

Repeat the test in a month. The answers will change, and that change is the most useful signal you have.

The Bigger Picture

It's tempting to treat AI visibility as one question: does ChatGPT recommend us or not? The Ahrefs data suggests it's really two questions running on different clocks.

One is slow. The picture a model holds of your practice is built over years, from everything written about immigration law and about your firm across the web. The other is fast. When someone asks a specific question, the answer can change the day your website says something clearer.

Small immigration practices can't control the first one quickly. They can control the second one this week — and for the families asking specific questions about fees, languages, and case types, the second one is where the decision often gets made.

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 Your AI Visibility Score → CLIENTORY.ORG