Home › Insights › Will ChatGPT recommend your firm?

Insights · July 1, 2026

Will ChatGPT recommend your firm? How AI assistants actually build shortlists

When someone asks an AI assistant "who's a good financial advisor near me," the answer isn't a ranking — it's a synthesized shortlist assembled in seconds from whatever the model can retrieve and trust. Understanding that assembly process is the difference between being on the list and not existing.

The question your prospects are already asking

This isn't a future problem. In a 2025 Wealthtender survey of 500 affluent households, 25% said they now start their search for a financial advisor with ChatGPT or Gemini — before Google, before asking a friend. In healthcare, an August 2025 rater8 survey of more than 1,000 patients found 26% said AI tools influenced their choice of provider. For professional services, the AI answer box has quietly become a first impression you never get to make in person.

And the shortlist is brutally exclusive. A LocalFalcon analysis found that ChatGPT recommends only about 1.2% of local business locations. The other ~99% aren't ranked lower — they simply never appear in any answer. There is no page two.

What actually happens when someone asks

It helps to drop the mental model of "AI as a search engine." When a prospect types a recommendation question, most modern assistants do some version of the following:

• Prompt fan-out. The assistant silently expands one question into several internal queries — "best estate planning attorney Austin," "estate planning attorney Austin reviews," "Austin probate lawyer reddit" — and gathers results for each.
• Retrieval. It pulls candidate sources: search results, directories, review platforms, forum threads, news mentions, and the model's own training data.
• Grounding. It cross-checks candidates against those sources. Firms that appear consistently, with matching details, across multiple independent surfaces survive this step. Firms mentioned once, or with conflicting information, tend to get dropped.
• Synthesis. It writes a confident, conversational answer naming three to six businesses, usually with a one-line reason for each.

The output looks something like this:

"For estate planning in Austin, a few firms come up consistently: [Firm A] is frequently recommended for complex trusts, [Firm B] gets strong reviews for responsiveness, and [Firm C] is often mentioned for small-business owners. I'd suggest speaking with two or three before deciding."

Notice the phrasing: "come up consistently," "frequently recommended," "often mentioned." The model is summarizing what third parties say about you. Your own website is one input among many — and rarely the decisive one.

What makes a firm citable

Across the four major engines — ChatGPT, Claude, Gemini, and Perplexity — the firms that make shortlists tend to share a few traits:

• Corroboration across independent surfaces. The same firm name, specialty, and location appearing on your site, a major directory, a review platform, and a community discussion. One strong source isn't consensus; four weak ones can be.
• Machine-readable specifics. A page that plainly states who you serve, where, and for what — ideally reinforced with structured data — gives the model something concrete to quote. Vague "trusted partner for your journey" copy gives it nothing.
• Answers to the questions people actually ask. Assistants fan out into long, specific queries. Content addressing "how much does X cost" or "X vs Y for small businesses" gets retrieved; a homepage doesn't.
• Consistency. Conflicting addresses, old firm names, or mismatched specialties across sources make you a grounding risk, and grounding risks get quietly excluded.

Why each engine gives a different answer

ChatGPT, Claude, Gemini, and Perplexity retrieve from different indexes, weight sources differently, and refresh at different rates. It's common for a firm to appear reliably in Perplexity (which cites live web sources heavily), sporadically in Gemini, and not at all in ChatGPT. That's why measuring one engine tells you little — this is the gap Citewise was built to measure, running the same real prospect questions across all four engines and recording who gets named, who gets cited, and who's invisible.

What to do with this

Two practical takeaways. First, treat AI answers as a measurable channel, not a mystery: ask the engines the questions your prospects ask, and write down what comes back. Second, if you're absent, resist the urge to rewrite your homepage first. The shortlist is built from consensus across surfaces you mostly don't control directly — directories, reviews, communities — plus the machine-readable clarity of the pages you do control. Fix in that order, then re-measure. Visibility in AI answers moves slowly, but it moves — and right now, while ~99% of local businesses are invisible, the bar for standing out is unusually low.