When someone asks ChatGPT, Gemini, or Perplexity for the best mortgage broker, accountant, or lawyer in their city, the assistant does not run a quick web search and read off the rankings. It draws on what it has already learned about the businesses in that category and, increasingly, pulls live sources to back up its answer. The name it gives is not random and it is not the loudest advertiser. It is the business the available evidence makes the safest to recommend.
That is the most useful way to understand how AI decides what to recommend: these systems are built to give answers they can defend. They favor names that show up consistently, are described the same way across many sources, and carry visible signals of quality like reviews and reputable citations. Below we break down the signals AI assistants actually weigh, why they pick your competitor instead of you, and what you can change so the model has a reason to name you.
How does ChatGPT decide what to recommend?
Large language models are trained to predict the most likely, most defensible response to a prompt. When the prompt is a recommendation request, the model is effectively asking itself: which name appears most often, in the most trusted places, described in the most positive and consistent way? It is pattern-matching against everything it has absorbed and, when connected to live retrieval, against the sources it can pull in real time.
This is why a single great landing page rarely moves the needle on its own. The model is not impressed by self-description; it is looking for corroboration. If your business is mentioned on industry directories, reviewed well on the platforms that matter, written about by third parties, and described consistently across all of them, you become a low-risk answer. If you are barely present, or your details conflict from one source to the next, the model has no reason to gamble on you. For the fuller framework behind this, see our guide to answer engine optimization.
The signals AI assistants weigh most
Across the audits we run, a handful of signals come up again and again as the difference between a business that gets named and one that stays invisible. None of them is a trick. Each one is a way for the model to verify that you are real, established, and well regarded.
1. Consensus across trusted sources
This is the single most important factor. AI assistants are looking for agreement. When your name, services, and reputation are described the same way on Google Business Profile, industry directories, review platforms, and editorial coverage, you give the model a coherent picture it can repeat with confidence. Conflicting or thin coverage does the opposite.
2. Reviews and ratings
Reviews are a public, third-party verdict on your quality, which is exactly what an assistant wants to lean on. A strong, recent review profile makes you far more likely to be recommended, and the specific language customers use often shows up in how the AI describes you. If buyers consistently praise your responsiveness, an assistant may volunteer that you are known for being responsive.
3. Citations from sources the model trusts
Being mentioned and linked by reputable sites, directories, and publications is how you earn the model's trust by association. The more credible sources point to you, the more the model treats you as an established option rather than an unknown. This is the core of getting picked up by retrieval-based assistants, and it is closely tied to how large language models find and cite sources.
4. Consistent business information (NAP and details)
Your name, address, phone number, hours, and service area need to match everywhere they appear. When they conflict, the model cannot be sure which version is correct, and uncertainty pushes you down the list. Clean, consistent details across every listing remove that doubt.
5. Structured, answer-first content and schema
Pages that answer real questions directly, in plain language, near the top, are far easier for an assistant to extract and reuse. Adding structured data (schema) labels your information so machines can read it without guessing. Together, answer-first writing and schema make your content quotable, which is what gets it pulled into responses.
| Signal | What it tells the AI | Where you build it |
|---|---|---|
| Consensus across sources | You are real and consistently described | Directories, listings, coverage |
| Reviews and ratings | Independent proof of quality | Google Business Profile, industry review sites |
| Trusted citations | Credible sources vouch for you | Directories, publications, partner sites |
| Consistent NAP details | No ambiguity about who you are | Every listing and your own site |
| Structured, answer-first content | Your answers are easy to extract | Your website plus schema markup |
Why AI recommends your competitor and not you
The most common reason is simple and a little frustrating: your competitor is more present in the sources AI reads, and you are not. They likely have more reviews, more matching directory listings, clearer pages, or coverage on sites the model already trusts. The model is not judging who is better at the actual work; it is judging who has the stronger, more consistent evidence trail. For more on the gaps that cause this, see what happens as AI search takes share from Google and where it leaves businesses that are not prepared.
The good news is that this is fixable, and it is fixable without outspending anyone. You are not bidding for placement. You are assembling the proof that you deserve to be named. That means closing the gaps your competitor happens to have filled first.
What you can change to get recommended by AI
Once you understand the signals, the work becomes concrete. These are the moves that, in our experience, move a business from invisible to recommended:
- Clean up your business information everywhere. Make your name, address, phone, hours, and services identical across Google Business Profile and every directory. Consistency removes doubt.
- Build a steady stream of reviews. Ask satisfied customers, respond to every review, and keep them recent. Quality and recency both matter.
- Earn citations from sources that matter in your field. Get listed in the directories and mentioned by the publications your industry trusts. Each credible mention adds weight.
- Rewrite key pages to answer questions directly. Lead with the answer, then explain. Cover the real questions buyers ask, in their words.
- Add structured data. Mark up your business, services, FAQs, and reviews so machines can read them cleanly.
- Check what AI says about you today. Ask the assistants the questions your buyers ask and note who gets named. That is your starting line.
A real example of how fast this can move
This is not theoretical. We worked with a Seattle mortgage broker who was effectively invisible to AI assistants when he started. By tightening his business information, building reviews, earning the right citations, and restructuring his content to answer real questions, he became the most recommended broker in his market in about six weeks, generating roughly 30 leads and four closed deals in that window. Nothing about that depended on a bigger ad budget. It depended on giving the models a reason to name him.
Do all AI assistants decide the same way?
The broad logic is shared: every major assistant wants defensible, well-supported answers. But the weighting differs. Perplexity leans heavily on live retrieval and visible citations, so fresh, linkable sources matter a lot. ChatGPT blends what it learned in training with live browsing depending on the prompt. Google's AI features sit on top of years of search infrastructure, so traditional ranking signals still carry weight. If you want to compare the engines side by side, our AI search glossary for business owners is a good orientation, and our breakdown of what an answer engine is covers how ChatGPT, Perplexity, Gemini, and Copilot each work.
The practical takeaway is that you do not optimize for one assistant at a time. You build the underlying evidence, consensus, reviews, citations, clean details, and clear content, and that foundation pays off across all of them at once.
The bottom line
AI assistants decide who to recommend by trusting the evidence. They reward the business that shows up consistently, is well reviewed, is cited by credible sources, and explains itself clearly. None of that is gamed; it is earned, and it is within reach for almost any business willing to do the foundational work. If AI is naming your competitors and skipping you, it is not a verdict on your quality. It is a gap in the proof, and gaps can be closed.