Insights/Guide
How ChatGPT Decides Which Firms to Recommend
ChatGPT names specific businesses when asked for recommendations. Here is the retrieval pipeline behind those answers — and the signals that get a firm onto the list.
When someone asks ChatGPT for "the best personal injury lawyer in Dallas," it names specific firms. Those names are not random and not paid placements — they are the output of a retrieval pipeline with measurable inputs. Understanding the pipeline tells you exactly where to compete.
The two modes that produce recommendations
ChatGPT answers recommendation questions in one of two modes.
Default mode answers from training data. No live lookup, no citations. The model repeats what it absorbed during training — which favors firms with long-standing, widely-distributed reputations: directory presence, press mentions, Wikipedia-adjacent visibility.
Search mode activates for queries that need current information, including most "best X near me" questions. It generates search queries, retrieves results from Bing's index, extracts passages, and synthesizes an answer with citations. Analyses of ChatGPT Search citations show approximately 87% overlap with Bing's top-10 results.
The practical consequence: Bing is the gate. A firm invisible to Bing is invisible to ChatGPT Search, no matter how well it ranks on Google.
Which sources does ChatGPT actually cite?
Citation studies through 2026 are consistent about the pattern. Wikipedia is the single largest cited domain — it accounts for roughly 48% of ChatGPT's top-10 citation volume. Major news publications (Reuters, Forbes, Business Insider) collectively form the largest source category. LinkedIn rose from the #11 most-cited domain to #5 between late 2025 and early 2026. Review platforms like G2 and sector directories appear consistently for "best provider" queries.
For a professional firm, this translates to a target list: the legal, accounting, and healthcare directories Bing ranks for your queries, review platforms with structured ratings, local press, and LinkedIn.
The signals that move a firm into the answer
Brand mention frequency. Studies measuring citation predictors find brand mentions — even unlinked ones — are among the strongest correlates of ChatGPT citation (correlation coefficients of 0.33–0.66 across query sets). The model recommends names it has seen described consistently, in many places.
Entity clarity. ChatGPT must resolve your firm as an unambiguous entity: one name, one location set, defined practice areas. Schema.org markup, consistent directory data, and a Wikidata entry remove the ambiguity that gets firms silently dropped from answers.
Extractable pages. When ChatGPT retrieves your site, it cites chunks: self-contained passages of roughly 40–150 words with specific facts. Pages where each section answers one question directly get quoted; pages that bury answers in narrative do not.
Third-party validation. Recommendation answers are risk-averse. The model prefers firms supported by multiple independent sources — a directory listing, a review profile, a news mention — over firms whose only evidence is their own website.
Why answers compound
ChatGPT's recommendations are sticky. Once a firm appears in trusted sources, it keeps being retrieved; once it keeps being retrieved, it keeps appearing in answers, which generates more mentions. Firms that establish AI visibility early in their market defend it cheaply. Firms that wait compete against an incumbent the model already believes.
To see where you stand today, start with how to check what AI says about your firm — the audit takes under an hour.
Frequently asked questions
Does ChatGPT use Google to find businesses?
No. ChatGPT Search retrieves from Bing's index. Studies show roughly 87% overlap between Bing's top-10 results and ChatGPT's citations, which makes Bing indexing a prerequisite for ChatGPT visibility.
Why does ChatGPT recommend the same firms repeatedly?
Engines weight sources they already trust, and answers reinforce themselves: a firm cited in trusted sources keeps getting retrieved, which keeps it in answers. This compounding is why early movers in a market hold their position.
Can a brand-new website get recommended by ChatGPT?
Rarely on day one. ChatGPT leans on accumulated authority signals. New sites typically appear first in Perplexity, which favors fresh content, and earn ChatGPT visibility over weeks to months as citations accumulate.
Does ChatGPT recommend firms differently in its default and search modes?
Yes. Default mode answers from training data without citations and favors long-established names. Search mode retrieves live results from Bing and can surface firms whose visibility improved recently.