How ChatGPT Decides Which Brands to Recommend

Introduction

Two demolition companies in Dallas ask ChatGPT the same question. One shows up in the answer. The other doesn't. Same city, same industry, same query. So what's different?

ChatGPT recommends brands using two systems. The first is the brand associations baked into its training data. The second is live retrieval from Bing's search index when a query needs fresh information. To get recommended, you need to be either deeply associated with your category before the model's training cutoff, or visible in the Bing-indexed sources ChatGPT pulls when it browses.

This post breaks down both systems with sourced data, explains how fan-out queries reshape the candidate pool, and gives you a clear playbook. We'll lean on research from SE Ranking, AirOps, and Profound, plus a real example we documented with one of our own clients. If you're new to AI search, start with our guide on Generative Engine Optimization (GEO).

How Does ChatGPT Choose Which Brands to Recommend?

ChatGPT chooses brands through two systems: brand associations learned during model training, and live retrieval from Bing's search index. Most brand-recommendation queries trigger search, so retrieval is the lever you can actively pull. Training data shapes which brands ChatGPT knows. Bing shapes which brands ChatGPT can see right now.

When ChatGPT can answer from training data alone, it pulls from associations it learned before the training cutoff. Every place your brand was mentioned alongside your category and competitors during that window helps shape what ChatGPT "knows" about you.

When ChatGPT decides a query needs fresh information, it searches the web through Bing. It then selects sources from those results to build its answer. Most "best [category] in [city]" or "top [product] for [use case]" queries fall into this second bucket.

Why Does ChatGPT Use Bing Instead of Google?

ChatGPT uses Bing because of OpenAI's strategic partnership with Microsoft (Bing's owner) and because Google's search index isn't licensed for external commercial use. Microsoft's Bing Web Search API was already built and openly available, and Microsoft's multi-billion-dollar investment in OpenAI made integration the natural choice. Google is a direct competitor in AI search, so its index was never realistically on the table.

The result, according to a Seer Interactive study, is that 87% of ChatGPT's citations match Bing's top organic results. Only 56% match Google's, with a median Google rank of 17 for those that do. If your brand doesn't rank in Bing, ChatGPT can't see it during retrieval.

That single fact reframes the whole problem. You can rank #1 in Google, dominate AI Overviews, and still be invisible to ChatGPT, because the underlying search index is different.

If you've ignored Bing, you're missing one of the most important channels for AI search visibility.

Bing's index is not Google's index

Bing and Google index different content with different priorities. Bing is more open to younger domains. It values explicit, declarative content. It weights social signals like forum threads and third-party reviews more heavily than Google does.

Pages that are over-optimized for Google often underperform in Bing. Pages that read like clear, helpful answers tend to do well in both.

What this means for SEO programs that ignored Bing

Most SEO programs treat Bing as an afterthought. That made sense when Bing was a small share of total search volume. It doesn't make sense now that Bing powers 87% of ChatGPT citations.

Verify your site in Bing Webmaster Tools. Submit your sitemap. Audit your top category pages for Bing-specific issues. Treating Bing as a separate channel, not an afterthought, is one of the highest-leverage things you can do for ChatGPT visibility.

What the Data Says About ChatGPT Citations

Three large-scale studies published between 2025 and early 2026 give us the clearest picture of how ChatGPT picks pages to cite. The pattern is consistent: distributed authority signals beat any single on-page tactic.

Referring domains are the single strongest predictor

SE Ranking analyzed 129,000 unique domains across 216,524 pages in 20 niches. They found that referring domain count is the single strongest predictor of ChatGPT citation likelihood. Sites with up to 2,500 referring domains averaged 1.6 to 1.8 citations. Sites with 350,000+ referring domains averaged 8.4. The researchers also identified a clear threshold: at 32,000 referring domains, citations roughly double, jumping from 2.9 to 5.6.

Most retrieved pages are never cited

AirOps studied 548,534 pages retrieved by ChatGPT across 15,000 prompts. Only 15% of retrieved pages were cited in a final response. The other 85% were found, evaluated, and discarded. Even more striking: 32.9% of all cited pages came from fan-out sub-queries, not the original prompt. ChatGPT does a lot of background searching you never see.

Source mix is shifting

A Profound study of 30 million citations from August 2024 to June 2025 showed Wikipedia as ChatGPT's dominant source. The picture changed quickly. In mid-September 2025, ChatGPT sharply reduced citations to Reddit (from a peak near 60% to roughly 10%) and Wikipedia (from above 55% to under 20%). Publishers like PR Newswire, Forbes, Medium, and LinkedIn picked up the slack. The lesson: source mix in ChatGPT is volatile, so don't over-index on any single channel.

Other research-backed findings

The SE Ranking study and a Search Engine Journal analysis of the top 20 ChatGPT citation factors surface several other signals worth optimizing for:

• Pages with section lengths of 120 to 180 words between headings averaged 4.6 citations.
• Articles over 2,900 words averaged 5.1 citations.
• Pages with First Contentful Paint under 0.4 seconds averaged 6.7 citations, vs. 2.1 for pages over 1.13 seconds. That's a 3x speed-driven gap.
• Content updated within the last 30 days got cited about 3.2x more often than older material.
• Domains with active profiles on Trustpilot, G2, Capterra, or Yelp had 3x higher citation probability than domains without.
• Pages ranking 1 to 45 in Google averaged 5 citations, while pages ranking 64 to 75 averaged 3.1.

What Is a Fan-Out Query, and Why Does It Matter for ChatGPT?

A fan-out query is a parallel sub-query ChatGPT generates by breaking your original prompt into subtopics. It runs those sub-queries against Bing, retrieves pages for each, and uses the combined pool to assemble its answer. The AirOps study found that 32.9% of ChatGPT's cited pages come from these sub-queries, not the original prompt. Fan-out is why topical depth beats head-term optimization.

Here's what that looks like in practice. We recently ran the prompt "Give me a list of the top demolition companies in Dallas, TX" across four AI platforms. JRP Demolition appeared prominently in three of them (ChatGPT, Gemini, and Google AI Mode), but not in AI Overviews. We documented the full breakdown in our post on why your brand can earn ChatGPT citations but miss AI Overviews and in our JRP Demolition case study.

For that ChatGPT result, the fan-out wasn't just "demolition companies in Dallas." It was a constellation of sub-queries: residential demolition specialists, commercial demolition contractors, top Dallas demolition reviews, and head-to-head comparisons of likely candidates. JRP was cited because it ranked well across several of those sub-queries, not just the head term.

Why fan-out changes optimization strategy

Most SEO programs concentrate effort on the head term, the page that targets the most generic version of the query. Fan-out makes that strategy incomplete. ChatGPT is just as likely to pull from a long-tail comparison page, a pricing breakdown, or a niche use-case article as from your main category page.

In the AirOps study, 85% of pages ChatGPT retrieved were never cited. The 15% that were cited skewed heavily toward content that answered specific sub-questions cleanly, not content that tried to answer everything at once.

How to optimize for fan-out queries

Build out the topical cluster around your category instead of stacking everything onto a single page. The pages that perform best as fan-out retrieval surfaces tend to share a few traits.

• Comparison pages ("your brand vs. competitor X") that answer one specific sub-question cleanly.
• Pricing pages with explicit, structured data and clear comparisons to typical alternatives.
• Use-case-specific landing pages, by industry, role, company size, or scenario, that match the sub-queries ChatGPT is likely to generate.
• FAQ pages and dedicated answer pages that address one question per page in plain language.
• Integration and feature-specific pages that match how buyers describe their actual problem, not how the category describes itself.

If your site has only a homepage and a generic "solutions" page, ChatGPT has very little to fan out into. Topical depth, measured by how many distinct sub-questions your site answers well, is a structural advantage that compounds over time.

What Are the 6 Signals That Get Your Brand Recommended?

Six signals carry the most weight when ChatGPT decides which brands to recommend: referring domain authority, entity recognition, Bing visibility, third-party review presence, content structure and freshness, and cross-source consensus. Each signal compounds. A brand that hits all six is a recommendation candidate. A brand that hits one or two rarely shows up.

How Do You Get ChatGPT to Recommend Your Brand?

To get recommended, focus on six tactical priorities: build referring domain diversity, strengthen your entity profile, fix your Bing presence, claim third-party review profiles, restructure your highest-value pages, and earn co-mentions in trade publications. The fastest gains usually come from a referring-domain push, plus a Bing audit, plus claiming the right review profiles.

1. Build referring domain diversity. Pursue links from a wide variety of authoritative domains: trade publications, industry databases, podcasts, news mentions, partnerships. Volume matters less than diversity. Concentrated link graphs from a handful of sources underperform distributed ones.

2. Strengthen your entity profile. Build out a Wikipedia or Wikidata entry where eligibility allows. Implement complete schema markup (Organization, LocalBusiness, Product). Make sure your NAP data is identical across every directory. Submit to authoritative industry databases.

3. Fix your Bing presence. Verify your site in Bing Webmaster Tools. Submit your sitemap. Audit your top category pages for Bing-specific issues. Bing weights H1 headers, exact-match keywords, and on-page structure more rigidly than Google.

4. Claim and optimize third-party review profiles. Trustpilot, G2, Capterra, Yelp, BBB, and category-specific directories triple your citation probability. Claim every relevant profile, fill it out completely, and pursue authentic reviews. For local brands, the Maps Pack equivalent is your Google Business Profile.

5. Restructure your highest-value pages. Aim for 120 to 180 words per section between headings. Add explicit Q&A sections. Use declarative claim sentences. Lead with the answer. Make sure pages targeting recommendation queries are over 2,000 words and updated within the last 30 days.

6. Earn co-mentions in the right places. Get your brand listed in third-party listicles for your category. Pursue mentions in trade publications and news articles. With Reddit and Wikipedia citation rates dropping in late 2025, lean harder on PR Newswire, Forbes, LinkedIn, Medium, and industry-specific outlets.

For a structured, end-to-end approach to all six, our GEO services handle the strategy, audit, and execution as one program.

Why Is ChatGPT Ignoring Your Brand?

If ChatGPT consistently fails to recommend your brand, the cause is almost always one of three things.

1. Your referring domain profile is too thin or too concentrated. Referring domain count is the single strongest published predictor of ChatGPT citation. Brands without a broad, distributed link profile rarely make the cut.

2. You're invisible in Bing. ChatGPT's retrieval system runs on Bing. If you don't rank there, you can't be cited when search runs.

3. You only appear in your own content. No third-party reviews, no listicle mentions, no industry citations means no cross-source consensus, which means no recommendation.

This is also where the difference between GEO and SEO matters. Each of these failure patterns has a clear playbook. The harder question is which one is biting your brand right now, and that's exactly what an audit answers.

Conclusion

ChatGPT recommends brands two ways: through training-data associations and through Bing-powered retrieval. Both reward distributed authority signals over single-page tactics. Fan-out queries make topical depth a real competitive moat, because they expand the candidate pool well beyond your head-term page.

The brands that get recommended consistently aren't the ones with the best landing page. They're the ones with broad backlink profiles, strong entity recognition, real Bing visibility, third-party validation, and content depth that answers many sub-questions cleanly.