---
title: "Do AI assistants favor big brands over smaller competitors?"
description: "Mostly yes when products look alike: AI assistants default to market leaders, but in tests a small, clearly stated advantage beat a famous name."
canonical: "https://underneath.agency/resources/do-ai-assistants-favor-big-brands"
published: 2026-10-07
updated: 2026-10-08
publisher: "Underneath (https://underneath.agency/agent)"
entity: "https://underneath.agency/.well-known/entity.json"
---
Guide · AI search

# Do AI assistants favor big brands over smaller competitors?

Yes: when nothing else separates the options, AI assistants lean hard toward the brands they already know. The lean is real but conditional, because in controlled tests a small, clearly stated advantage was enough to beat a famous name. For a challenger brand, the task is to give the assistant a reason to pick you, and to put that reason where the assistant will read it.

## The short version

1. In 50 unbranded soda questions to ChatGPT and Perplexity, major brands took 62.2% of brand mentions and niche brands 9.0% ([Chen and colleagues](https://arxiv.org/abs/2509.08919), University of Toronto, 2025).
2. When ten skincare products had identical specifications, three AI models picked the one real brand in all 670 valid trials ([Chu and Hou](https://arxiv.org/abs/2606.17443), 2026).
3. In the same tests, a rating edge of just +0.075 stars was enough for an unknown brand to win half the time.
4. In one vendor’s data on 102 brands, household names appeared in 73% of unbranded AI answers on day one and niche brands in 11% ([Kumar](https://arxiv.org/abs/2606.20065), 2026).
5. In [our own study](https://underneath.agency/research/brand-entity-ai-recommendations-study) of four assistants, independent coverage was the strongest signal: each tenfold rise in independent sites naming a brand went with 4.7 times the odds of a recommendation.

## How strong is the big-brand lean in real AI answers?

It is strong enough to see clearly in live tests of several assistants. [Chen and colleagues](https://arxiv.org/abs/2509.08919) asked ChatGPT and Perplexity 50 unbranded questions, such as “most popular cola brand” or “top cola brands in the US”. They then sorted every brand named into major, niche or other.

| Assistant | Major brands | Niche brands | Other |
|---|---|---|---|
| ChatGPT | 56.3% | 12.3% | 31.4% |
| Perplexity | 67.9% | 5.8% | 26.3% |
| Both combined | 62.2% | 9.0% | 28.8% |

Coca-Cola alone drew 107 mentions on Perplexity. The authors suggest that prominent sources and what the assistants already “know” jointly pull answers toward major labels. That is their reading, not something they measured, and questions such as “most popular” invite famous names by design.

A second, larger data set points the same way. Kumar, a co-founder of the AI visibility company Ranqo, analyzed 102,025 AI answers about brands tracked on the company’s platform between March and May 2026. On each brand’s first tracking run, global household names appeared in 73% of unbranded category answers, and niche or small brands in 11%. This is a vendor’s own customer data, not a random sample, and the brand tiers were assigned by hand.

## Is it the brand name itself that wins?

In controlled tests, yes: with nothing else to go on, the assistants picked the name they recognized. [Chu and Hou](https://arxiv.org/abs/2606.17443) gave three AI models (GPT-4o-mini, Claude Sonnet and Gemini 3 Flash) lists of ten skincare products with identical ratings, prices, reviews and ingredients. One was a real brand such as CeraVe; the other nine were invented names.

Across 670 valid trials, the real brand was recommended every single time, in English and in Chinese. The product list was supplied in the question, with no live web search. So this measures what the assistant brings from its training, not what it finds online. Skincare makers can see what this means for them in [how skincare brands get found in AI answers](https://underneath.agency/resources/skincare-brands-ai-search).

An older test by [Pfrommer and colleagues](https://arxiv.org/abs/2406.03589) at UC Berkeley found a similar pull in some assistants. Across 50 product categories, GPT-4 Turbo and Llama 3 were heavily swayed by what they already knew about product names, and GPT-4 Turbo paid little attention to the product pages it was given. Those were 2024-era models in a research setup, not today’s consumer ChatGPT.

## Can a small, clear advantage beat a famous name?

Yes, in the same tests the brand advantage collapsed once a rival was visibly better. When Chu and Hou gave an invented brand better specifications than the real one, the assistants still chose the real brand only 1.7% to 4.6% of the time.

The tipping point was tiny. An unknown brand won half the time with a rating just +0.075 stars higher, 1.6 times as many reviews, or a 7.3% lower price. Across all their conditions, product details explained 82.4% of how the assistants ranked the products, and the brand name only 1.2%.

The authors call this a “conditional monopoly”: the big brand wins as a tiebreaker when the information looks the same. The catch is that their test handed the assistant clean, comparable facts for every product. In real AI answers, a smaller brand’s facts first have to be found.

## Is “big” the same as “popular” to an AI assistant?

Not exactly: the brands AI assistants favor are not always the ones consumers know best. [Malthouse and colleagues](https://arxiv.org/abs/2609.16304) at Northwestern University asked six AI models for up to five brands in five categories, 40 times each, with web search switched off. They compared the results with Kantar BrandZ scores for how readily consumers think of each brand.

The match was loose. Disney Cruise Line, which scored 111 on that measure, was almost never recommended, while Viking, at 70, was recommended prominently. In cordless drills and hiking jackets, the models favored premium brands such as DeWalt, Milwaukee, Patagonia and Arc’teryx over mass-market names such as Black+Decker and L.L.Bean. That pattern held in only two of the five categories. We look at why [well-known brands miss AI recommendations](https://underneath.agency/resources/why-well-known-brands-miss-ai-recommendations) and how to check your own.

A hotel experiment by [Baig and colleagues](https://arxiv.org/abs/2606.16344) shows another way established players gain ground. Across twelve AI models choosing between made-up hotels, a high review count (2,100 against 45) raised the chance of being recommended by 8.3 percentage points. The authors note that this kind of weighting advantages established properties over new entrants with thin review histories.

## Where does the big-brand advantage come from?

Mostly from how widely other people write about a brand, as far as current data can tell. In [our study of brand entity signals](https://underneath.agency/research/brand-entity-ai-recommendations-study), ChatGPT, Gemini, Perplexity and Claude answered the same 80 US buyer questions. Of the options all four named for a question, 60.0% had an English Wikipedia article, against 20.6% of the options only one assistant named.

Most of that gap went away once we accounted for how prominent each brand already was. The strongest signal we measured was independent coverage: how many websites, other than the brand’s own, named it in the pages the assistants cited. Each tenfold increase went with 4.7 times the odds of being recommended. That is an association, not proof of cause.

Chen and colleagues reached a similar view from the source side. On questions about well-known brands, 93.5% of the sources ChatGPT drew on were earned media, meaning third-party publications and reviews rather than the brand’s own site or social posts. Paid reach is a separate question, weighed in [whether ad spend helps AI recommendations](https://underneath.agency/resources/does-ad-spend-help-ai-brand-recommendations).

## What should you do about it?

Treat AI visibility as a contest challengers start from behind, and give assistants a specific, checkable reason to choose you.

1. **Measure unbranded questions.** Ask the buyer questions your customers ask, without naming yourself, several times on each assistant. Being recognized by name tells you little about being recommended.
2. **State your advantage as plain, comparable facts.** Ratings, review counts, prices and specifications moved the assistants far more than brand names did in controlled tests. Publish them where they are easy to find and compare.
3. **Earn independent coverage.** Reviews, comparisons and mentions on other sites carried the strongest link to recommendation in our data.
4. **Do not invent authority.** Chu and Hou found that made-up clinical claims swayed the assistants, but they flag such claims as potential false advertising and call for platforms to check them.
5. **Track each assistant separately.** Perplexity leaned toward major soda brands more than ChatGPT did in the same test.

If you want help turning this into a plan, see [how we approach generative engine optimization](https://underneath.agency/services/generative-engine-optimization).

## What does the research not tell us yet?

The research shows a clear lean toward familiar brands, but it leaves real gaps.

- **The tests are narrow.** The soda test covered one category and two assistants. The skincare tests supplied the products in the question instead of letting assistants search.
- **The causes are inferred.** Whether training data, prominent sources or something else drives the lean is the researchers’ interpretation, not something they observed.
- **Vendor data needs independent checks.** The tier figures come from one company’s customers, who skew toward software, fintech and Indian consumer brands.
- **No proven lasting fix exists yet.** A [critical survey of 45 studies](https://arxiv.org/abs/2607.14035) found no technique with a stable, cross-platform effect on whether content is discovered in the first place.

## Frequently asked questions

### Does ChatGPT prefer well-known brands?

In unbranded tests, usually yes. In 50 soda questions, 56.3% of the brands ChatGPT named were major labels and 12.3% niche ones; Perplexity leaned further, at 67.9% major.

### Can a small brand beat a big brand in AI recommendations?

Yes, if the assistant can see a clear advantage. In controlled skincare tests, an unknown brand with slightly better ratings, more reviews or a lower price won most head-to-head comparisons.

### Do AI assistants simply recommend the most popular brands?

Not reliably. A Northwestern study found Disney Cruise Line almost never recommended despite high consumer awareness, while some premium brands were favored over bigger mass-market ones.

### Why do AI assistants default to market leaders?

Researchers point to what assistants learned in training and to which sources are most prominent online. Our own data points to independent coverage: brands that many other sites write about were recommended far more often.

## Sources

- Chen, Wang, Chen and Koudas (2025), [Generative Engine Optimization: How to Dominate AI Search](https://arxiv.org/abs/2509.08919), arXiv:2509.08919.
- Chu and Hou (2026), [Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems](https://arxiv.org/abs/2606.17443), arXiv:2606.17443.
- Kumar (2026), [Generative Engine Optimization at Scale: Measuring Brand Visibility Across AI Search Engines](https://arxiv.org/abs/2606.20065), arXiv:2606.20065.
- Pfrommer, Bai, Gautam and Sojoudi (2024), [Ranking Manipulation for Conversational Search Engines](https://arxiv.org/abs/2406.03589), arXiv:2406.03589.
- Malthouse, Lee, Yang, Pal and Feng (2026), [Evaluating Brand Retrieval and Ranking in Large Language Model Recommendations](https://arxiv.org/abs/2609.16304), arXiv:2609.16304.
- Baig et al. (2026), [Whose hotel does the AI recommend? An algorithm audit of reputation signals in LLM-assisted hotel selection](https://arxiv.org/abs/2606.16344), arXiv:2606.16344.
- Martinez (2026), [Optimizing Visibility in Generative Engines: A Critical Survey of Generative Engine Optimization (2023-2026)](https://arxiv.org/abs/2607.14035), arXiv:2607.14035.
- Underneath (2026), [Do Wikipedia and schema make AI assistants recommend a brand?](https://underneath.agency/research/brand-entity-ai-recommendations-study)

---

This is the Markdown twin of https://underneath.agency/resources/do-ai-assistants-favor-big-brands. The HTML page is canonical. Publisher: Underneath, https://underneath.agency/agent. Site index: https://underneath.agency/llms.txt.