Why does a competitor appear more often than my brand in AI answers?

Direct answer

Engines pick their sources through their own search retrieval, starting from a query the model generates. Research shows an effect on the citation of content an engine has already retrieved, on the condition that the source is in the context to begin with.

What engines document about how they pick sources

Google documents that, to appear in AI Overviews and AI Mode, a page has to be indexed and eligible to be shown with a snippet in Google Search, and states that there are no additional requirements or special optimizations¹. It also describes AI Mode issuing several related searches concurrently across subtopics and bringing those results together into one response².

Microsoft documents that Microsoft 365 Copilot Chat generates a short query of a few words and sends it to Bing, and that the Sources button shows the exact query that was sent³. Google documents that the Gemini API with search decides whether to search, generates one or more queries, and returns annotations linking spans of text to their sources⁴. OpenAI documents that its web search tool returns inline citations plus a field with every URL retrieved, wider than the citations displayed⁵.

What these pages share is the mechanism: an intermediate query, generated by the model, and a retrieval step that decides what enters the context. The brand that appears is the one that survives that retrieval for that query.

Which crawler each engine looks at

Each provider separates the agent feeding search from the agent feeding training. OpenAI documents OAI-SearchBot for appearing in ChatGPT search results and GPTBot for model training, and clarifies that ChatGPT-User does not determine whether content may appear in search⁶. Perplexity documents PerplexityBot for surfacing and linking sites in its results, and states it is not used to train models⁷. Anthropic documents Claude-SearchBot as the agent that navigates the web to improve search result quality⁸.

The practical consequence is one of configuration. A site blocking a provider's search agent stays outside that surface, whatever the quality of its content. Google also documents that the nosnippet directive prevents content from being used as a direct input for AI Overviews and AI Mode, and that max-snippet limits how much may be used⁹.

What research says about what can be changed

The foundational work on optimizing for generative engines reports, on its own benchmark, visibility gains of up to 40% in generative engine responses¹⁰. A survey of 45 studies published in 2026 concluded that the figure holds in laboratory conditions, with the page already among the sources the engine has at hand, and that the works reviewed leave it undemonstrated that any of it brings lasting traffic¹¹.

The same survey reports the most reproducible findings in the field: topical relevance and position within the context, with generic heuristics transferring poorly across contexts¹¹. And a 2026 work separates two things that often get mixed up: a document's contribution to the answer and its citation, which is what sends traffic back to the creator¹².

Which sources engines cite, according to whoever measured

Public measurements come from companies selling AI visibility monitoring, so each figure travels with its origin and stays unmixed with the others.

According to a study by Profound, a company that sells AI visibility analytics, across 680 million citations between August 2024 and June 2025, in ChatGPT Wikipedia held 7.8% of all citations, in AI Overviews Reddit held 2.2% and in Perplexity Reddit held 6.6%¹³. According to a study by Peec AI, a company that sells AI visibility monitoring, across 30 million sources in five engines in the United States, the ten most-cited domains were Reddit, YouTube, LinkedIn, Wikipedia, Forbes, G2, Yelp, Facebook, Medium and TechRadar, published as an order and without percentages¹⁴. According to a study by Semrush, a company that sells SEO software and AI visibility monitoring, across more than 230,000 prompts between July and October 2025, Reddit and Wikipedia led the most-cited domains and their rates moved sharply week to week¹⁵.

The three measure with their own prompt panels, which are incomparable between them. What they hold together is the shape: a few third-party surfaces concentrate a large part of the citations.

What happens to traffic even when a citation appears

The Pew Research Center, a research centre with no SEO product, measured across 68,879 searches by 900 US adults in March 2025 that a click on a source cited inside an AI summary happened on 1% of visits¹⁶.

How I measure it in a specific case

I replicate the Observatory's design over the brand's own category. I run a bank of unbranded prompts, repeated across several models, and I get two comparable series: which brands get named and which domains get cited in those same answers.

I hold the comparison together with three conditions. I write the prompts without naming brands, so the appearance belongs to the model. I repeat them several times per model, to separate the signal from the noise of a single run. And I run them on the same models and in the same window for both brands.

Before that I make two checks, once, taken from the official documentation. I confirm that robots.txt allows each provider's search agent, which differs from its training agent⁶⁷⁸. And I check whether a nosnippet or max-snippet directive limits the use of the content as a direct input for Google's AI features⁹.

How to get AI to name you more often than your competitor

1. Start with access, ahead of content. Check in your robots.txt that each provider's search robot can get in, because an inherited block leaves you out of that surface however good your writing is.

2. Check your pages' snippet directives too. A nosnippet or a short max-snippet set years ago to protect the content now stops Google from using it in its answers.

3. Write one page per category question, with the answer in the first paragraph. That is the most reproducible finding in the studies: it matters that the page answers exactly the question, and it matters where that answer sits.

4. Back every answer with a figure of your own, dated and sourced. It is worth it because a figure with a name and a date is what an engine can attribute to you, and attribution is what sends visits back.

5. Take care of what forums, encyclopaedias and social networks say about you. The public measurements agree that a few surfaces belonging to somebody else gather a large part of the citations, so those pages weigh as much as yours.

6. Measure with the same bank of questions every month, leaving it unchanged. The discipline is worth it because that way, if something moves, you know it was the site and the questions stayed the same.

Which AuraMetrics modules address this

See which sources AI cites when it answers about the category. Prompt Tracking shows, for each prompt and model, the sources the answer cited, and AI Market Leaders groups them by brand.

Measure the brand with the same category prompts. Prompt Tracking runs prompts on the five engines.

Review the site's technical signals with the GEO Score and the GEO Framework. The GEO Score scores the site from 0 to 100 on the four pillars of the GEO Framework.

Frequently asked questions

Why do competitors appear where our company should, in AI answers?

Engines document that the source is picked by retrieval from a query the model generates, and a preference criterion between brands stays unpublished¹³. Public measurements show that a few third-party surfaces concentrate a large part of the citations¹³.

How do large language models understand and describe brands?

The observable part is the retrieval step: an intermediate query and a set of sources that enter the context⁴⁵. Research separates a document's contribution to the answer from its citation, which is what sends traffic back to the creator¹².