The short answer
AI search engines assemble answers from three pools: community conversations (Reddit, YouTube, Quora), brand-owned pages (your site, marketplace listings, price comparisons), and independent editorial and reference (press, Wikipedia, review directories). Which pool carries your industry is largely determined by what your buyers are asking, not by how well you polish your own site.
An August 2026 analysis by Trendos, published on Search Engine Journal, looked at 107 million AI answers across ChatGPT, Perplexity, Gemini, and Google AI Overviews. Community content made up about half of the top cited sources in every industry studied. After that, the mix splits sharply: IT and software services lean on independent editorial (47%), consumer goods lean almost entirely on brand and marketplace pages (46%), and communication services sit in the middle with a heavy community and editorial tilt.
So the useful question is not "how do I optimize for AI search" but "which pool does my industry cite, and am I actually visible there?" This article walks through the three pools, the numbers behind them, and the 20-minute audit that tells you where you stand.
The three pools, and why engines trust them
Before the numbers mean anything, it helps to name the pools and understand what they signal.
Community and UGC. Reddit, YouTube, Quora, forums, social platforms. These sources carry first-hand experience. When an engine needs to know how something actually behaves in the wild, a thread where real people report real results is stronger evidence than any marketing page.
Brand, retail, and owned. First-party company sites, Amazon and marketplace listings, reviewing and price-comparison resources. For a physical product, this content is often the answer itself: a listing with specs, photos, and reviews answers the question without needing a third party.
Independent editorial and reference. Press, Wikipedia, B2B directories such as Clutch, G2, and Gartner. These sources hold no stake in the sale. For a high-consideration purchase like enterprise software, an engine treats vendor claims as unverified and looks for someone else to confirm them.
The three pools: every AI answer is assembled from one of these. Engines pick whichever type of source answers the question most credibly.
What the numbers actually show
The study averaged the share of the top 10 citation sources in each engine for three industries.
Industry | Community & UGC | Brand / retail / owned | Independent editorial & reference |
|---|---|---|---|
IT & solutions services | 51% | 2% | 47% |
Consumer goods | 50% | 46% | 4% |
Communication services | 50% | 10% | 40% |
Average share of each engine's top 10 sources; rows rounded, engines weighted equally.
The same three pools, re-sorted by industry. The mix does not shift slightly; it flips.
Read that table in reverse and the pattern is brutal in both directions. For IT brands, your own website is the least important place to be: 2%. For consumer-goods brands, the entire editorial industry is nearly absent: 4% of citations.
The reasons are mechanical. A B2B buyer has a large budget, a named competitor list, and a long evaluation cycle, so an answer that says "trust the vendor's homepage" feels useless. An engine instead cites G2, Clutch, Gartner, SourceForge, and trade press because they validate rather than claim. A consumer-goods buyer asks "is this product any good," and the product listing already contains the answer, so the engine treats marketplace pages as answer copy, not as ad copy. Communication services (telecom, media) sit in the middle because reputation is the entire product: what subscribers and journalists say about you outweighs what you say about yourself.
Each engine reads sources differently
The same industry mix looks different in each engine, because each one has a different default trust model.
Engine | Where it leans | In practice |
|---|---|---|
Google AI Overviews | Community and social | YouTube and Reddit dominate every industry, often four of five top sources |
ChatGPT | Editorial and reference | Wikipedia, trade media, and reference lead; Reddit still strong |
Perplexity | Community plus proof | Reddit and YouTube on top, review directories and press close behind |
Gemini | Industry-dependent | Editorial and reference for services; brand and retail for physical products |
This is why a single-engine test is a bad proxy. Your visibility can look excellent in Gemini and nonexistent in Google AI Overviews, and both facts are true at the same time.
The one source that never leaves the list
Reddit appears in the top three citation sources for every industry across all four engines, and it is the most-cited source in more than half of those lists. No other domain repeats that consistently.
That does not mean "post on Reddit." It means conversations in relevant subreddits are the floor of your AI search visibility. Engines cite threads where someone already asked the question and got grounded answers. Promotions get ignored, and spam is the fastest way to make a subreddit hostile to your brand. The practical move: find out what people already say about your category, and build content that the best of those answers would genuinely link to or improve upon.
Find your own mix in about 20 minutes
Your industry may differ from the three studied, and your sub-category could deviate significantly. The fix is a small, repeatable audit.
- Write down 10 real questions your buyers ask an AI before a purchase decision.
- Run each question in ChatGPT, Perplexity, Gemini, and Google AI Overviews.
- Record the domains cited in each answer, one row per domain.
- Tally the recurring domains and sort them into the three pools.
- Mark where your brand appears, where a competitor appears, and which pool is driving the answer.
The first pass takes longer, because you are building the sheet. After that, a recheck takes about 20 minutes, and monthly is plenty. A one-page GEO citation audit wraps this five-step check into a single report, so you skip the manual domain-copying and get a before-and-after record of where you appear.
What the pattern tells you to do
The playbook follows directly from whichever pool dominates your industry.
If independent editorial and reviews dominate (IT, software, services): claim and fully complete your G2, Clutch, Gartner, and SourceForge profiles: categories, features, pricing, screenshots. Keep a steady stream of current reviews; recency matters more than the total count. Earn trade-media coverage by publishing citable original data or genuinely informed comparisons of a real problem.
If community and reputation dominate (telecom, media, and anything reputation-sensitive): keep an accurate, well-sourced Wikipedia entry updated through third-party sources, which is also how Wikipedia's conflict-of-interest rules want it. Shift budget from landing-page polish to earned press. Watch the subreddits, forums, and review threads where your brand is actually discussed.
If owned and marketplace pages dominate (consumer goods): audit your Amazon and marketplace listings: titles, bullets, specs, Q&A, reviews. Add structured, current reviews to your product pages and prompt buyers to mention specific use cases. Build a second front on YouTube and Reddit with genuinely helpful review content and real participation.
One more practical note: treat your citation mix as an input to content planning, not as a campaign result. The pattern is stable until the engines change their default trust model, so revisit it on a cadence rather than reacting to a single bad week.
What the study does not tell you
The 107 million answers make this a large sample, but the summary has real limits worth knowing before you turn it into a strategy.
- The figures are averages of the top 10 sources, so they describe the typical answer in each industry, not every answer.
- Only three industries were analyzed. Yours may be close, or it may not be, which is why the five-step audit matters more than the table.
- The research came from Trendos, which also sells citation tracking as a service. Read it as a map, not a verdict, and let your own market's data be the confirmation.
Mistakes that waste the effort
- Running one generic AI-search checklist. B2B electronics and a home goods store need opposite strategies. Start from the pool that dominates your industry.
- Polishing only your own pages. For IT and services, the engine almost never cites the vendor's site. Time spent there is worth little until independent coverage exists.
- Treating Reddit as a placement channel. Engines cite threads with genuinely useful experience. Low-effort promotional posts do not rank, and they damage the community standing that actually drives citation.
- Checking one engine and calling it done. Google AI Overviews and ChatGPT can show entirely different citation sets. Run the audit across all four, or you will optimize for a hallucinated average.
FAQ
Which AI search engine cites the most community content? Google AI Overviews. YouTube and Reddit dominated the top sources for every industry studied, often filling four of the five slots.
Why does Reddit get cited in almost every industry? Reddit threads contain first-hand experience in a conversational format that engines can quote directly. It is the most consistently cited domain across industries and engines, appearing in the top three everywhere.
My industry was not in the study. Can I still use it? Yes. Use the study to understand the mechanism: community, editorial, and owned pages each answer a different type of question. Then run the five-step audit with your own buyer questions to see which pool your industry actually cites.
Does every brand need a Wikipedia article? Only if your industry is editorial-driven, and even then the entry must be accurate and sourced from third parties. Wikipedia's conflict-of-interest rules mean self-editing is the fastest way to get the entry cleaned up or removed.
How often should I recheck my citation mix? Roughly monthly if you are testing content or earning coverage, or after a major engine update. The mix moves slowly, and a monthly check is enough to catch drift before it becomes a gap.
Author: Isabel Grant, Researcher of 2,000+ AI Citation Patterns at Auspia. Isabel writes about where AI answers get their sources, how citation patterns differ by industry, and what those patterns mean for content strategy.
