When someone asks ChatGPT, Perplexity, or Google AI Mode about your brand, 93.4% of the sources they cite are NOT your website. They are Reddit threads, review sites, news articles, and competitor blogs. Your brand's AI reputation is mostly written by other people.
There is a question every brand should be asking right now: when AI talks about us, where is it getting its information?
The answer, based on our analysis of 2,169 citations across three AI platforms, is uncomfortable. For every 1 time AI cites your own website, it cites someone else's website about 14 times. Reddit alone is the #1 source for brand queries in 18 out of 18 industry verticals we tested.
This post covers three things we discovered: (1) the third-party citation dominance and what drives it, (2) how Reddit sentiment directly predicts whether AI recommends or ignores your brand, and (3) why "one-size-fits-all" optimization advice is wrong because every industry has a completely different cited page profile.
🔍 THE CLAIM WE TESTED
In March 2026, AirOps (a company that sells AI optimization tools) published a widely-shared finding: 85% of brand mentions in AI responses come from third-party pages, not the brand's own website. This fueled an entire consulting industry around "LLM seeding," the idea that brands need to get mentioned on other websites to be visible in AI.
We wanted to verify this with our own data. Was 85% accurate? Was it higher? Lower? And more importantly, what kind of third-party sources dominate?
🧪 WHAT WE DID
We asked 50 brand-related questions to three AI platforms:
The questions (50 total across 18 industries):
- Validation questions (25): "Is [Brand] worth it?" / "Is [Brand] actually good?"
- Discovery questions (25): "Best [product category] for [use case]?"
- Comparison questions (mixed in): "[Brand A] vs [Brand B]"
The brands we tested were selected for variety in Reddit reputation:
- Brands Reddit loves: Fidelity, Bitwarden, ProtonVPN
- Brands Reddit dislikes: Robinhood, LastPass, Temu
- Brands Reddit is split on: Tesla, BetterHelp, Airbnb, HubSpot, Casper
The platforms: ChatGPT (paid), Perplexity (free tier), Google AI Mode
Total data: 150 scrapes (50 questions x 3 platforms), producing 2,169 citations. Every citation URL was classified as either first-party (the brand's own website) or third-party (everyone else).
📊 THE HEADLINE: 93.4% THIRD-PARTY
| What we measured | Result |
|---|---|
| Total citations analyzed | 2,169 |
| Citations from third-party sources | 2,025 (93.4%) |
| Citations from the brand's own website | 144 (6.6%) |
| AirOps claimed | 85% third-party |
| Our finding exceeds their claim by 8.4 percentage points |
AirOps was right that third-party sources dominate. But they actually understated the problem. The real number is 93.4%, not 85%. For every 1 citation a brand's own website receives, third-party sources receive about 14.
Which AI platform is friendliest to brands?
| Platform | Citations from brand's own site | Citations from third parties |
|---|---|---|
| Google AI Mode | 9.0% | 91.0% |
| Perplexity | 5.0% | 95.0% |
| ChatGPT | 4.2% | 95.8% |
Google AI Mode gives brands the best shot at first-party citation (9.0%), likely because Google's own search index ranks brand websites well for branded queries. ChatGPT is the toughest (4.2%). It appears to actively prefer third-party sources even when the brand's own site ranks.
Does the type of question matter?
| Question type | Brand's own site cited | Third parties cited |
|---|---|---|
| Discovery ("best X for Y") | 3.2% | 96.8% |
| Validation ("is X worth it") | 8.6% | 91.4% |
| Comparison ("X vs Y") | 9.6% | 90.4% |
Discovery questions are the worst for brands. When someone asks "what is the best password manager," AI platforms almost never cite the winning brand's own website. They cite review sites and comparison articles. Even comparison questions ("Bitwarden vs LastPass") only cite the brands themselves 9.6% of the time.
🏆 REDDIT IS THE #1 SOURCE FOR BRAND QUERIES
The single most-cited domain for brand-related AI queries is reddit.com.
| Domain | Citations | Type |
|---|---|---|
| reddit.com | 231 | UGC |
| youtube.com | 70 | UGC |
| forbes.com | 43 | Media |
| techradar.com | 22 | Review |
| chime.com | 22 | First-party |
| nerdwallet.com | 19 | Review |
| cnet.com | 19 | Review |
| pcmag.com | 16 | Review |
Reddit accounts for 10.6% of ALL brand-query citations. It appears as either the #1 or #2 cited domain in 18 out of 18 industry verticals we tested. The only first-party domain in the top 10 is chime.com, and that is likely because Chime's content is structured more like an educational resource than a product page.
Some industries are especially Reddit-dependent:
| Industry | Reddit's share of citations |
|---|---|
| Meal kits (HelloFresh, Blue Apron) | 29.1% |
| Pet products (Chewy, BarkBox) | 26.7% |
| Email services (ProtonMail, Gmail) | 26.7% |
| Mattresses (Casper, Purple) | 11.4% |
In meal kit and pet product categories, nearly 1 in 3 AI citations comes from a Reddit thread. What Reddit users say about HelloFresh IS what ChatGPT says about HelloFresh.
For our full analysis of Reddit's role in AI search, see Reddit's Influence on AI Search.
🤔 WHY BRANDS ALMOST NEVER GET CITED ON THEIR OWN SITES
This is not a failure of optimization. It is a structural problem. AI platforms prefer third-party sources for brand queries because those sources do four things a brand's own website cannot:
| What AI platforms want | Third-party sources | Brand's own website |
|---|---|---|
| Compare alternatives | Review sites compare 10 products side-by-side | A brand site only talks about itself |
| Unbiased perspective | Reddit users, journalists have no financial stake | A brand site is inherently promotional |
| Cover multiple brands | A "best of" article covers the whole category | A brand site covers one brand |
| Community validation | Reddit upvotes, user reviews signal real opinions | A brand site shows curated testimonials |
A brand's own website fails on all four dimensions. The problem is not that the site is poorly built. The problem is that AI platforms are looking for the kind of content brand sites structurally cannot provide.
The industry matters enormously
| Industry | Brand's own site cited | What to do |
|---|---|---|
| Finance (Fidelity, Robinhood) | 14.0% | Educational content works. Build reference-style resources on your own domain. |
| SaaS (HubSpot, Salesforce) | 12.5% | Documentation and comparison pages work. Be the reference for your category. |
| E-commerce (Shopify, Temu) | 12.0% | Product education works. Technical depth earns citations. |
| Mattress (Casper, Purple) | 0.8% | Almost impossible to get first-party citations. Focus entirely on review coverage and Reddit. |
| Audio (Bose, Sony) | 0.0% | Zero first-party citations in our data. Review sites and YouTube own this space. |
Finance and SaaS brands get their own sites cited 12 to 14% of the time because they publish substantial educational content that functions as a reference resource. Consumer product brands (mattresses, audio equipment) get essentially zero first-party citations. Their product pages are optimized for selling, not for answering the question "which product should I buy?"
🔴 PART 2: REDDIT SENTIMENT PREDICTS AI RECOMMENDATIONS
We expanded our analysis to 674 AI responses with full answer text to test something specific: does what Reddit thinks about a brand predict what AI platforms say about that brand?
The answer is yes, with a twist.
| Reddit's opinion of the brand | AI says positive things | AI says negative things | Ratio |
|---|---|---|---|
| Reddit LOVES the brand | 35% | 3% | 12:1 positive |
| Reddit is MIXED | 38% | 5% | 8:1 positive |
| Reddit HATES the brand | 19% | 10% | 2:1 positive |
The key insight: AI almost never goes negative
Even when Reddit universally dislikes a brand, AI platforms still say more positive things than negative things (19% positive vs 10% negative). AI platforms almost never explicitly warn you away from a brand.
The real effect of negative Reddit sentiment is not criticism. It is silence. Reddit-loved brands get recommended in 35% of AI responses. Reddit-hated brands get recommended in only 19%. The gap is not that AI says bad things about Reddit-hated brands. It is that AI stops recommending them at all.
Real examples from our data
Brands Reddit loves (and AI recommends consistently):
- Fidelity: 100% AI-positive (6 out of 6 responses)
- Vanguard: 100% AI-positive (3 out of 3)
- Pipedrive: 100% AI-positive (3 out of 3)
Brands Reddit hates (and AI is ambivalent about):
- TurboTax: AI is split (2 positive, 2 negative). Reddit overwhelmingly dislikes it.
- BetterHelp: AI is surprisingly favorable (3 positive, 2 mixed). Reddit has serious concerns.
- LastPass: AI still mentions it positively (2 out of 10 responses), despite Reddit unanimously recommending people switch after multiple security breaches. This suggests training data lag.
Real client brand data
We also tested three real-world client brands across 4 platforms (530 responses). The brands span three different verticals (skincare, advertising services, and marketing technology):
| Brand | Industry | AI mentions it | Positive | Top competitor in AI |
|---|---|---|---|---|
| Brand A | Skincare (DTC) | 20% of responses | 4% | The Ordinary (15%) |
| Brand B | Advertising Services | 8% of responses | 3% | Helium 10 (20%) |
| Brand C | Marketing Technology | 0.5% of responses | 0% | Semrush (19%) |
Google AI Mode consistently showed the highest brand mention rates across all three clients. Claude was consistently the lowest. ChatGPT showed the most variance (21% for Brand A but 0% for Brand B).
The competitive landscape revealed by AI responses is telling. For advertising service queries, AI platforms recommend software tools (Helium 10, Jungle Scout, Pacvue), not service agencies. For marketing technology queries, AI recommends established SaaS tools (Semrush, Surfer SEO), not newer companies. AI platforms cite tools over services and established players over newcomers.
📏 PART 3: EVERY INDUSTRY HAS A DIFFERENT "CITED PAGE" PROFILE
We merged our data with the Experiment M dataset to build profiles of what cited pages actually look like in each industry. The sample: 8,043 unique pages across 14 industries.
The finding: one-size-fits-all advice is wrong for almost every industry.
| Industry | Typical cited page length | Statistics per 1,000 words | FAQ schema usage | Technical jargon |
|---|---|---|---|---|
| Consumer Electronics | 3,422 words | 6.3 | 7% | Low |
| Health | 3,302 words | 7.0 | 12% | Very low |
| E-commerce | 3,317 words | 9.0 | 17% | Medium |
| Technology | 3,095 words | 4.8 | 11% | High (2.0/1k) |
| SaaS/B2B | 3,040 words | 6.2 | 23% | High (2.5/1k) |
| Finance | 2,841 words | 11.2 | 21% | Low |
| Automotive | 1,866 words | 3.8 | 8% | None |
| Pet | 2,235 words | 5.7 | 14% | None |
What this means in plain English
Word count varies 2x. Automotive cited pages are about 1,900 words. Consumer Electronics cited pages are about 3,400 words. The generic advice "write 2,000 words" is wrong for 10 of 14 industries.
Statistics density varies 3x. Finance pages contain 11.2 data points per 1,000 words. Automotive pages contain only 3.8. If you write finance content without numbers, it will not get cited. If you write automotive content, numbers matter much less.
FAQ schema usage varies 6x. SaaS/B2B pages use FAQ schema 23% of the time. Fitness pages use it 4% of the time. FAQ schema is a strong play in SaaS and Finance, but barely used in consumer product content.
Technical jargon is industry-specific. SaaS (2.5 per 1,000 words), Technology (2.0), and E-commerce (2.6) use significant technical language. Pet (0.0), Food (0.0), and Automotive (0.0) use none at all. Technical depth helps in tech industries and is irrelevant in consumer industries.
For our full vertical-by-vertical optimization guide, see AI SEO by Vertical.
✅ WHAT BRANDS SHOULD ACTUALLY DO
1. Accept that third-party sources control your AI narrative
93.4% of brand-query citations come from other websites. This is not going to change through website optimization alone. Your AI reputation is a PR problem, not an SEO problem.
2. Invest in earned media (especially for consumer brands)
Get reviewed by the sites AI platforms cite. Our review site analysis found that garagegymreviews.com and trustpilot.com are the only two domains cited by ALL five access methods we tested. Techradar, Wirecutter, PCMag, NerdWallet, and RTINGS are consistently cited across platforms.
Trustpilot is the easiest win. It is free to claim your profile and it is cited by all five access methods.
3. Take Reddit seriously
Reddit is the #1 cited domain for brand queries. What Reddit says about your brand IS what AI says about your brand. This means genuine, long-term community participation. Not astroturfing. Not paid posts. Real users sharing real opinions about your product.
For the complete Reddit strategy, see Reddit Strategy for AI Visibility.
4. Match your strategy to your industry
Finance and SaaS brands (12 to 14% first-party rate) can invest in first-party educational content that gets cited directly. Build reference-style resources, comparison pages, and data-driven guides on your own domain.
Consumer product brands (0 to 5% first-party rate) should focus almost entirely on third-party coverage. Your budget is better spent on review site outreach and Reddit/community presence than on optimizing product pages for AI.
5. Monitor what AI is actually saying about you
AI platforms rarely say negative things directly. The real risk is silence: they stop recommending you. Track your brand's mention rate and sentiment across ChatGPT, Perplexity, Claude, and Google AI Mode. For monitoring tools, see our AI Citation Monitoring comparison.
For a free check of your page-level AI optimization: AI Visibility Quick Check
❓ FREQUENTLY ASKED QUESTIONS
Is 93% third-party really accurate?
Yes. We classified all 2,169 citations by matching URL domains against target brand domains. 144 matched (6.6% first-party), 2,025 did not (93.4% third-party). This exceeds AirOps' claim of 85% third-party by 8.4 percentage points. The pattern held across all 3 platforms and all 18 industry verticals.
Can I get my brand's own website cited more?
It depends on your industry. Finance and SaaS brands achieve 12 to 14% first-party citation rates by publishing substantial educational content (not product pages). Consumer product brands (mattresses, audio, pet products) get 0 to 5%. The structural problem is that AI platforms prefer sources that compare multiple options, and brand websites only cover one option.
Does Reddit sentiment really affect AI recommendations?
Yes. Brands Reddit loves get 12:1 positive-to-negative AI mentions. Brands Reddit hates get only 2:1. But the mechanism is not that AI criticizes disliked brands. AI platforms rarely go explicitly negative. Instead, they stop recommending you. Negative Reddit sentiment suppresses positive AI mentions rather than generating negative ones.
What if Reddit has wrong information about my brand?
This is a real risk. We found that AI still mentions LastPass positively in some responses, despite Reddit unanimously recommending people switch after multiple security breaches. This suggests training data lag. The reverse is also possible: outdated negative Reddit threads could suppress your brand even after you have fixed the issue. Genuine, ongoing Reddit engagement is the only way to update the narrative over time.
Should I hire an "LLM seeding" agency?
The data supports the premise (third-party sources dominate AI citations). But the mechanism is not mysterious. It is traditional PR and earned media, with AI platforms as the audience instead of journalists. Getting mentioned on review sites, building genuine Reddit presence, and earning media coverage all contribute. Whether you need an agency for this depends on your team's PR capabilities.
🔬 METHODOLOGY NOTES
Part 1 (Third-party analysis):
- 50 queries across 18 verticals, 3 AI platforms, 2,169 total citations
- Brand domain matching: heuristic (e.g., "HubSpot" -> hubspot.com)
- Citation source types: First-party, UGC/Social, Review/Authority, News/Media, Other
- Collection period: April 3 to 5, 2026
Part 2 (Sentiment analysis):
- 674 responses with full answer text (144 from sentiment re-run + 530 from VPS database)
- Sentiment classification: regex-based NLP (positive/negative/mixed/neutral signal extraction)
- 100+ brands across 18+ verticals, 3 real client brands across 4 platforms
Part 3 (Vertical profiles):
- 8,043 unique pages (4,544 from Experiment M + 3,535 newly crawled cited pages)
- 14 canonical verticals normalized from 40+ sub-verticals
- 43 features per page including word count, headings, schema, statistics density, technical terms
📚 REFERENCES
- Lee, A. (2026c). "I Rank on Page 1: What Gets Me Cited by AI?" Preprint. Paper | Dataset
- Lee, A. (2026a). "Query Intent, Not Google Rank: What Best Predicts AI Citation Behavior." Preprint v5. DOI
- Lee, A. (2026b). "Reddit Doesn't Get Cited (Through the API)." Preprint v3. DOI
- AirOps (2026). "85% of Brand Mentions in LLM Responses Come From Third-Party Pages." Industry Report.
- Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., & Deshpande, A. (2024). "GEO: Generative Engine Optimization." KDD 2024. DOI