How AI Models Decide Which Brands to Recommend (And Why Yours Might Not Make the List)

Forget everything you know about how Google ranks websites. AI models don’t maintain a ranked index of pages. They don’t care about your Domain Authority. They don’t follow backlinks the way Googlebot does. Instead, they do something far more interesting — and far harder to game: they build consensus.

The consensus layer is the mechanism by which large language models aggregate information from dozens — sometimes hundreds — of independent sources to form a coherent understanding of a brand, product, or topic. When someone asks ChatGPT "What’s the best project management tool for remote teams?", the model doesn’t look at one page. It synthesizes everything it has absorbed during training and retrieval: Reddit discussions, Quora answers, G2 reviews, TechCrunch articles, Wikipedia entries, comparison blog posts, and YouTube transcripts. If multiple independent sources agree that a particular tool is strong, that signal gets amplified. If only the brand’s own website makes the claim, it gets discounted. Our AI Visibility Index study proved this empirically: the top 10 most visible businesses — Asana (score: 91), Zoho CRM (87), Jira (86), Pipedrive (85), Monday.com (85), HubSpot CRM (84), ClickUp (83), Trello (81), WebFX (81), and The Dominguez Firm (73) — all have massive third-party footprints across Reddit, reviews, and press. Every single one appeared in all 5 AI models tested.

According to Joel House, founder of MentionLayer and author of AI for Revenue, "The consensus layer is the single biggest paradigm shift since PageRank. Brands that built their entire strategy around controlling their own pages are now discovering that AI doesn\'t care about owned media — it only trusts what other people say about you."

This is fundamentally different from Google’s PageRank model. PageRank asks: "Who links to this page?" The consensus layer asks: "What do diverse, independent sources agree about?" A brand with 10,000 backlinks but zero Reddit mentions, zero press coverage, and zero third-party reviews is practically invisible to the consensus layer. Meanwhile, a brand with 50 genuine Reddit discussions, a handful of press articles, and strong review presence across G2 and Trustpilot becomes part of what AI considers established knowledge.

Search Engine Land’s research on this topic nailed it: AI models weight information proportionally to how many independent, credible sources corroborate it. One source saying you’re great is noise. Fifteen unrelated sources saying you’re great is signal. That’s the consensus layer in action, and it’s the single most important concept in Generative Engine Optimization.

The practical implication is massive. Traditional SEO let you control the narrative because you controlled your pages. AI doesn’t care about your pages. It cares about what everyone *else* says about you. If you’ve spent the last decade optimizing your website and ignoring your earned media footprint, you’ve been building a castle on a foundation that AI doesn’t even look at. Want to see where you stand? Run the 60-second AI visibility test right now.

Not all sources carry equal weight in the consensus layer. Through extensive testing across ChatGPT, Perplexity, Gemini, and Claude, we’ve mapped out what we call the AI Trust Hierarchy — a ranking of source types by how much influence they have on AI recommendations.

At the top sits earned media, which accounts for roughly 90% of all citations in AI-generated answers. This includes news articles, industry publications, analyst reports, and editorial mentions from sites the AI considers authoritative. When Perplexity answers a question about CRM software, it’s pulling from TechCrunch reviews, Forbes roundups, and industry comparison articles — not from Salesforce’s homepage. The reason is straightforward: earned media represents third-party validation. An editorial team decided your brand was worth writing about, which is a powerful trust signal.

"We\'ve tracked over 14,000 AI-generated brand recommendations, and the pattern is unmistakable: 90% of citations trace back to earned media," says Joel House. "If you\'re spending 80% of your marketing budget on paid channels and 5% on earned media, your AI visibility budget is inverted."

Wikipedia and Wikidata occupy the second tier. ChatGPT specifically leans heavily on Wikipedia — 47.9% of its citations reference Wikipedia content. If your brand has a Wikipedia page with accurate, well-sourced information, you’ve got a significant advantage. Wikidata is equally important because it provides the structured entity data that AI models use to understand relationships between brands, categories, and concepts.

[Reddit and community forums](/blog/reddit-most-important-platform) sit in the third tier, but their influence is growing fast. Perplexity cites Reddit in 46.7% of its top results. For product-related queries, Reddit appears in 95% of Google’s results (which Gemini and ChatGPT also reference). The reason AI trusts Reddit is that it represents authentic user opinions with built-in quality signals (upvotes, comment depth, community moderation). A heavily-upvoted Reddit comment recommending your product carries more weight than a sponsored blog post.

Review platforms like G2, Trustpilot, and Capterra form the fourth tier. AI models reference these when answering comparison queries because review data provides structured, quantifiable sentiment. Star ratings, review volume, and review recency all feed into the consensus. Brands with strong review profiles get mentioned; brands with thin or absent review presence get overlooked.

[Schema markup and structured data](/blog/schema-markup-ai-search) play a supporting role. While AI models don’t directly parse your schema the way Google does, the structured data helps AI understand what your brand *is* — its category, products, location, and relationships. Organization schema, Product schema, and FAQ schema all contribute to entity clarity.

At the bottom of the hierarchy sits brand-owned content — your website, your blog, your social media profiles. This isn’t zero-weight, but it’s minimal. AI models are specifically designed to discount self-promotional content. Your About page saying you’re "the leading provider" carries almost no influence compared to a TechCrunch article that independently reaches the same conclusion. For the complete data on which sources each AI platform cites most, see the AI Citation Index.

Here’s a stat that should change how you think about AI visibility: only 6% of AI brand mentions result in actual recommendations. And of all brand mentions across AI platforms, only 31% are positive. The rest are neutral (listed among alternatives), mixed, or negative.

This is what we call the 6% problem. Getting mentioned by AI is hard enough. Getting *recommended* is an order of magnitude harder. There’s a massive gap between "ChatGPT knows your brand exists" and "ChatGPT tells users to choose your brand."

"In our experience running AI visibility campaigns at MentionLayer, we\'ve found that most brands fixate on getting mentioned at all, when the real game is crossing the threshold from mention to recommendation," says Joel House. "That threshold requires specificity of praise across diverse source types — generic mentions from a single platform won\'t cut it."

So what separates a mention from a recommendation? We’ve analyzed thousands of AI responses and identified the specific signals that trigger recommendation language — phrases like "I’d recommend," "your best option," "the top choice for," and "stands out because." The pattern is consistent across models.

Specificity of praise is the biggest differentiator. When AI mentions a brand neutrally, it uses generic language: "Brand X is a project management tool." When AI recommends a brand, it cites specific advantages: "Brand X is particularly strong for remote teams because of its async communication features and timezone-aware scheduling." That specificity comes from source material that contains specific praise — Reddit comments that explain *why* they love the product, reviews that detail particular use cases, press articles that highlight differentiators.

Frequency across source types is the second signal. A brand mentioned on Reddit AND in reviews AND in press articles triggers recommendation language more often than a brand mentioned only in one source type. The consensus layer doesn’t just count mentions — it evaluates the *diversity* of mentions. Five Reddit threads plus three G2 reviews plus one press article beats twenty Reddit threads alone.

Recency and consistency round out the formula. AI models weight recent sources more heavily. A brand that was praised two years ago but has no recent mentions gets listed but not recommended. And if recent mentions contradict older ones (quality decline, service issues), AI picks up on the inconsistency and downgrades the recommendation.

The takeaway: don’t just aim for AI mentions. Aim for recommendation-quality presence across diverse source types with specific, positive, recent content. That’s the 6% target. Tracking your progress on this is what the Share of Model metric is designed for.

If your brand isn’t showing up in AI recommendations, it’s almost certainly because of one or more of these five gaps. I’ve audited over 200 brands at this point, and the same problems show up again and again.

1. No third-party mentions. This is the number-one killer. Your website can be perfect. Your SEO can be strong. But if nobody else is talking about you — on Reddit, in press, in reviews — AI has nothing to build consensus from. We audited a SaaS company with $5M ARR and great Google rankings. They had zero Reddit threads discussing their product, zero press mentions in the last 12 months, and 4 total reviews across all platforms. To AI models, they barely existed. Their competitor with $2M ARR but an active Reddit community and regular press coverage was recommended far more often. The AI Visibility Index data tells the same story at scale: 71-73% of SaaS businesses in our 200-company sample were completely invisible to AI, while brands like Asana (score 91), Zoho CRM (87), and HubSpot CRM (84) dominated because they have thousands of Reddit discussions, hundreds of G2 reviews, and consistent press coverage.

2. Inconsistent [entity authority](/blog/what-is-entity-authority-ai). Your LinkedIn says you’re a "marketing platform." Your Google Business Profile says "advertising agency." Your Crunchbase listing says "ad tech company." Your website says "growth marketing solution." AI models process all of these descriptions, and when they conflict, the model can’t form a clear consensus about what you actually are. The result: you get skipped in favor of brands with consistent entity data. We’ve seen entity inconsistency meaningfully drop a brand’s AI mention rate.

3. Missing from review platforms. G2, Trustpilot, and Capterra are among the most-cited sources in AI product comparisons. If your competitors have 200+ reviews on G2 and you have zero, you’re invisible in an entire category of AI queries. It’s not just about star ratings — it’s about presence. Having 15 reviews at 4.2 stars beats having zero reviews at any rating.

4. No press or earned media coverage. Remember that 90% stat? If you have no press mentions, feature articles, podcast appearances, or thought leadership pieces, you’re missing the single largest input to AI recommendations. A brand without earned media is like a candidate with no references — nobody will vouch for you, so you don’t get the job.

5. [Website blocking AI crawlers](/blog/robots-txt-ai-crawlers). This one is surprisingly common and incredibly damaging. Some brands have robots.txt rules that block AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended). Others use aggressive bot detection that blocks anything that isn’t a standard browser. If AI can’t access your website content, it can’t reference your product descriptions, pricing pages, or feature lists — even when other sources mention you. Check your robots.txt and server logs. You might be blocking the very systems you need to embrace.

Building AI consensus isn’t a single tactic — it’s a systematic campaign across six pillars. The AI Visibility Audit framework gives you the full scoring methodology, but here’s the strategic overview of how to build the consensus that gets you recommended.

Start with the two pillars that matter most. Our AI Visibility Index study across 1,004 businesses proved that Domain Authority (r=0.337) and Google Review Count (r=0.333) are virtually tied as the strongest predictors of AI visibility — together they explain more variance than every technical signal combined. High DA with low reviews (37% visible) beats low DA with high reviews (26% visible), making DA the foundation and reviews the multiplier. The study also identified a "1,000 Review Cliff" where visibility jumps from 39% (at 500 reviews) to 54% (at 1,000+). Until you cross that threshold, other optimizations have limited ceiling.

Start with the pillar where you have the biggest gap relative to competitors. For most brands, that’s citations (Reddit/Quora presence) or press coverage. These are the highest-leverage pillars because they’re both direct inputs to AI models AND they influence other pillars. A press article that mentions your brand also generates backlinks (entity signal), gets shared on Reddit (citation signal), and provides the kind of third-party validation that AI weights most heavily.

The fastest path to AI visibility is [citation seeding](/blog/citation-seeding-playbook). The Citation Seeding Playbook covers this in detail, but the core idea is simple: find the Reddit and Quora threads that AI models already reference for your industry keywords, then contribute genuine, valuable responses that naturally mention your brand. This works because you’re not trying to create new content that AI might discover someday — you’re placing your brand directly into the sources AI already trusts. See how the full audit-discover-seed-monitor cycle works inside MentionLayer.

Entity consistency is a quick win. Audit your brand’s presence across Google Business Profile, LinkedIn, Crunchbase, Wikipedia, and your top 5 industry directories. Make every description consistent: same business category, same value proposition, same founding year, same headquarters. Add Organization schema to your website. This won’t directly trigger recommendations, but it removes confusion that actively blocks them.

Reviews and [press](/blog/digital-pr-ai-era) take longer but compound. Getting your first 20 reviews on G2 takes weeks of outreach. Landing your first feature article might take a month of pitching. But these investments compound: each review and each press mention is a permanent addition to the consensus layer. Unlike paid ads that stop the moment you stop spending, earned media keeps working indefinitely.

Expect 4-8 weeks minimum for measurable AI presence changes. AI models don’t update their knowledge in real-time. ChatGPT’s training data refreshes periodically. Perplexity pulls live search results but still caches. Gemini uses Google’s index which has its own crawl schedule. After you seed citations and earn press, you need to wait for the models to absorb the new information. Run your Share of Model tracking weekly to catch the inflection point.

The brands that win in the AI era aren’t the ones with the best websites. They’re the ones with the broadest, most consistent earned media footprint. That’s the consensus that AI rewards.

Want to see how much consensus AI has actually built around your brand? Our free AI visibility audit checks where you stand across the sources AI trusts — earned media, Reddit, reviews, and entity signals — and emails you the results in about 20 minutes.