Why Your AI Visibility Fix Works on One Platform and Fails on Another

Most AEO advice treats AI search as a single channel. Fix your schema. Build directory presence. Publish authoritative content. Then collect the improvement -- on all platforms, together.

That's not how the underlying systems work.

In tracking AI citation behaviors across platforms (knowledge/platform-citation-behaviors.md, Scout, May 2026 through July 2026), we analyzed the most comprehensive third-party dataset available on the question: Yext's 2026 AI Visibility study, which catalogued 17.2 million citations across ChatGPT, Perplexity, and Gemini. Its core finding changes how any serious fix plan should be structured.

Each platform gets the majority of its business citations from a different source type:

- ChatGPT: third-party directories account for 49% of citations - Gemini: brand-owned websites account for 52% of citations - Perplexity: niche expert directories, not general-purpose ones

A fix targeted at directories improves ChatGPT and may do nothing for Gemini. A fix targeted at brand site structure improves Gemini and may not touch ChatGPT. A fix that helps both of those may still miss Perplexity entirely. These are not parallel tracks that eventually converge -- they require different interventions.

One caveat worth stating upfront: Yext is a directory and listings company, so their methodology may have a slight bias toward overweighting directory presence. The platform-specific breakdown is still the most granular public dataset available on this question, and it aligns with what we observe in our own audit data.

ChatGPT cites from third-party directories

ChatGPT draws primarily from its training data. When answering a business query, it resolves the entity by pattern-matching across sources that were indexed before its training cutoff. The sources that dominate that resolution process are the ones most heavily and consistently represented in its training corpus.

Third-party directories fill that role at a 49% citation share. Yelp, the Better Business Bureau, G2, Capterra, and industry-specific registries form the citation foundation for most business categories. The brand's own website -- regardless of how well-built it is -- accounts for a materially smaller share.

We see this in Sourcepull audit data as well. ChatGPT misattribution rates drop for businesses with strong directory presence, because directory citations provide a stable entity anchor for the model. A business with consistent name, address, and phone data across Yelp, BBB, and two industry directories gives ChatGPT a well-grounded entity to cite. A business with a well-structured brand site and thin directory coverage gives ChatGPT weaker resolution material.

The practical implication: a client whose ChatGPT score is low and Gemini score is high does not need website work. They need directory presence -- specifically in the directories that appear in ChatGPT's citation patterns for their category, not just the most popular general-purpose platforms.

Gemini cites from brand-owned websites

Gemini shows the inverse weighting. Brand-owned websites account for 52% of Gemini citations in the Yext data. Gemini is more Google-native in its architecture: it indexes and weights brand domains, structured data, E-E-A-T signals, and Google Business Profile data more heavily than its peers.

In audit comparisons, a business with a well-structured, schema-annotated website and thin directory coverage consistently scores better on Gemini than on ChatGPT. A business with strong Yelp and BBB presence but a poorly-structured website tends to score better on ChatGPT than on Gemini. The correlation holds across the audits we've run.

For Gemini specifically, Google Business Profile is the highest-leverage single fix. GBP data feeds directly into Gemini's local responses. Business categories, service areas, hours, and service descriptions inside GBP are inputs Gemini uses when answering local queries. Schema on the brand website reinforces those GBP signals -- structured data telling Gemini what the entity does, where it operates, and what services it offers amplifies the GBP signal rather than substituting for it.

A client asking "how do I improve my Gemini visibility" has a materially different fix plan than a client asking the same question about ChatGPT. Conflating the two produces a fix plan that partially works on neither.

Perplexity cites from neither of those

Perplexity is a live retrieval system. It fetches approximately 10 pages per query and synthesizes citations from what it retrieves in real time. That structural difference changes what "fixing visibility" means.

First, the directories that matter for ChatGPT -- general-purpose, high-authority platforms like Yelp, BBB, and Angi -- are not primarily where Perplexity goes. Per the Yext breakdown and our own directory research (knowledge/directory-presence-impact.md, April 2026, updated through July 2026), Perplexity favors niche expert directories specific to the industry: legal directories for law firms, medical registries for healthcare, fintech databases for financial firms, trade-specific registries for contractors. Yelp still appears in Perplexity citations across industries -- our research documents it showing up in every industry tested. But in competitive verticals, general directory presence alone is insufficient; the niche expert databases are the differentiating layer.

Second, crawlability matters more for Perplexity than for any other platform we track. Because Perplexity retrieves live pages, a business website that blocks PerplexityBot in robots.txt, or sits behind a Cloudflare policy restricting AI crawlers, will not appear in the citation pool -- regardless of how good the schema is or how many directories list the business. For ChatGPT, a blocked crawler is a reduced training signal; for Perplexity, it is a complete exclusion.

Third: Reddit's citation share on Perplexity dropped approximately 86% after October 2025, following litigation over unauthorized scraping. Any fix plan still recommending Reddit as a Perplexity visibility lever is working from outdated assumptions. The niche directory strategy has functionally replaced Reddit as the primary Perplexity lever for most business categories.

The 87% contractor problem makes this concrete

Our tracking of home services AI citation data (knowledge/home-services-ai-citation-data.md, Scout session 45, created June 2026, updated July 20, 2026) documents what platform divergence looks like in a real market.

The 5W HVAC & Plumbing AI Visibility Index 2026 found that 87% of independent HVAC and plumbing contractors have zero AI citation share in their own metro -- even those with 800 or more five-star Google reviews. National chains (Roto-Rooter, ARS Rescue Rooter, Mr. Rooter) hold approximately 19% of all consumer-intent AI citations across the vertical combined. Local contractors with decades of community presence are cited at a 1.2% rate across ChatGPT's local business citation pool.

The explanation is embedded in the platform breakdown. Google Reviews are rendered via JavaScript and are largely inaccessible to ChatGPT, Perplexity, and Claude crawlers. Review volume doesn't transfer into AI citations unless those reviews also exist on AI-indexable platforms like Yelp and BBB. A contractor with 800 Google Reviews and no Yelp profile has 800 reviews that the citation layer cannot reach.

National chains have strong directory coverage, well-structured brand websites, and complete GBP profiles. They're visible on all three platforms for different reasons. Local contractors strong on Google Reviews but weak on directory coverage are invisible to ChatGPT and underperforming on Perplexity -- not because they can't compete, but because they're not listed where ChatGPT looks, and their site structure isn't optimized where Gemini looks.

This extends beyond HVAC and plumbing. Our research found the same pattern in roofing, electrical, and general contracting. Any independent home service contractor faces the same citation concentration problem, and the fix plan differs per platform.

What this means for building a fix plan

Per-platform scores are the starting point, not an aggregate. An aggregate AI visibility number can mask a business that is genuinely visible on Gemini but invisible on ChatGPT. The platform gap identifies where to start.

ChatGPT gaps point to directory work -- specifically the directories that appear in ChatGPT's citation patterns for the business category. Yelp and BBB are baseline; the right answer also includes the industry-specific registries that dominate ChatGPT's citation patterns for that vertical.

Gemini gaps point to brand site structure, schema, and GBP completeness. Of the platforms we track, schema investment has the clearest return on Gemini. GBP category and service accuracy is the lever most businesses haven't pulled.

Perplexity gaps require identifying the niche expert directories for the category. Crawlability verification comes first -- a schema fix on a bot-blocked site has no citation impact on a live-retrieval platform. Then build content that's accessible, fresh, and structured clearly enough to survive Perplexity's synthesis layer.

A Sourcepull Signal Check breaks out scores per platform and surfaces which gap is largest -- the answer is often different from what business owners expect, and it's usually not the same platform where they thought they were weakest.

---