Do AI Platforms Recommend the Same Brands? A 2,395-Answer Study

If you've checked how one AI platform talks about your brand, it's tempting to assume the others say roughly the same thing. We wanted to measure whether that's true, so we asked four major AI platforms the same 200 buyer-intent questions, three times each, and compared every answer.

They mostly don't agree. And the gap between Perplexity and the rest is wide enough that "we rank well in AI search" doesn't mean much until you say which AI.

What we did

We wrote 200 buyer-intent questions, 20 in each of 10 categories: SaaS, local services, ecommerce, finance, health, B2B services, travel, education, home services, and professional services. None of them named a brand. They're the kind of question a buyer actually types: "Which payroll provider is best for a business with 8 hourly employees?" or "What's the best-rated dentist for a nervous patient in Denver, CO?"

Each question went to ChatGPT, Claude, Gemini, and Perplexity three times, with web search enabled, between September 4 and September 24, 2026. That's 2,400 requests, and 2,395 of them returned a usable answer. For each answer we extracted the ranked list of brands it recommended and every source it cited. The full method and its limits are at the end.

Finding 1: the platforms rarely agree on the #1 brand

The simplest question: when a buyer asks for a recommendation, do the platforms put the same brand first?

Pair by pair, agreement on the top pick looks like this:

Even the closest pair, Gemini and ChatGPT, puts a different brand first almost half the time. Any pairing with Perplexity agrees on the top pick about one time in five.

Finding 2: Perplexity recommends a different set of brands

The top pick is only part of an answer. We also compared the full set of brands each platform named for the same question, using Jaccard overlap: the share of brands two answers have in common out of all the brands either one named. A score of 1.0 means identical lists; 0 means nothing in common.

ChatGPT, Claude, and Gemini form a loose cluster. Perplexity sits well apart from all three: roughly five out of every six brands it names are ones the other platform in the pair didn't mention.

There's an important detail behind that cluster. In this study, ChatGPT, Claude, and Gemini got their web results from the same search tool, so they often saw the same sources (identical source lists on 397 of 597 comparable answers). Perplexity ran its own search. Two things follow:

• Sources drive a large share of what gets recommended. The platforms retrieving from the same web results converged; the one retrieving on its own went its own way.
• Even with the same sources, the models still disagree. Given largely identical search results, ChatGPT, Claude, and Gemini still left out about 40% of each other's brands and picked different #1 brands most of the time (all three agreed on the top pick on 33% of prompts). How each model reads and ranks the same evidence matters too.

In the consumer apps, each platform runs its own search stack, so real-world divergence between ChatGPT, Claude, and Gemini is unlikely to be smaller than what we measured here.

Finding 3: Perplexity's answers change the most between runs

We asked every question three times per platform and compared each platform's answers with its own earlier answers.

Even the steadiest platforms changed about a quarter of their brand list between identical requests minutes apart. Perplexity changed well over half. A single check of any AI platform is a sample, not a measurement. It takes repeated checks over time to know where you actually stand.

Finding 4: most of Perplexity's citations point to third-party sites

Perplexity cited sources heavily: 8,970 citations across its 598 answers, about 15 per answer. Only 19.6% pointed to the websites of brands it recommended. The other 80.4% went to third parties.

Its most-cited domains:

App store listings, publisher roundups, Reddit threads, and review sites do much of the work. If Perplexity matters to your buyers, what those sources say about you likely matters as much as what your own site says.

We report citations for Perplexity only. Because ChatGPT, Claude, and Gemini shared a search tool in this study, their citation lists reflect that tool rather than how each platform cites sources in its own app.

Finding 5: agreement depends heavily on the industry

Health and B2B services, where a few well-known names dominate, saw the most agreement. Local and professional services saw the least. Those answers depend on geography and fragmented markets, so each platform assembles its own shortlist. With 20 prompts per category, treat these as directional rather than precise.

The platforms also differed in how many brands they name. Claude averaged 5.1 brands per answer, Gemini 4.6, Perplexity 4.0, and ChatGPT 3.9. On shorter lists, missing the cut is more likely.

What this means for your brand

• Measure each platform separately. An 11.5% top-pick agreement rate means your standing on one platform tells you little about the others. ChatGPT, Perplexity, and Gemini each need their own read.
• Treat Perplexity as its own channel. It recommends different brands, changes its answers more often, and leans on third-party sources. Reviews, app store listings, Reddit discussions, and publisher roundups carry real weight there.
• Get into the sources, not just onto your own site. The platforms that shared search results converged on similar brands. Being present in the pages AI platforms retrieve is one of the most direct ways to influence what they say.
• Track over time, not once. With run-to-run stability between 0.37 and 0.75, a single check can mislead in either direction. Trends over weeks are what show whether your work is moving the answer.
• Know your category's baseline. In local and professional services, near-total disagreement is normal. Winning your market across platforms means building visibility platform by platform.

Want to see where you stand right now? Run a free AI visibility check across ChatGPT, Perplexity, and Gemini. For daily tracking across the major AI platforms we cover (and growing), see DabaRank's plans.

Get the data

The aggregate results are free to use with attribution to DabaRank and a link to this page:

• Brand overlap by platform pair (CSV)
• #1-brand agreement (CSV)
• Run-to-run consistency (CSV)
• Average brands named per answer (CSV)
• Perplexity top cited domains (CSV)
• Per-category breakdown (CSV)

Methodology and limitations

Prompts. 200 neutral, non-branded buyer-intent questions, 20 per category across 10 categories. Location-based prompts use a fixed list of cities.

Platforms and models. Each question went through each platform's API with web search enabled: ChatGPT (gpt-4o-mini), Claude (Claude Haiku 4.5), Gemini (Gemini 2.5 Flash-Lite), and Perplexity (Sonar), at temperature 0.7, with the same instruction asking each model to list its recommended brands in ranked order. For ChatGPT, Claude, and Gemini, the same web-search tool supplied results; Perplexity used its own search.

Repeats. Each question ran three times per platform, 2,400 requests in total, of which 2,395 returned a usable answer. Cross-platform comparisons (findings 1, 2, and 5) use each platform's first successful answer per question, so repeats don't double-count. Run-to-run consistency (finding 3) compares all repeats for the same platform and question.

Measures. Brand overlap is the Jaccard similarity of the brand sets two answers named. #1 agreement compares the first-ranked brand after normalizing names. Citation share classifies each cited domain as either the site of a brand named in that answer (an approximate match of the domain against those brand names) or a third party.

Limitations.

• API answers, not app answers. The consumer apps may use different or larger models, personalization and memory, and different search systems. Individual users can see different answers from the ones we recorded.
• Shared retrieval for three platforms. Because ChatGPT, Claude, and Gemini shared a search tool here, their mutual overlap likely overstates how similar they are in their own apps, and we excluded their citation data.
• Automated brand extraction. Brands were parsed from each answer's ranked list and normalized automatically. Unusual spellings or sub-brands can occasionally be split or merged.
• A snapshot in time. Models and search indexes change constantly. These results describe September 2026.
• Sample size per category. 20 prompts per category is enough to show direction, not to rank industries precisely.