ChatGPT、Perplexity、Gemini 和 Google AI Overviews 會推薦哪些品牌？200 次回答分析

AI-generated: an AI agent drafted this post from real runs of a tool I built with it. All the numbers come from 200 answers collected on 26 October 2026, asked from the United States.

More and more buyers ask an AI assistant "what's the best X?" before they open a search engine. So I asked four of them the same 50 buying questions, and counted which brands they named, in which order, and which websites they relied on.

The setup

• 10 categories: CRM, project management, email marketing, website builders, password managers, VPNs, note-taking apps, web hosting, running shoes and language-learning apps.
• 5 questions per category, written the way buyers type them: "best CRM for small business", "best free password manager", "cheapest website builder"... In 9 categories, one of the five was "[leader] alternatives". It names the brand, so I left it out of the rankings below. That leaves 41 open questions.
• 4 engines: ChatGPT with web search, Perplexity (sonar), Gemini and Google AI Overviews. Google showed an AI Overview for 48 of the 50 questions.
• One leader and 7 or 8 competitors tracked per category, for example HubSpot against Salesforce, Pipedrive, Zoho and four others.

The data comes from AI Visibility Tracker, an Apify Actor I built. It returns one row per question and engine, with the brands named, their order, the sources and the answer.

1. The engines mostly agree on who comes first

On 17 of the 41 open questions, all four engines named the same brand first. On 15 more, three of the four did. For most buyer questions, there is one brand the AI layer thinks of first, whatever the assistant.

They split on the vaguest questions. For "best note-taking app", Google put OneNote first, ChatGPT Obsidian, Perplexity OneNote and Gemini Apple Notes. For "most secure VPN", the four answers started with Proton VPN, Mullvad, NordVPN and Mullvad.

2. Being the best-known brand is not enough

HubSpot, 1Password and Wix own their categories: they are named in almost every answer and first in most of them. Other famous brands are named but rarely recommended first:

• Mailchimp appears in only 4 of 15 email-marketing answers. Klaviyo wins "for ecommerce" on all four engines, Brevo wins "free" and Beehiiv wins "for creators".
• Bluehost is named in half the hosting answers and never first. Hostinger comes first in 14 of 16.
• Nike is named in 13 of 19 running-shoe answers, but first only once. Asics and Brooks lead.

3. The question decides the winner

Add one word to the question and the winner changes:

• "free": Bitwarden for password managers, ClickUp for project management and OneNote for notes, each first on all four engines. Brevo comes first on three of the four for email;
• "cheap": Hostinger for hosting and for website builders, and Surfshark for VPNs, all on four of four;
• "for teams", "for creators", "for ecommerce": Notion, Beehiiv and Klaviyo, each first on all four engines.

If you are not the category leader, the questions with a qualifier are the ones you can win.

4. Each engine reads different websites

"Brand websites" are the sites of the brands I tracked.

• Google's AI Overviews lean on video and forums. YouTube is cited in 44 of the 48 overviews, Reddit in 25.
• ChatGPT reads the brands' own pages. 43% of its sources are brand websites, and 49 of its 50 answers cite at least one.
• Perplexity reads review sites and the press: PCMag, Forbes, Zapier and TechRadar.
• Gemini cites the fewest sources, and only 7% of them are brand websites.

Across all engines, the most cited website is zapier.com. Its "best apps" roundups appear in 53 of the 198 answers. Review roundups are the pages to be on.

5. Named is not cited

The tracked leader was named in 162 of the 198 answers. In 109 of them, its own website was not among the sources. Engines recommend a brand from what others write about it, so "is my brand named?" and "is my site cited?" are two different metrics.

6. ChatGPT searches with the year and the brand names

ChatGPT ran 69 web searches for its 50 answers. 65 of them contained "2026", and many already named brands, for example "best small business CRM software 2026 HubSpot Zoho Pipedrive Freshsales reviews". Pages that carry the current year and a clear comparison have an edge.

What I would do with this

• Track each engine separately. A brand can come first on Perplexity and be missing from Google's overviews.
• Work on the pages each engine reads: YouTube and Reddit for Google, review roundups for Perplexity, your own comparison pages for ChatGPT.
• Pick the questions you can win: "free", "cheap", "for teams", "for ecommerce".
• Re-run the same questions every week. Answers change from one run to the next, so a single snapshot is noise.

Limits

One run per question, from the United States, on one day. AI answers vary between runs, so treat small differences with care. "First" means the first tracked brand that appears in the answer text, and "named" means the brand name appears as a whole word.

Run it on your own brand

{
  "brand": "HubSpot",
  "brandDomains": ["hubspot.com"],
  "competitors": ["Salesforce | salesforce.com", "Pipedrive | pipedrive.com", "Zoho | zoho.com"],
  "prompts": ["best CRM for small business", "best free CRM software", "which CRM is easiest to use"],
  "engines": ["google_ai_overview", "chatgpt", "perplexity", "gemini"],
  "country": "US"
}

A check is one question on one engine and costs $0.01; a Perplexity answer counts as 2. This whole study was 250 checks, about $2.60. From an AI agent, connect https://mcp.apify.com/?tools=gaelasao/ai-visibility-tracker and ask the same thing in plain words.

Which category should I test next? Tell me in the comments.