The AI Recommendation Index

Which brands ChatGPT, Gemini, Claude, and Perplexity recommend across 25 product and software categories. Q4 2026 edition.

By Jarred Smith · Data collected September 29–30, 2026 · Published October 2026 · Next update January 2027

I asked four AI engines 1,600 buying questions across 25 categories, from CRM and payroll software to mattresses and running shoes, with web search on and the wording held identical. I got 1,503 usable answers back, and in them the engines named 1,039 different brands. This page ranks every category by how often each brand shows up, shows where each engine gets its information, and measures how often the engines agree with each other (not very often, as it turns out).

The short version: a handful of brands in each category show up in nearly every answer, the four engines lead with the same brand only 26% of the time, and ChatGPT pulls 65% of its citations from the websites of the brands it recommends while the other three engines lean on reviews and media coverage.

Key findings from the Q4 2026 index

1. ChatGPT cites the brands' own websites far more than the other engines do. When ChatGPT named a brand, 65% of its citations pointed to a website belonging to a brand in that same answer. The share was 12% for Gemini, 10% for Perplexity, and 6% for Claude. The result held in both independent runs of the study (67% and 64%).
2. The four engines named the same brand first in 26% of cases. Across 299 question runs answered by all four engines, they led with the same brand 77 times. In 21 cases, each engine led with a different brand.
3. Asking the same engine again often changes the answer. The same engine gave the same first brand to the same question 70% of the time. Across all four times a question was asked, an engine stuck with one first brand in 52% of cases.
4. Most categories have a clear leader and a long tail. In 19 of 25 categories, the leading brand appeared in at least 90% of answers. On average only four or five brands per category appeared in half the answers or more, while 45% of the brands named in a category showed up only once.
5. The engines agree on the category leader a little over half the time. All four engines had the same most-mentioned brand in 14 of 25 categories. In CRM, ChatGPT mentioned Salesforce most while the other three favored HubSpot. In email marketing, ChatGPT and Perplexity favored Klaviyo and Claude and Gemini favored Mailchimp.
6. The buyer changes the answer. In a separate test with six buyer profiles, the six buyers were not all led to the same brand in 39 of 40 question and engine combinations. Gusto showed up in 82% of payroll answers for a US founder and in none for buyers in London or Toronto.
7. The wording of the question changes how many brands you see. "What are the best X options?" produced about 10 brands per answer. A use-case question produced about 6, and a question asking the engine to compare the top two produced about 4.

AI brand rankings by category

Pick a category to see the ten brands the engines named most often, then switch between engines to see how each one differs. The percentage is the share of answers in that category that named the brand at least once. You can search for a brand, link straight to a category, or grab a badge if your brand made a top five.

Where AI engines get their information

The engines cited 18,742 web pages across the study, and they don't draw from the same places. Perplexity cited about 20 sources per answer and Gemini about 13, while ChatGPT and Claude cited about 8 each. The biggest difference is in whose pages get cited. ChatGPT leaned heavily on the brands' own websites (product pages, pricing pages, help centers), and Claude, Gemini, and Perplexity leaned on third-party reviews and coverage.

The most-cited third-party sources across all four engines were G2 (585 citations, counting learn.g2.com), Forbes (512), TechRadar (335), Zapier (319), PCMag (302), and RTINGS (266).

For a brand, this splits the work in two. Your own site carries a lot of weight with ChatGPT, so pricing, comparison, and use-case pages need to say plainly who the product is for. For the other three engines, the reviews and roundups that mention you do more of the lifting.

How often AI engines agree on which brand to recommend

Every question in the index was asked four times per engine, in two separate runs launched two minutes apart. That makes it possible to measure agreement two ways, between engines and within the same engine.

Claude agreed most often with Gemini (59% on the first brand named) and Perplexity (58%). ChatGPT agreed least with the others, and least of all with Perplexity (41%). Some brands depend heavily on which engine the buyer happens to use. Blue Apron appeared in 93% of Gemini and Perplexity meal kit answers and in none from Claude. Mullvad appeared in 93% of ChatGPT's VPN answers and 13% of Claude's. Red Canary appeared in 94% of ChatGPT's MDR answers and 19% of Perplexity's.

How the buyer changes the AI recommendation

A second study asked the same 10 software questions on behalf of six buyers: a founder of a 10-person SaaS company in Austin, an office manager at a 25-person construction company in Tulsa, a marketing director at a 200-person manufacturer in Chicago, a head of growth at a 60-person ecommerce company in London, an IT manager at a 400-person financial services firm in Toronto, and a VP of operations at a 2,000-person healthcare company in Boston. That produced 694 usable answers.

Brand lists for two different buyers overlapped by 27%. For the same buyer asking twice, the overlap was 48%. The engines were reading the buyer's size, budget, and location, and some of the swings made sense. Gusto only serves US employers, and it didn't come up for the London or Toronto buyers at all. Zendesk showed up in 92% of help desk answers for the healthcare VP and 8% for the construction office manager. Sage Intacct appeared in every accounting answer for the VP and none for the office manager, who got Zoho Books and Wave instead.

What this means for brands

Four things stand out to me from this data, and they line up with what I see when I run audits.

First, one check in one engine tells you very little. With first-brand agreement at 26% across engines and 70% within the same engine, a brand needs repeated checks across all four engines before the result means anything. That's the same pattern I found in the AI Verdict Study, where 29% of brand and engine pairings flipped between runs.

Second, the leaders in these categories show up almost everywhere, and the gap after the top four or five is steep. For a challenger, getting into the "short list" answers matters more than winning the "best overall" slot.

Third, your own website and third-party coverage do different jobs depending on the engine. ChatGPT reads your pages, and the other three read what others say about you. A brand that has only worked on one of those will look very different from one engine to the next.

Fourth, the buyer's details decide a lot. A brand that fits small US businesses will not show up for a mid-size company in Toronto, and that's the engine doing its job. Tracking visibility with one generic prompt hides that.

If you want to see where your own brand stands, the free AI Visibility Quick Check takes about 20 minutes, and the AI Visibility Audit runs this kind of testing on your category for you. I wrote more about why this happens in Explainable.

Frequently asked questions

Which AI engine recommends the most brands?

ChatGPT named the most brands per answer in the Q4 2026 index, about 8.6 on average. Gemini named 7.5, Claude 5.3, and Perplexity 5.2.

Do ChatGPT, Gemini, Claude, and Perplexity recommend the same brands?

They often don't. The four engines led with the same brand in 26% of the 299 question runs all four answered, and they shared the same most-mentioned brand in 14 of 25 categories.

Where does ChatGPT get its brand recommendations?

With web search on, 65% of ChatGPT's citations in this index pointed to websites belonging to brands named in the same answer. Claude, Gemini, and Perplexity cited brand websites for 6% to 12% of their citations and relied more on review sites and publishers such as G2, Forbes, TechRadar, and PCMag.

Does AI give the same answer if you ask twice?

Not reliably. The same engine named the same brand first 70% of the time when the identical question was repeated, and it held one first brand across four repeats 52% of the time.

Does the buyer's company size or location change AI recommendations?

Yes. When the same question came from six different buyer profiles, the brands recommended overlapped by 27%, compared with 48% for the same buyer asking twice. Location mattered most for products sold in one country, such as Gusto payroll.

How is the AI Recommendation Index measured?

Each category was asked four ways (best options, a specific use case, a problem-first question, and a top-two comparison) on ChatGPT, Claude, Gemini, and Perplexity with web search on, four times each. A brand's mention rate is the share of usable answers in the category that named it at least once. Full method and limits are below.

How often is the index updated?

Quarterly. The next edition is scheduled for January 2027, and each edition will show which brands moved.

Method and limits

Questions. 25 categories (16 software and service categories, 9 consumer product categories), each asked four ways: "What are the best [category] options right now?", a specific use case, a problem-first question, and a request to compare the top two brands. No brand names appear in the questions.

Engines and settings. ChatGPT (gpt-5-2025-08-07 with web search), Claude (claude-haiku-4-5-20251001 with web search), Gemini (gemini-2.5-flash with Google Search grounding), and Perplexity (sonar-pro). All calls went through the API with no system prompt, no buyer profile, and no memory. API models can differ from the consumer apps people use every day.

Runs. Two identical runs of 800 calls each were launched on September 29, 2026, two minutes apart, and completed on September 30. Each question went to each engine twice per run. Of 1,600 calls, 1,503 produced usable answers (778 and 725 per run). Failures were collection errors spread evenly across engines.

Measurement. Brands were extracted from each answer by a separate model, then merged across spelling variants and product lines (for example, "monday.com" and "Monday.com", or "Tesla Model Y" counted as Tesla). Publications and programs that appeared as brands, such as NerdWallet, Edmunds, Sleep Foundation, and Priority Pass, were removed. A citation counts as a brand's own site when its domain belongs to a brand named in that answer. "Named first" is the first brand mentioned in the answer.

Buyer study. Ten B2B software questions asked on behalf of six buyer profiles, on all four engines, three times each (720 calls, 694 usable), run September 29–30, 2026. Each buyer and category combination rests on about 12 answers, so treat individual percentages in that section as directional.

Limits. This is a snapshot of two days. Results come from API calls and don't account for personalization, chat history, or location settings in consumer apps. Some Gemini citations couldn't be resolved to a source and are left out of the source counts. Mention rate measures presence in answers, not sentiment or purchase outcomes.

Download the data and cite the index

The full rankings (660 brand and category rows with per-engine mention rates) are free to download and use with attribution.

Jarred Smith is the author of Explainable: Why AI Recommends Some Brands & Ignores Others, an Amazon bestseller on AEO, GEO, and SEO. He's a marketing leader with nearly 20 years of experience across healthcare, public media, retail, and environmental services, and he runs AI visibility audits and workshops for marketing teams.