Research

We asked ChatGPT, Gemini and Google AI 210 buying questions. They picked the same #1 brand 27% of the time.

839 answers, 21 categories, 4 engines. Method, brand lists and raw data are linked at the bottom.

Preliminary: this is day 1 of a 7-day run. We'll update every number with the full week on 24 November and log the changes at the bottom.

Same #1 brand on all four engines
27%54 of 199 questions
Recommendations only one engine made
16%623 of 3,845
Categories where an engine has a different favourite
16 of 21the leader isn't #1 everywhere
Answers recommending Supercite in our own category
0 of 39our starting line

People now ask AI what to buy. "Best CRM for a small team." "Best travel insurance with medical coverage." "Best running shoes for wide feet." The answer names four or five brands, and most people never look further.

So we asked the four AI engines that answer the most of these questions the same 210 buying questions, on the same day, from the US. Then we counted who got recommended.

TL;DR

  • The engines rarely agree on the top pick. All four engines named a brand for 199 of the questions. They put the same brand first on 54 of them (27%).
  • They do agree on who's in the conversation. In every category, one or two brands show up in most answers: HubSpot in 93% of CRM answers, Proton VPN in 98% of VPN answers, QuickBooks in 95% of accounting answers.
  • But the leader isn't the leader everywhere. In 16 of 21 categories, at least one engine puts a different brand on top. ChatGPT's favourite small-business credit card is American Express. Google's AI Overviews and AI Mode prefer Chase.
  • Google's two AI products disagree with each other about as much as they disagree with ChatGPT. AI Overviews and AI Mode share 53% of their recommendations for the same question. ChatGPT and Gemini share 58%.
  • 16% of all recommendations came from only one engine. For that same question, the other three engines didn't name that brand. If you check one engine, you miss part of the picture.
  • One review site often decides a category. NerdWallet was cited in 83% of the small-business credit card answers. RunRepeat was cited in 79% of the running shoe answers, and SafeHome.org in 73% of the home security answers.
  • We aren't on our own list. Supercite tracks AI visibility, so we tested our own category too. We were recommended in 0 of 39 answers. More on that below.

Explore the data

Who AI recommends, by category

  1. 1HubSpot93%
  2. 2Pipedrive68%
  3. 3Zoho65%
  4. 4monday.com25%
  5. 5Streak23%
  6. 6EngageBay18%
  7. 7Freshsales18%
  8. 8Salesforce18%
  9. 9Attio15%
  10. 10Less Annoying CRM15%

Share of the 40 answers that recommend each brand, with the four engines weighted equally.

What we did, in one paragraph

We wrote 10 buying questions for each of 21 categories: 8 software categories, 7 consumer products and 6 services. None of the questions names a brand. We asked all 210 of them on ChatGPT, Gemini, Google AI Overviews and Google AI Mode on 17 November 2026, as a US user in English. That gave 839 answers: one AI Overview came back empty. For each answer, we counted the brands it recommends. That means a brand put forward in a heading, in bold, as a link, in a table row or at the start of a list item. A brand mentioned only in passing doesn't count: in "Cloudways runs on Google Cloud", Cloudways counts and Google Cloud doesn't. The full method and its limits are at the bottom.

1. Four engines, four different top picks

Take one question: "best protein powder for women."

EngineRecommended, in order
ChatGPTPromix, Ascent, Ritual, Orgain, Transparent Labs
GeminiTransparent Labs, Truvani, Ritual, Nutricost
Google AI OverviewsLegion, Nutricost, Transparent Labs, Ritual, Naked Nutrition
Google AI ModeRitual, Transparent Labs, Orgain, Optimum Nutrition

There are four different #1s. Only two brands, Ritual and Transparent Labs, made all four lists.

That isn't one odd question. Across the study, the four engines put the same brand first on 54 of 199 questions. It varies a lot by category:

CategoryMost recommended2nd3rdSame #1 on all four engines
AI search visibility toolsSemrush (66%)Otterly.AI (53%)Profound (51%)0 of 7
CRMHubSpot (93%)Pipedrive (68%)Zoho (65%)4 of 10
Email marketingMailerLite (73%)Brevo (63%)Mailchimp (43%)2 of 10
Project managementClickUp (63%)Asana (55%)Trello (55%)3 of 9
Accounting softwareQuickBooks (95%)Zoho Books (80%)Xero (70%)3 of 10
Website buildersSquarespace (85%)Wix (85%)WordPress (43%)4 of 10
AI meeting notetakersFireflies.ai (85%)Otter.ai (68%)Fathom (68%)3 of 10
Help desk softwareHelp Scout (68%)Freshdesk (65%)Zendesk (58%)2 of 10
Running shoesBrooks (83%)ASICS (78%)HOKA (63%)0 of 8
MattressesHelix (70%)Nectar (58%)Brooklyn Bedding (50%)2 of 10
Protein powderTransparent Labs (80%)Dymatize (53%)Optimum Nutrition (48%)2 of 10
VPNsProton VPN (98%)Surfshark (80%)NordVPN (75%)4 of 10
MoisturizersCeraVe (78%)La Roche-Posay (65%)Neutrogena (48%)1 of 6
Espresso machinesBreville (73%)De'Longhi (70%)Gaggia (25%)7 of 10
Dog foodPurina (75%)Hill's (75%)Royal Canin (60%)2 of 9
Online therapyTalkspace (95%)BetterHelp (68%)Grow Therapy (65%)1 of 10
Meal kitsHome Chef (58%)Green Chef (55%)HelloFresh (50%)6 of 10
Web hostingHostinger (70%)SiteGround (65%)Cloudways (60%)1 of 10
Travel insuranceAllianz (73%)Travelex (53%)Seven Corners (50%)2 of 10
Small business credit cardsChase Ink (75%)American Express (73%)Capital One Spark (68%)2 of 10
Home security systemsRing (85%)SimpliSafe (83%)ADT (50%)3 of 10

Percentages are the share of answers that recommend the brand. We average the four engines with equal weight, so each engine counts the same. "Same #1" only counts questions where all four engines recommended at least one brand.

Espresso machines and meal kits are the most settled: the four engines put the same brand first on 7 and 6 of 10 questions. Running shoes and our own category are the least settled: not one question got the same #1 from all four.

2. Everyone agrees on the shortlist, not on the order

The top of each category is crowded but stable. Across the 21 categories, the three most-recommended brands take between 32% and 60% of all recommendations; the median is 46%. Most answers name four or five brands:

ChatGPTGeminiAI OverviewsAI Mode
Brands recommended per answer4.94.64.44.6

So if you're a top-three brand in your category, you're probably in the answer. Whether you're the first name, the one the answer calls "my recommendation" or "best overall", depends on which engine the buyer happens to use.

The category leader came first on every engine in only 5 of 21 categories: CRM (HubSpot), running shoes (Brooks), protein powder (Transparent Labs), VPNs (Proton VPN) and online therapy (Talkspace). In the other 16, at least one engine has a different favourite:

  • Small-business credit cards: ChatGPT recommended American Express in 9 of 10 answers and Chase Ink in 7. AI Overviews and AI Mode both lean to Chase Ink.
  • Travel insurance: ChatGPT recommended Seven Corners in 8 of 10 answers. AI Overviews recommended Allianz in 9 of 10.
  • Home security: ChatGPT and AI Mode recommend SimpliSafe most. Gemini recommended Ring in all 10 answers.
  • Accounting: ChatGPT put Zoho Books in all 10 answers, one more than QuickBooks. The other three engines favour QuickBooks.

Some of these gaps are one or two answers wide (see the limitations). But they are consistent with the "same #1" numbers above: the engines read the same market differently.

3. Google's two AI products don't agree with each other either

You might expect Google AI Overviews and Google AI Mode to give the same answer. Both are Google and both search Google. They don't. Here is how much of each pair's recommendations overlap for the same question:

Engine pairOverlap
ChatGPT and Gemini58%
ChatGPT and AI Mode55%
AI Overviews and AI Mode53%
Gemini and AI Mode52%
ChatGPT and AI Overviews51%
Gemini and AI Overviews49%

Overlap is the Jaccard index of the two recommendation lists: the brands both named, divided by the brands either named. It's averaged over questions.

The most similar pair is ChatGPT and Gemini, from two different companies. The least similar is Gemini and AI Overviews, both from Google. Ranking well in one Google product tells you little about the other.

4. One in six recommendations comes from a single engine

We counted every brand an engine recommended, question by question, and checked whether any of the other three engines named it for the same question. 623 of 3,845 recommendations (16%) were unique to one engine.

That's the case for tracking more than one engine. A brand can appear in ChatGPT and nowhere else for the same question, or the other way round. Some brands are only ever named by one engine: 108 of the 483 brands that got recommended at all were recommended by one engine only.

5. The sites behind the answers

AI engines cite their sources, so we can see what they read. In many categories, one site does most of the work:

CategoryMost-cited siteShare of answers citing it
Small business credit cardsnerdwallet.com83%
Running shoesrunrepeat.com79%
Home security systemssafehome.org73%
Online therapyhelpguide.org70%
Travel insurancesquaremouth.com68%
Mattressessleepfoundation.org59%
CRMreddit.com46%
Web hostingreddit.com45%
Meal kitsbonappetit.com45%
AI meeting notetakersjusttalkingtech.medium.com43%

The full table for all 21 categories is in the data download.

A few things stand out:

  • A single blog can decide a category. One Medium blog was cited in 43% of the AI notetaker answers. It was in 7 of 10 Gemini answers and 7 of 10 AI Mode answers.
  • Google's engines lean on Reddit and YouTube. ChatGPT, in our capture, didn't. YouTube was cited in 59% of AI Overviews answers, and Reddit in 38% of AI Overviews answers and 35% of AI Mode answers. ChatGPT cited neither in any of its 210 answers. It cited review and publisher sites (NerdWallet, Tom's Guide, Healthline, TechRadar) and the brands' own sites.
  • ChatGPT links to the brands it recommends. Google mostly doesn't. When ChatGPT recommended a brand, it linked that brand's own website 57% of the time. Gemini did it 5% of the time, AI Overviews 13% and AI Mode 11%. On Google's engines, a recommendation is mostly built from what other sites say about you.
ChatGPTGeminiAI OverviewsAI Mode
Sites cited per answer7.55.65.93.7
Answers citing reddit.com0%28%38%35%
Answers citing youtube.com0%12%59%24%
Answers linking the recommended brand's own site57%5%13%11%
Average answer length (characters)2,9623,6601,9673,008

6. Our own category: we're not on the list

Supercite tracks how AI recommends brands, so "AI search visibility tools" is our category. We asked questions like "best tools to track brand visibility in ChatGPT" and "affordable AI visibility tracking tool for small businesses."

Supercite was recommended in 0 of the 39 answers. Here's who was:

BrandAll enginesChatGPTGeminiAI OverviewsAI Mode
Semrush66%70%80%56%60%
Otterly.AI53%80%60%33%40%
Profound51%60%50%44%50%
Peec AI46%70%40%33%40%
Ahrefs41%20%60%33%50%
SE Ranking28%20%40%22%30%
Scrunch25%70%0%11%20%
AthenaHQ20%40%10%11%20%

We're a new product, so this is what we expected. It's also the most useful row in the study for us, because it's the starting line. Scrunch is a good example of section 4: ChatGPT recommended it in 7 of 10 answers and Gemini in none.

We'll track this category every day and publish what we do and what changes, including if nothing does.

What this means if you sell something

  1. Check every engine, not one. One in six recommendations comes from a single engine. Each engine's top pick differs from the category leader in 16 of 21 categories. One engine is a sample, not the market.
  2. Find the site your category leans on. If one review site is cited in 70–80% of answers, getting onto that site's list may matter more than anything on your own website. Look at the sources the answers cite for your questions; they're usually right there.
  3. Google and ChatGPT need different work. ChatGPT links brands' own sites and review sites. Google's AI leans on Reddit, YouTube and third-party reviews. A plan that only covers your own site covers about half of it.
  4. Being in the answer and being first are different goals. If you're already in the top three, the work is about being the first name, and that changes by engine.

We wrote up what the research says about how to do each of these: How to get ChatGPT to recommend your business.

Download the data

Licensed CC BY 4.0: free to use, including commercially, as long as you link back to this page.

Method and limitations

Questions. 10 per category, 210 in total. They are written the way people ask: "best X for Y", "X that does Z", "is it worth paying for X". None of them names a brand. The full list is in questions.csv.

Engines. ChatGPT, Gemini, Google AI Overviews and Google AI Mode. Each was asked every question once on 17 November 2026, for a US location in English. The answers were collected through the same pipeline Supercite uses for customers: DataForSEO, with Bright Data as fallback. One AI Overview came back empty, so there are 839 answers, not 840.

Counting brands. We counted brands, not products: "Gel-Kayano 31" counts as ASICS. We used a hand-built list of 700 brands with their product names (brands.json). A brand counts once per answer, in its own category only. It counts as recommended when the answer puts it forward: in a heading, in bold, as a link, in the first column of a table, or at the start of a list item (including "Best value: Brand X"). A brand mentioned only in passing ("built on Google Cloud") counts as named, not recommended. 5% of named brands were passing mentions. The download has both. Not counted as brands: AI engines, platforms (Reddit, YouTube, Zoom, Shopify inside a help desk answer), retailers, publishers, certifications, protocols and insurers that come up inside another category. Image captions are left out, because in ChatGPT they are the titles of other sites' pages.

Agreement. "Same #1" compares the first recommended brand from each engine for the same question. It only counts questions where all four engines recommended at least one brand: 199 of 210. Overlap is the Jaccard index of two engines' recommendation lists for the same question.

Citations. A site counts once per answer. The "brand's own site" check matches the cited domain to the brand's name or alias, so it misses brands whose domain doesn't match their name.

Limitations. Please read these before quoting a number.

  • This is one day. AI answers change from day to day. Each engine's share in a category is out of 10 answers, so a 10-point difference between two engines is one answer. Treat category-level engine differences as directional until the full week is in.
  • One question, one answer. We asked each question once per engine per day. Repeating a question gives somewhat different answers, which is why we're running it for a week.
  • US and English only. Other countries and languages give different answers.
  • Recommended isn't the same as endorsed. Our rule can't tell "best overall: X" from a bolded "avoid X". We searched every bolded phrase for words like "avoid", "skip" and "worst". There were 29 hits, and only one was about a brand ("Avoid GnuCash").
  • Our list can miss brands. We checked every name that came up three or more times against the list. A brand named once or twice may be missing.
  • Answers aren't traffic. We measured what the engines say, not how many people ask or what they click.

Data. Everything is in the downloads above: the per-brand shares, every recommendation, every citation, the questions, the brand list and the script.

FAQ

Which AI engine should I care about most?

The one your buyers use, and you probably don't know which one that is. The engines disagree enough that a brand can lead on one and be missing from another. Start by checking all four for the questions your buyers ask.

Why don't the engines agree?

They search different indexes (ChatGPT uses Bing and its own crawl, Google's engines use Google), they read different sources (Google's lean on Reddit and YouTube), and each one writes its own summary of what it found. Small differences in sources become different top picks.

Does being #1 on Google mean being #1 in AI answers?

Not reliably. Google's two AI products disagree with each other on about half of their recommendations, and both cite third-party sites more than brand sites.

Is a brand's share stable?

Not completely. Answers change from day to day, which is why this study runs for a week and why we report shares rather than single answers.

Can I use the data?

Yes. Use it however you want and link back to this page.

Is Supercite in the results?

No: 0 of 39 answers in our own category. We're publishing that because it's true and because it's our baseline.

Update log

  • 17 November 2026. First version, using day-1 data (839 answers).
  • 24 November 2026 (planned). Full week, about 5,900 answers. Every number gets updated, plus how stable each engine's picks are from day to day.

Supercite asks these questions about your business every day on ChatGPT, Gemini, AI Overviews and AI Mode. It shows who gets recommended instead of you, which sites the answers lean on, and what to fix. Start a 3-day trial.

All posts →

Make AI search your next revenue channel

Start free trial