When a B2B founder asks an AI assistant who to hire or what to buy, the answer is a short list built from a few company websites, a handful of LinkedIn pages and other agencies' blog posts. On 13 November 2026 I ran 50 founder questions through ChatGPT, Google AI Mode and Perplexity and collected 110 answers. They named 393 different agencies, tools and people. Windmill Growth and Understory Agency led LinkedIn ghostwriting, Taplio led LinkedIn tools, Omniscient Digital and iPullRank led GEO, and no single name appeared in more than 18 of the 110 answers.
This is an original study, and it is mine. I run Wise Mango (a LinkedIn branding service) and Klype (a marketing intelligence product), and both sit in the categories I tested. I checked for them in every answer and report what I found plainly, including where the result is probably inflated by my own accounts. The full list of questions was written before the run and is published in the dataset, including every question where my brands did not appear.
The raw data is free to reuse: get ai-answers-b2b-2026-11.csv by email. It has one row per name per answer: the date, the engine, the question and its group, the position of each name, every cited domain, and notes on personalisation and my own brands. The question list, ai-answers-b2b-2026-11-questions.json, comes with it.
Get the data: ai-answers-b2b-2026-11.csv
766 rows, one per name per answer, from ChatGPT, Google AI Mode and Perplexity on 13 November 2026, plus the 50-question list as JSON. Free to reuse with a link to this page. I email you the download link.
Or
See where your company stands in these answers
A free check of whether ChatGPT, Perplexity and Google AI Mode name your company when buyers ask about your category.
How to cite this study. Kaushik, S. (2026, November 13). Which agencies and tools does ChatGPT recommend? 50 B2B buyer questions in ChatGPT, Google AI Mode and Perplexity. shrutikaushik.com. https://shrutikaushik.com/writing/which-agencies-tools-chatgpt-recommends-b2b-2026
Short form for articles: "In a November 2026 study of 110 AI answers to B2B buyer questions, Shruti Kaushik found that ChatGPT and Google AI Mode shared an average of fewer than three named firms per question (shrutikaushik.com)." Please link to this page, not to the data file. The data is free to reuse with attribution.
#How the study worked
- Engines: ChatGPT (web search on, in a Temporary Chat), Google AI Mode (the
udm=50search mode) and Perplexity (default mode). One fresh query per question per engine, no follow-up prompts, nothing typed except the question. - Date: every answer was collected on 13 November 2026, in a Chrome browser in India.
- Questions: 50, in five groups of 10, written the way founders type them: LinkedIn ghostwriters and agencies; founder-led marketing and personal branding; getting named by ChatGPT and AI search (GEO and AEO); go-to-market planning and GTM tools; and LinkedIn content tools and Taplio-type alternatives.
- Coverage: Google AI Mode 50 of 50, ChatGPT 50 of 50, Perplexity 10 of 50. Perplexity showed "You've reached your free search limit" at question 11 and still showed it when checked again at the end of the run, so it was stopped. Perplexity results cover the ghostwriting group only and are reported separately where that matters.
- What was recorded: every agency, tool or person the answer highlighted in bold, as a heading or as the first cell of a table row, in the order shown; and every external domain the answer linked to.
- Personalisation: all three engines were used in my own signed-in browser. ChatGPT addressed me as the owner of my companies in 8 of its 50 answers, so ChatGPT figures for my own brands should be read as personalised. Details are in the Methodology section.
#The headline numbers
| Measure | Google AI Mode | ChatGPT | Perplexity | All |
|---|---|---|---|---|
| Answers collected | 50 | 50 | 10 | 110 |
| Answers that named at least one firm | 41 (82%) | 47 (94%) | 8 | 96 (87%) |
| Name mentions | 306 | 349 | 57 | 712 |
| Different names | 220 | 226 | 48 | 393 |
| Average names per answer (when any) | 7.5 | 7.4 | 7.1 | 7.4 |
| Domain citations (one per domain per answer) | 399 | 271 | 62 | 732 |
| Different domains cited | 251 | 179 | 30 | 392 |
| Average domains per answer | 8.0 | 5.4 | 6.2 | 6.7 |
| Answers with no linked source | 3 | 5 | 0 | 8 |
Four patterns stand out before looking at any single name.
- ChatGPT almost always gives a shortlist. 47 of its 50 answers named firms, against 41 of 50 in Google AI Mode. Google was more likely to answer "who can help me" questions with categories ("fractional CMOs", "GEO and AI SEO agencies") and no names. It did that for 9 questions, including "who can help my brand show up in ai search results" and "who can help me build a go-to-market plan for my startup".
- Google AI Mode links to more sources. It cited 8.0 domains per answer against 5.4 for ChatGPT. ChatGPT mostly linked the firms it named. Google linked a wider mix of other companies' blog posts, LinkedIn, YouTube and Reddit.
- The long tail is enormous. 271 of the 393 names (69%) appeared in only one answer. 279 of the 392 cited domains (71%) appeared in only one answer.
- Shortlists are about seven names long in every engine, so a firm is competing for one of roughly seven slots per question.
#The most-named agencies, tools and people
The table counts the number of answers that named each brand at least once, out of 110. It shows the 25 most-named brands, extended to 27 because the last places are tied at 5.
| Rank | Name | Answers | Google AI Mode | ChatGPT | Perplexity | Main question group |
|---|---|---|---|---|---|---|
| 1 | Understory Agency | 18 | 8 | 8 | 2 | Ghostwriting (11) |
| 2 | Windmill Growth | 17 | 6 | 6 | 5 | Ghostwriting (15) |
| 3 | Taplio | 14 | 6 | 8 | 0 | LinkedIn tools (13) |
| 4 | Supergrow | 11 | 5 | 6 | 0 | LinkedIn tools (10) |
| 4 | AuthoredUp | 11 | 6 | 5 | 0 | LinkedIn tools (11) |
| 6 | Rethoric | 10 | 6 | 2 | 2 | Ghostwriting (7) |
| 7 | Omniscient Digital | 8 | 3 | 5 | 0 | GEO and AEO (8) |
| 7 | Postiv AI | 8 | 3 | 5 | 0 | LinkedIn tools (8) |
| 9 | Wise Mango (author's business) | 7 | 1 | 6 | 0 | Ghostwriting (5) |
| 9 | Boldface | 7 | 2 | 5 | 0 | Founder branding (5) |
| 9 | iPullRank | 7 | 3 | 4 | 0 | GEO and AEO (7) |
| 12 | Stop The Scroll | 6 | 3 | 2 | 1 | Ghostwriting (5) |
| 12 | HubSpot | 6 | 3 | 3 | 0 | GTM (6) |
| 14 | Storyarb | 5 | 0 | 4 | 1 | Ghostwriting (4) |
| 14 | Column Content | 5 | 5 | 0 | 0 | Ghostwriting (4) |
| 14 | YellowInk Digital | 5 | 0 | 2 | 3 | Ghostwriting (5) |
| 14 | The Wise Idiot | 5 | 3 | 1 | 1 | Ghostwriting (4) |
| 14 | GrowthSpree | 5 | 2 | 3 | 0 | GTM (4) |
| 14 | The People Company | 5 | 2 | 3 | 0 | Founder branding (5) |
| 14 | Kalungi | 5 | 2 | 3 | 0 | GTM (4) |
| 14 | Winning by Design | 5 | 2 | 3 | 0 | GTM (5) |
| 14 | Apollo.io | 5 | 3 | 2 | 0 | GTM (5) |
| 14 | Clay | 5 | 4 | 1 | 0 | GTM (5) |
| 14 | ContentIn | 5 | 2 | 3 | 0 | LinkedIn tools (5) |
| 14 | Kleo | 5 | 2 | 3 | 0 | LinkedIn tools (5) |
| 14 | MagicPost | 5 | 3 | 2 | 0 | LinkedIn tools (5) |
| 14 | Metricool | 5 | 2 | 3 | 0 | LinkedIn tools (5) |
Wise Mango is in this table, and it should be read with the caveat in the section on my own brands: 6 of its 7 mentions came from ChatGPT, on an account whose answers addressed me as the owner of Wise Mango and Klype. In Google AI Mode it was named once in 50 answers, and in Perplexity it was named in none of the 10.
The brands that led the whole study were the ones named by more than one engine. Understory Agency and Windmill Growth were named by all three engines, and together they account for 35 of the 712 mentions. Column Content shows the opposite pattern: all 5 of its mentions came from Google AI Mode, and ChatGPT never named it.
The first name in an answer matters because it is the one a reader sees first. Windmill Growth was named first in 7 answers, Supergrow in 5, Rethoric in 4, and Understory Agency, Omniscient Digital, Kalungi and Taplio in 3 each.
#Who each kind of question names
The five question groups produced almost separate markets. Only a few names crossed groups: Understory Agency appeared in ghostwriting, founder branding and GTM answers, and Taplio and Supergrow each made one appearance in a ghostwriting answer.
LinkedIn ghostwriters and agencies
This is the only group run in all three engines, so it has 30 answers (10 per engine). It is also the most concentrated group in the study.
| Name | Answers (of 30) | Google AI Mode | ChatGPT | Perplexity |
|---|---|---|---|---|
| Windmill Growth | 15 | 5 | 5 | 5 |
| Understory Agency | 11 | 4 | 5 | 2 |
| Rethoric | 7 | 4 | 1 | 2 |
| Stop The Scroll | 5 | 2 | 2 | 1 |
| Wise Mango (author's business) | 5 | 1 | 4 | 0 |
| YellowInk Digital | 5 | 0 | 2 | 3 |
| Storyarb | 4 | 0 | 3 | 1 |
| RevenueZen | 4 | 0 | 4 | 0 |
| Postlit Media | 4 | 0 | 3 | 1 |
| Column Content | 4 | 4 | 0 | 0 |
Windmill Growth was named in half of all ghostwriting answers and was the only firm that every engine named in 5 of its 10 answers. The rest of the list depends heavily on the engine: Column Content was a Google name, RevenueZen a ChatGPT name, YellowInk Digital mostly a Perplexity name.
Narrow questions surfaced individual people rather than agencies. "linkedin ghostwriter for fintech founders" in Perplexity named eight individual writers and cited only linkedin.com. "linkedin ghostwriter for enterprise executives" in Perplexity did the same. Questions about price ("affordable linkedin ghostwriting service for an early stage startup") brought in marketplaces and lower-cost providers such as Jennavi, Upwork and Fiverr.
"linkedin ghostwriting agency or freelancer, which is better for a founder" produced no named firms in any of the three engines, and "who should i hire to write my linkedin posts as a startup founder" produced none in Google AI Mode or Perplexity. Comparison and "how do I choose" questions get advice; "best X for Y" questions get names.
Founder-led marketing and personal branding
This was the most fragmented group: 120 different names in 20 answers (Google AI Mode and ChatGPT), with no name in more than 5.
| Name | Answers (of 20) | Google AI Mode | ChatGPT |
|---|---|---|---|
| Boldface | 5 | 2 | 3 |
| The People Company | 5 | 2 | 3 |
| Understory Agency | 5 | 3 | 2 |
| Ohh My Brand | 4 | 2 | 2 |
| Midas Touch | 4 | 1 | 3 |
| FableSquare | 4 | 2 | 2 |
| Shake Content | 3 | 3 | 0 |
| Rethoric | 3 | 2 | 1 |
| Profile | 3 | 1 | 2 |
| Montaz Medias | 3 | 2 | 1 |
| Scribblers India | 3 | 1 | 2 |
Several of these are India-based agencies. Only one question asked for India ("personal branding agency for founders in india"), but ChatGPT also split its lists into "India" and "international" sections for questions 12, 13 and 14, which did not mention a country. That is one of the signs of location or account personalisation discussed in the Methodology.
The "experts to learn from" question named people, not firms. Google AI Mode named 11 (Chris Walker, Dave Gerhardt, Justin Welsh and Rand Fishkin among them) and ChatGPT named 14, with Dave Gerhardt, Justin Welsh, Adam Robinson and Rand Fishkin appearing in both.
Getting named by ChatGPT and AI search (GEO and AEO)
| Name | Answers (of 20) | Google AI Mode | ChatGPT |
|---|---|---|---|
| Omniscient Digital | 8 | 3 | 5 |
| iPullRank | 7 | 3 | 4 |
| DerivateX | 4 | 2 | 2 |
| Nakama | 4 | 1 | 3 |
| Otterly.AI | 4 | 2 | 2 |
| Peec AI | 4 | 2 | 2 |
| Profound | 3 | 1 | 2 |
| Ahrefs Brand Radar | 3 | 1 | 2 |
| Semrush AI Visibility Toolkit | 3 | 1 | 2 |
| TripleDart | 3 | 2 | 1 |
| Skale | 3 | 2 | 1 |
| RevvGrowth | 3 | 1 | 2 |
Agencies and tracking tools split cleanly. Questions about agencies named Omniscient Digital, iPullRank, DerivateX and Nakama; questions about tracking named Otterly.AI, Peec AI, Profound and the AI features of Ahrefs and Semrush. The how-to questions ("how do i get my company recommended by chatgpt", "should a b2b startup hire a geo agency or an seo agency") named no firms in either engine. This group also had the highest share of Reddit and YouTube citations, covered below.
Go-to-market planning and GTM tools
| Name | Answers (of 20) | Google AI Mode | ChatGPT |
|---|---|---|---|
| HubSpot | 6 | 3 | 3 |
| Winning by Design | 5 | 2 | 3 |
| Apollo.io | 5 | 3 | 2 |
| Clay | 5 | 4 | 1 |
| GrowthSpree | 4 | 2 | 2 |
| Kalungi | 4 | 1 | 3 |
| 6sense | 4 | 2 | 2 |
| Refine Labs | 3 | 2 | 1 |
| Attio | 3 | 2 | 1 |
| Demandbase | 3 | 2 | 1 |
| ZoomInfo | 3 | 1 | 2 |
GTM answers mixed consultancies (Winning by Design, Kalungi, GrowthSpree, Refine Labs) with software (HubSpot, Apollo.io, Clay, 6sense). 101 different names appeared in 20 answers. For "best b2b marketing intelligence platform", Google AI Mode named ZoomInfo, Demandbase, 6sense, Contify, Clay, Cognism and ZenABM; ChatGPT named Demandbase, 6sense, ZoomInfo, AlphaSense, Crayon, Similarweb and Klue. The two lists shared three names.
LinkedIn content tools and Taplio-type alternatives
| Name | Answers (of 20) | Google AI Mode | ChatGPT |
|---|---|---|---|
| Taplio | 13 | 5 | 8 |
| AuthoredUp | 11 | 6 | 5 |
| Supergrow | 10 | 4 | 6 |
| Postiv AI | 8 | 3 | 5 |
| ContentIn | 5 | 2 | 3 |
| Kleo | 5 | 2 | 3 |
| MagicPost | 5 | 3 | 2 |
| Metricool | 5 | 2 | 3 |
| Buffer | 4 | 1 | 3 |
| Oiti | 3 | 2 | 1 |
Software questions were the most predictable. Only 54 different names appeared in 20 answers, and the top four tools were each named in at least 8. Even "best taplio alternatives" kept Taplio's neighbours in a tight set: Supergrow, AuthoredUp, ContentIn and Kleo appeared in both engines' answers. For LinkedIn tools, the shortlist is close to settled; for agencies, it is not.
#How concentrated the answers are
Across all 712 mentions, the ten most-named brands hold 15.6% of mentions, and 69% of all names appear in a single answer. On the Herfindahl-Hirschman scale used for market concentration (0 to 10,000), the whole study scores 55, which is extremely fragmented.
| Measure | Google AI Mode | ChatGPT | All engines |
|---|---|---|---|
| Different names | 220 | 226 | 393 |
| Names that appeared in only one answer | 174 (79%) | 160 (71%) | 271 (69%) |
| Share of mentions held by the top 10 names | 17.0% | 16.6% | 15.6% |
| Herfindahl-Hirschman index (0 to 10,000) | 69 | 69 | 55 |
The overall number hides real concentration inside each category:
| Question group | Answers | Different names | Most-named brand | Its share of answers |
|---|---|---|---|---|
| LinkedIn content tools | 20 | 54 | Taplio | 13 of 20 (65%) |
| LinkedIn ghostwriting | 30 | 83 | Windmill Growth | 15 of 30 (50%) |
| GEO and AEO | 20 | 64 | Omniscient Digital | 8 of 20 (40%) |
| GTM planning and tools | 20 | 101 | HubSpot | 6 of 20 (30%) |
| Founder branding | 20 | 120 | Boldface, The People Company, Understory Agency | 5 of 20 each (25%) |
In other words, AI shortlists are concentrated within a category and almost empty of overlap between categories. A software category with a known leader produces the same five tools again and again. A service category such as founder branding produces a different list nearly every time.
#The three engines barely agree
Google AI Mode and ChatGPT both named firms for 41 of the 50 questions. On those 41 questions they shared an average of 2.7 names out of roughly seven each, and they shared no names at all on 7 questions. The average Jaccard similarity of their lists (shared names divided by all names) was 0.24.
Across the whole study, Google AI Mode named 220 different brands and ChatGPT 226. Only 82 brands were named by both; 138 were named only by Google and 144 only by ChatGPT.
Perplexity, on the 10 ghostwriting questions it answered, matched the others no better. All three engines shared at least one name on 6 of those 10 questions, and Windmill Growth was most often that shared name.
For a founder, this means "are we recommended by AI" has no single answer. A firm can be on ChatGPT's list and missing from Google's for the same question on the same day, and the reverse.
#Which sources decide AI visibility
These are the domains the answers linked to most often. Each count is the number of answers that cited the domain at least once, out of 110.
| Rank | Domain | Answers | Google AI Mode | ChatGPT | Perplexity |
|---|---|---|---|---|---|
| 1 | linkedin.com | 36 | 25 | 4 | 7 |
| 2 | youtube.com | 20 | 20 | 0 | 0 |
| 3 | wisemango.io (author's business) | 14 | 4 | 6 | 4 |
| 4 | reddit.com | 12 | 12 | 0 | 0 |
| 5 | understoryagency.com | 10 | 4 | 4 | 2 |
| 5 | taplio.com | 10 | 3 | 7 | 0 |
| 7 | postiv.ai | 9 | 4 | 5 | 0 |
| 8 | hubspot.com | 8 | 5 | 3 | 0 |
| 8 | supergrow.ai | 8 | 3 | 5 | 0 |
| 10 | demandbird.com | 7 | 3 | 2 | 2 |
| 10 | windmillgrowth.com | 7 | 2 | 5 | 0 |
| 10 | magicpost.in | 7 | 2 | 2 | 3 |
| 10 | postlitmedia.com | 7 | 0 | 3 | 4 |
| 10 | getaob.com | 7 | 6 | 1 | 0 |
| 10 | clay.com | 7 | 4 | 3 | 0 |
These are the pages that decide who gets named. Three things are worth noticing.
LinkedIn is the most cited source, but mostly by Google and Perplexity. linkedin.com was linked in 25 of 50 Google AI Mode answers (50%) and 7 of 10 Perplexity answers, but only 4 of 50 ChatGPT answers (8%). For the ghostwriting group alone it was cited in 14 of 30 answers.
YouTube and Reddit appeared only in Google AI Mode. youtube.com was cited in 20 of Google's 50 answers (40%) and reddit.com in 12 (24%). Neither was linked by ChatGPT or Perplexity in this run. The GEO and AEO group drew on them most: youtube.com in 9 of 20 answers and reddit.com in 7.
Some of the most-cited sites were not the firms being recommended. getaob.com was cited in 7 answers and demandbird.com in 7, but neither firm was a regular name in the lists. yoursocialstrategy.co (5), foundera.co (5), viralbrain.ai (4), onfolio.com (4) and executivepresence.io (4) behaved the same way. These are sites that publish comparisons of other providers, and AI engines read them to build their own shortlists. Only 8 domains were cited by all three engines: linkedin.com, wisemango.io, understoryagency.com, demandbird.com, magicpost.in, reachly.co, gtm8020.com and thewiseidiot.in.
The most-cited domains by question group:
| Question group | Most-cited domains (answers) |
|---|---|
| LinkedIn ghostwriting (30 answers) | linkedin.com (14), wisemango.io (11), demandbird.com (7), postlitmedia.com (7), understoryagency.com (6), windmillgrowth.com (6), jennavi.co (6) |
| Founder branding (20) | linkedin.com (8), thepeoplecompany.me (5), fablesquare.com (4), ohhmybrand.com (4), shakecontent.com (3), wisemango.io (3) |
| GEO and AEO (20) | youtube.com (9), reddit.com (7), revvgrowth.com (5), nakama.in (4), beomniscient.com (3), quora.com (3), derivatex.agency (3), ahrefs.com (3) |
| GTM (20) | clay.com (7), hubspot.com (6), winningbydesign.com (4), youtube.com (4), kalungi.com (4), linkedin.com (4), apollo.io (4), demandbase.com (4) |
| LinkedIn content tools (20) | taplio.com (10), postiv.ai (9), supergrow.ai (8), linkedin.com (7), authoredup.com (6), youtube.com (5) |
#What kinds of pages get cited
I sorted all 732 domain citations into types. A citation counts as the brand's own site when the domain belongs to a brand named in that same answer.
| Source type | Citations | Share | Answers citing it |
|---|---|---|---|
| Own site of a brand named in the same answer | 378 | 51.6% | 85 |
| Sites of other firms not named in that answer (agency and vendor blogs) | 241 | 32.9% | 65 |
| 36 | 4.9% | 36 | |
| Big-vendor blogs, publishers and analysts | 23 | 3.1% | 13 |
| YouTube | 20 | 2.7% | 20 |
| 12 | 1.6% | 12 | |
| Directories and marketplaces | 11 | 1.5% | 10 |
| Other social and document sites (Instagram, Quora, Facebook, Medium, Scribd, SlideShare) | 11 | 1.5% | 11 |
| G2 and Clutch | 0 | 0% | 0 |
Vendors' own sites are the biggest single source. Of the 93 answers that both named firms and linked sources, 85 (91%) linked at least one named firm's own website. The engines differ sharply here. In ChatGPT, 86% of linked citations went to the named firms' own sites, because ChatGPT usually links each name in its list. In Google AI Mode the figure was 33%, and 44% of its citations went to other firms' sites. In Perplexity, 60% went to other firms' sites and 19% to the named firms.
Third-party agency and vendor blogs are the second source. 241 citations (33%) went to the sites of companies that were not named in that answer. The pattern above (getaob.com, demandbird.com, yoursocialstrategy.co and others) suggests many of these are "best of" and comparison posts, but this study recorded domains rather than full page addresses, so I cannot say exactly what share were listicles.
Review directories barely matter for these questions. G2 and Clutch were cited 0 times. All directories and marketplaces together made 11 citations: ghostwriterindex.com (4), upwork.com (2), and sortlist.com, growthmentor.com, goodfirms.co, toptal.com and techjockey.com (1 each).
Publishers and analysts are a small slice. The 23 citations in that group were hubspot.com (8), ahrefs.com (3), gartner.com (3), semrush.com (2), zapier.com (2) and five others with one each. No major business publication was cited in this run.
#Where Shruti Kaushik, Wise Mango and Klype appeared
I run Wise Mango and Klype, and this page is on my own site, so this section reports every appearance, including the ones that flatter me less. The engines were used in my own signed-in browser, which matters for how to read these numbers.
| Check | Google AI Mode (50) | ChatGPT (50) | Perplexity (10) |
|---|---|---|---|
| Wise Mango named in the list | 1 (question 8) | 6 (questions 1, 2, 8, 10, 11, 18) | 0 |
| wisemango.io cited | 4 (questions 1, 8, 9, 18) | 6 (the same 6 answers) | 4 (questions 1, 2, 4, 9) |
| Shruti Kaushik named in the answer text | 1 | 4 | 0 |
| Klype named in the list | 0 | 1 (question 45) | 0 |
| klype.ai linked | 0 | 1 (question 45) | 0 |
| shrutikaushik.com cited | 0 | 0 | 0 |
| Answer addressed me as the owner of Klype or Wise Mango | 0 | 8 | 0 |
Positions in the lists: in Google AI Mode, Wise Mango was 4th of 8 for "best linkedin ghostwriters for technical founders". In ChatGPT it was 1st of 6 for "linkedin ghostwriter for enterprise executives", 1st of 5 for "best founder-led marketing agency for b2b startups", 2nd of 9, 3rd of 4, 5th of 6 and 6th of 6 in the other four. Klype was 5th of 5 in ChatGPT's answer to "ai linkedin ghostwriter tool that sounds like me", and that answer added a caveat about Klype's own comparison article.
What I take from this, honestly:
- The ChatGPT numbers for my brands are not reliable. In 8 answers ChatGPT spoke to me as the owner of Klype or Wise Mango, for example by offering "Delhi NCR-based options" or asking whether I was researching competitors for Wise Mango. The interface showed that a temporary chat "can reference memory, plugins, and custom instructions". A stranger asking the same questions would very likely see Wise Mango less often.
- Being cited is not the same as being recommended. In Google AI Mode and Perplexity, wisemango.io was cited 8 times but Wise Mango was named only once. In 7 of those answers my pages fed the answer (mostly price and comparison pages) while other firms made the list. In Google's answer to question 1, the text attributed a comparison to Wise Mango and then listed other agencies.
- Klype is invisible in its own category. For "best b2b marketing intelligence platform", neither Google AI Mode nor ChatGPT named Klype; ChatGPT mentioned it only as my own product. For GTM planning software, the same.
- My personal site was cited nowhere. shrutikaushik.com appeared in 0 of 110 answers.
#Five takeaways for a B2B founder
These follow from the numbers above. They are observations from one day and one browser, not laws.
1. Get onto the lists other people publish. A third of all citations (241 of 732) went to the sites of firms that were not on the shortlist themselves, and sites such as getaob.com and demandbird.com were cited in 7 answers each while rarely being named. The engines build their lists by reading these pages. Being included in other people's comparisons, with a clear one-line description of who you serve, is the most direct route onto an AI shortlist.
2. Make your own site say exactly who you are for. 91% of answers that named firms and linked sources linked at least one named firm's own site, and in ChatGPT 86% of linked citations were the named firms' own pages. When an engine names you, it checks you against your own site. A page that states the buyer, the service and the proof in plain sentences gives it something to quote.
3. Treat LinkedIn as a source, and YouTube and Reddit as Google sources. linkedin.com was the most cited domain overall (36 of 110 answers), cited in half of Google AI Mode answers and 7 of 10 Perplexity answers. YouTube (40% of Google answers) and Reddit (24%) were cited only by Google. Posts that answer a buyer question directly, in public, become evidence for the engines.
4. Check each engine separately. Google AI Mode and ChatGPT shared fewer than three names per question on average and none at all on 7 questions. Only 82 of 364 names appeared in both. Being named in one says little about the other, so any tracking needs to cover each engine you care about.
5. Win the narrow question before the broad one. 69% of all names appeared once. Specific questions ("for fintech founders", "for enterprise executives", "for technical founders") produced different, smaller names than broad "best agency" questions. Category leaders took the broad questions. A smaller firm's realistic entry point is a page built for one specific buyer question, and checking it from a clean account, because personalisation can make your own results look better than a stranger's.
If you want to see what the engines say about your own company for your own buyer questions, the AI visibility check runs that test, and the AI answer audit covers specific prompts.
#How to cite this study
Suggested citation:
Kaushik, S. (2026, November 13). Which agencies and tools does ChatGPT recommend? 50 B2B buyer questions in ChatGPT, Google AI Mode and Perplexity. shrutikaushik.com. https://shrutikaushik.com/writing/which-agencies-tools-chatgpt-recommends-b2b-2026
Dataset citation:
Kaushik, S. (2026). AI answers to B2B buyer questions, November 2026 [Data set]. https://shrutikaushik.com/writing/which-agencies-tools-chatgpt-recommends-b2b-2026#get-data
Journalists and list authors are welcome to quote any figure here. Please state the date (13 November 2026) and the engines, and note that answers vary by user, location and time.
#Methodology
Questions. 50 questions were written before the run and saved as a data file: 10 per group, phrased as a founder would type them, in lower case. Apart from LinkedIn, the only brands named in the questions were Taplio, AuthoredUp and Supergrow (tools group), Clay (one GTM question) and the AI engines themselves (GEO group). The list is in the CSV and in the questions file, both sent from the data form on this page.
Engines and settings.
- Google AI Mode: opened as
google.com/search?udm=50&q=<question>in a fresh page. The answer was read after its text had stopped changing for at least 4.5 seconds. - ChatGPT: opened as a Temporary Chat with web search selected (
chatgpt.com/?temporary-chat=true&hints=search&q=<question>), on the author's free account. The question was sent as prefilled and the answer read after it stopped changing for 6 seconds. - Perplexity: opened as
perplexity.ai/search/new?q=<question>on the author's free account in its default mode. Cited domains were read from the answer's Links tab.
Date and coverage. All answers were collected on 13 November 2026 from a Chrome browser in India. Google AI Mode answered 50 of 50 questions and ChatGPT 50 of 50. Perplexity answered questions 1 to 10, then displayed "You've reached your free search limit" at question 11, and showed the same message when checked again at the end of the run. It was stopped, as no limit was bypassed and no account was changed. Questions 11 to 50 are listed in the CSV for Perplexity with the status "not collected".
Retries. Five answers needed another attempt, and each is marked in the CSV status column. ChatGPT sometimes returned "Your account changed. Send your message again" instead of an answer (questions 5, 14 and 34), and those questions were sent again. Two ChatGPT captures were lost when a browser command timed out (questions 34 and 35) and were run again. One Google AI Mode capture (question 28) was taken before the answer had finished and was re-run. For ChatGPT question 15, the list of names was captured in full but the cited domains were cut off after the first three; two attempts to run it again returned the same "account changed" message, so the first capture is kept and the CSV notes the gap.
What counts as a name. A company, product or person shown in bold, as a heading, or as the first cell of a table row in the answer. Labels that were not firms (price ranges, feature names, section titles, follow-up questions) were removed by hand. Engines that are named as the subject of a question (for example "Perplexity" and "Google AI Overviews" in question 29) were not counted as recommendations. Small spelling variants were merged ("Understory" and "Understory Agency", "In Motion" and "InMotion Marketing", "Semrush AI Toolkit" and "Semrush AI Visibility Toolkit"). A firm mentioned only in plain sentence text, without bold or a table row, is not counted. Counts are the number of answers naming a brand at least once.
What counts as a citation. Every distinct external domain linked from the answer. Each domain counts once per answer. LinkedIn country subdomains were merged into linkedin.com and other subdomains into their parent domain. In ChatGPT, some sources appear as unlinked labels next to a sentence; those were not captured as domains, so ChatGPT citation counts are a lower bound. In one ChatGPT answer (question 42) an unlinked label read "Klype AI"; this is noted in the CSV.
Source types. LinkedIn, YouTube, Reddit, Instagram, Quora, Facebook, Medium, Scribd and SlideShare were classed as platforms. Directories and marketplaces were G2, Clutch, GoodFirms, Sortlist, Capterra, Techjockey, Ghostwriter Index, Upwork, Toptal, GrowthMentor and Fiverr. Big-vendor blogs, publishers and analysts were Gartner, HubSpot, Ahrefs, Semrush, Zapier, Business.com, SitePoint, Razorpay, Geekflare, Adobe, OpenAI Help and one university site. A domain was counted as a named brand's own site when its name matched a brand named in the same answer. Everything else was counted as another firm's site.
Personalisation. All three engines were used while signed in to the author's accounts, in the author's browser, in India. Google says AI Mode can personalise answers using past searches and activity. For ChatGPT, OpenAI's Temporary Chat FAQ says a temporary chat will not access memories but will still follow custom instructions, while the ChatGPT interface on the day of the run said a temporary chat can reference memory, plugins and custom instructions. In practice, 8 ChatGPT answers addressed the asker as the owner of Klype or Wise Mango, and 7 used Indian rupees, Delhi NCR or India-first sections (questions 2, 4, 9, 12, 13, 14 and 37). ChatGPT results for the author's own brands should therefore be treated as personalised and are reported separately. Anyone repeating these questions from another account, location or day will see different lists.
Brand checks. Every answer's text was searched for "Wise Mango", "wisemango", "Shruti Kaushik", "Klype" and "Kurattor", without case. The CSV column own_brand_check records what was found and in what role (named in the list, cited, mentioned in text, or addressed as the asker's company).
Limits. One run per question per engine, on one day, in one browser and one country. Perplexity covers only one group. Name extraction depends on formatting. Domains were recorded, not full page addresses, so this study cannot classify individual pages as listicles. The author runs two of the brands measured.
Data. ai-answers-b2b-2026-11.csv (sent by email from the form on this page): 766 rows, one per name per answer (or one row for an answer with no names). Columns: date, engine, question_id, question_group, question, rank, name, cited_domains, own_brand_check, status, notes.
#Sources
- Original data: ai-answers-b2b-2026-11.csv, answers from ChatGPT, Google AI Mode and Perplexity collected on 13 November 2026.
- Question list: ai-answers-b2b-2026-11-questions.json, written 13 November 2026 before the run.
- OpenAI Help Center, Temporary Chat FAQ.
- Google Search Help, Personal Intelligence: How AI Mode in Search personalizes responses for you.
- Earlier study on this site: Who AI names when B2B founders ask for LinkedIn ghostwriters and GTM help, 92 questions in Google AI Mode, 9 and 10 November 2026.
- Earlier study on this site: Who does AI cite for B2B buyer questions? 50-query study, Google AI Mode, 17 October 2026.