Study
Six independent CRMs are named in one AI answer out of ten, and never by ChatGPT
Founder of Epovest
Measured on Sep 2, 2026 AI engines: ChatGPT, Claude, Perplexity Questions asked in English
A small business looking for a CRM asks an assistant, and takes the shortlist it gets back. We put twelve of those buying questions to ChatGPT, Claude and Perplexity on the same day, three times each, and counted who was named.
What came back is not one market. It is three.
Four names carry every list: Pipedrive, Zoho CRM, HubSpot CRM and Freshsales appear on all three engines, in that order of weight. Below them, the assistants stop agreeing, and the disagreement is not a matter of a few points.
What each assistant recommends
Share of voice counts how much of the named-vendor space each name takes, engine by engine, over its 36 answers.
| CRM | ChatGPT | Claude | Perplexity |
|---|---|---|---|
| Pipedrive | 23.9% | 15.6% | 21.6% |
| Zoho CRM | 19.4% | 16.2% | 13.8% |
| HubSpot CRM | 14.9% | 12.3% | 13.8% |
| Freshsales | 14.2% | 5.0% | 4.3% |
| monday CRM | 11.9% | 5.0% | 0% |
| Salesflare | 0% | 7.8% | 10.3% |
| Nutshell | 0% | 7.3% | 6.9% |
| Less Annoying CRM | 2.2% | 6.7% | 8.6% |
| OnePageCRM | 0% | 3.9% | 7.8% |
| Bigin | 3.7% | 4.5% | 3.4% |
| Insightly | 0% | 3.9% | 2.6% |
| Streak | 3.0% | 2.8% | 1.7% |
| Attio | 3.0% | 1.1% | 0% |
| Scoro | 2.2% | 1.1% | 2.6% |
| NetHunt | 0% | 2.2% | 0.9% |
| EngageBay | 0% | 2.2% | 0% |
| Capsule CRM | 1.5% | 1.7% | 0% |
| Close CRM | 0% | 0% | 1.7% |
| Keap | 0% | 0.6% | 0% |
Read the ChatGPT column downward and the pattern arrives on its own: after the five largest names and three smaller ones, it stops. Salesflare, Nutshell, OnePageCRM, Insightly, NetHunt and EngageBay hold between 3.9% and 10.3% of the answers on Claude, on Perplexity, or on both. On ChatGPT they hold none.
That is not a small brand being ranked low. It is a buyer being shown a market that does not contain them.
The corpus explains it
Every answer in the run was archived with the sources it cited, and the three engines do not read the same web. Over the same 36 answers each, Perplexity cited 680 pages, Claude 249, and ChatGPT 99. What they cite differs more than how much.
| ChatGPT | n | Claude | n | Perplexity | n |
|---|---|---|---|---|---|
| hubspot.com | 24 | nutshell.com | 14 | reddit.com | 63 |
| salesforce.com | 12 | blog.salesflare.com | 11 | monday.com | 30 |
| pipedrive.com | 11 | onepagecrm.com | 10 | uschamber.com | 27 |
| zoho.com | 10 | salesforce.com | 7 | blog.salesflare.com | 26 |
| techradar.com | 7 | insightly.com | 7 | onepagecrm.com | 24 |
| monday.com | 5 | zeeg.me | 6 | nutshell.com | 24 |
Three different diets, and each column reads like the column of names it produces. ChatGPT builds its answers largely from the large vendors' own sites, and names the large vendors. Claude reads the independents' blogs, and names the independents. Perplexity reads Reddit before anything else, 63 citations out of its 680, then vendor blogs.
Across the whole run, the five most cited domains are reddit.com (63), monday.com (39), nutshell.com (38), blog.salesflare.com (37) and onepagecrm.com (34). Four of the five are vendors writing about their own market.
Being read is not being named
Put the two tables side by side and one line stands out. On Perplexity, monday.com is the second most cited domain, 30 citations, and monday CRM is named in 0% of that engine's answers. Its pages are used to write answers that recommend somebody else.
Nutshell, Salesflare and OnePageCRM sit in the same place on ChatGPT: among the most cited domains of the study, named in none of its answers.
The run also records the pages an engine opened without quoting them. At the top of that list, bigcontacts.com is read 14 times and named in no answer at all, on any of the three engines. Then zapier.com (8), onepagecrm.com (7), monday.com (7) and zeeg.me (6).
What this changes for what you do
There is no single answer to win. A buyer who opens ChatGPT and a buyer who opens Claude are shown two different shortlists of the same market, on the same day, and a strategy aimed at one of them is invisible on the other. The first thing to know is which assistant your buyers actually use.
Your own writing is part of the corpus. The independents that exist at all in these answers are the ones whose posts are cited, between 24 and 38 times each across the run. That is not a side effect of publishing; on this market it is the mechanism.
And the gap between being read and being named is where the next move sits. A domain that is cited but never mentioned is one whose pages answer the question without carrying the brand into the answer: the name is missing from the sentence the engine keeps, not from the web.
Method
Twelve questions were written the way a small business asks them, from "best CRM for a small business in 2026" to "best simple CRM without a long onboarding", including comparison and price questions. Each was put three times to ChatGPT, Claude and Perplexity through their official APIs on 2 September 2026, giving 108 answers, each archived with its cited sources and its date. Twenty-one CRM names were counted. Share of voice is the share of named-vendor mentions a name takes on that engine.
Three draws rather than one, because two identical calls do not return identical answers: on a single-draw run of the same twelve questions earlier that day, Pipedrive measured 47.4% on Perplexity, against 21.6% over three draws. A rate rests on repetition, which is why the resolution is part of the method rather than a detail of it.
The engines went to the web for nearly every answer: 36 times out of 36 on Perplexity, 35 on ChatGPT, 33 on Claude.
Measured with Epovest.