Gist Research · CRM · October 2026
AI can cite your content and recommend your competitor.
In repeated CRM answers, Perplexity linked to one vendor’s content while suggesting another vendor’s product. The source list only told half the story.
01 / Gist Research
A citation can look like a win. We checked the recommendation.
We examined ChatGPT, Perplexity and Google’s AI Overviews using questions about choosing and using CRM software.
For a software company, a link in an AI answer can look like good news: its website has appeared as a source. But a source link alone does not tell the company whether its product was recommended.
How it works
The answer can include product suggestions and source links.
AI suggests a product
That’s a recommendation.
AI links to a source
That’s a citation.
From choosing to using
Our questions covered CRM and related software, from selecting a tool to working with it every day.
We saved multiple answers to each question across three AI services, counted source links and reviewed selected groups for product recommendations.
The comparisons use the same 30 CRM questions across services. Full counts and methods are in the methods below.
02 / Gist Research
Two CRM companies got the links. Their rivals got the recommendations.
Perplexity linked to Capsule and Salesflare articles while recommending Salesforce, HubSpot and other CRM products.
“What kind of CRM fits a B2B sales team with several decision-makers involved in each deal?”
Capsule
capsulecrm.com/blog/b2b-crm/
Salesflare
blog.salesflare.com/best-b2b-crm
Neither company is named in the answer text.
- Salesforce
- Large or highly customized sales operation
- HubSpot
- Sales and marketing alignment
- Pipedrive or Zoho CRM
- Smaller team, simpler way to track deals
“For a large or highly customized sales operation, consider Salesforce; for sales-and-marketing alignment, HubSpot; and for a smaller team needing a simpler visual pipeline, Pipedrive or Zoho CRM.”
Perplexity · saved answer index 3 · CRM-X02-1 [E2]
saved responses to this question showed the same relationship.
Different wording; the same source links each time.
A product suggested for a particular need counts as a recommendation. This is one question, not a rate for the whole CRM category. The links do not establish which page supplied the product advice.
03 / Gist Research
Cited vendors missed the shortlist on three CRM questions.
Two Perplexity questions showed this in every saved answer. A third question produced the same outcome in eight of 23 Google AI Overview responses.
CRM suppliers linked but not recommended
Responses suggesting competing products, out of all responses to that question
View the data table
| Comparison | Answers | Denominator |
|---|---|---|
| Small-business CRM · Perplexity | 12 | 12 |
| Complex B2B CRM · Perplexity | 28 | 28 |
| Minimum CRM setup · AI Overviews | 8 | 23 |
Small-business CRM
Salesflare and NetHunt were linked as sources but not named in the advice. The 12 responses used different wording and the same links.
Complex B2B CRM
The Capsule and Salesflare example from the previous section: 28 differently worded responses, with the same source links.
Minimum CRM setup
Nutshell and NetHunt were linked while other products were suggested. Seven of the eight qualifying responses were identical: only two different texts showed the pattern.
Selected cases, not a category-wide rate. The AI Overview set contains 17 distinct texts in 23 responses. Companies in a row share those responses.
04 / Gist Research
ChatGPT cited Salesforce and HubSpot without naming them in the answer.
Across the 30-question ChatGPT comparison, we checked whether cited companies also appeared by name in the answer.
Linked as a source, absent from the written advice
ChatGPT · 30 CRM questions · totals count responses linking to each company
View the data table
| Comparison | Answers | Denominator |
|---|---|---|
| Salesforce | 168 | 732 |
| HubSpot | 128 | 501 |
questions had links to Salesforce without its name in the written advice.
questions had links to HubSpot without its name in the written advice.
Some questions call for instructions, not product choices.
For service businesses that rely on referrals, ChatGPT linked to Salesforce in 40 of 44 responses and named it in six. The reviewed answer discussed customer history, reminders and referrals. [E5]
These counts check exact names, not recommendations or every alternative product name. Names may still be visible in the linked-source display. Both company totals come from the same 1,016 ChatGPT responses.
05 / Gist Research
SoftwareGlimpse appeared in 26 of 48 CRM trial answers.
ChatGPT linked to its trial guide and evaluation checklist on one of the 30 questions, and nowhere else in that comparison.
ChatGPT questions included a link to SoftwareGlimpse.
“During a CRM trial, what daily tasks should sales reps test to compare usability and the time spent keeping records up to date?”
Websites linked in answers about CRM trials
ChatGPT · 48 responses · each used different wording
View the data table
| Comparison | Answers | Denominator |
|---|---|---|
| SoftwareGlimpse | 26 | 48 |
| Salesforce | 24 | 48 |
| Zoho | 6 | 48 |
| HubSpot | 3 | 48 |
The study did not measure company size, traffic or market share. The two-answer difference from Salesforce is not a statistically established lead.
06 / Gist Research
ChatGPT often cited Salesforce. AI Overviews often cited YouTube.
The same 30 CRM questions put Salesforce in 72% of ChatGPT responses and YouTube in 46% of Google AI Overview responses.
Share of responses linking to each website
Same 30 CRM questions · a response can link to several sites
ChatGPT 1,016 answers
Perplexity 616 answers
AI Overviews 876 answers
View the complete data table
| Engine | Answers | Salesforce | HubSpot | YouTube |
|---|---|---|---|---|
| ChatGPT | 1,016 | 72% | 49% | 0% |
| Perplexity | 616 | 16% | 18% | 0% |
| AI Overviews | 876 | 24% | 12% | 46% |
ChatGPT questions included links to Salesforce.
AI Overview questions included links to YouTube.
Source comparisons also depended on how repeated responses were counted.
The collection methods are in the methods below. Giving each ChatGPT question equal weight yields 71.4% for Salesforce. Keeping one AI Overview response per question and exact text leaves YouTube in 277 of 626 (44.2%). These are observed source patterns, not proof that a content format earns citations.
07 / Gist Research
A 12-to-4 citation lead became a 4-to-4 tie.
On one Google AI Overviews question, counting each answer text once put OnePageCRM level with Salesforce.
We asked when teams should use multiple sales pipelines. The 19 saved responses contained 11 different answer texts. The comparison below shows the counts before and after keeping one response for each text.
12OnePageCRM
4Salesforce
4OnePageCRM
4Salesforce
| Source | All saved answers | One per distinct text |
|---|---|---|
| NetHunt | 17/19 | 9/11 |
| OnePageCRM | 12/19 | 4/11 |
| Pipedrive | 11/19 | 4/11 |
| Salesforce | 4/19 | 4/11 |
NetHunt remained the most cited of the four companies in both views. OnePageCRM’s advantage over Salesforce came from the repeated responses. [E10]
Google AI Overviews, one question. We retained the first saved response for each exact text. Identical wording can have different links; different wording does not prove independent source selection.
08 / Gist Research
Perplexity suggested Folk CRM’s rivals in eight of 48 answers.
Every response to the multiple-pipeline question cited Folk CRM through the same source list; the product advice varied.
Eight responses presented competing products as options. Two described a style of product by naming a brand; excluding those leaves six. The other 40 did not present named product options under our review rules. [E15]
One automated score missed two suggested products
One ChatGPT answer about multiple pipelines described three products as “worth evaluating.” The automated score recorded only HubSpot, missing Pipedrive and Salesforce. Reading the answer prevented us from treating those companies as cited but passed over. [E17]
For the 48-response set, collection ended before brand-mention shares met the study’s stability threshold. Across Perplexity, 25 of 30 questions kept the same source links despite different wording in every response. These are short collection windows, not trends over days. [E16]
09 / Gist Research
The citation count didn’t tell us which CRM was recommended.
Reading the answers revealed who was suggested, which questions brought each source into view, and whether the result repeated.
In the B2B sales example, Capsule and Salesflare appeared in the source links while Salesforce and HubSpot appeared in the product advice. That relationship held across all 28 saved Perplexity responses. The cited companies were visible, but their products were not suggested. [E2]
Cited publishers could miss the product shortlist
Other CRM answers also cited vendors without suggesting their products. The source totals showed the publishers’ presence; the text identified the competing products being suggested. [E1, E2, E3]
A site’s presence could depend on one question
SoftwareGlimpse’s ChatGPT citations were concentrated on the CRM trial question. The category total alone would not identify that task or the guide and checklist linked in those answers. [E9]
The report’s central finding is a recurring separation between being cited and being recommended. A single answer showed an example; comparing saved responses established where it repeated and where it varied.
These are selected recommendation cases, not a measured rate across all CRM questions. The study did not test content changes or sales outcomes.
10 / Gist Research
About 60 questions, asked repeatedly across three AI services.
The exact dataset contains 58 questions and 3,869 saved answers. Comparisons across services use the same 30 CRM questions.
| Question set | Questions | ChatGPT | Perplexity | AI Overviews |
|---|---|---|---|---|
| CRM + related software | 28 | 340 | 500 | 521 |
| Focused CRM questions | 30 | 1,016 | 616 | 876 |
| Total | 58 | 1,356 | 1,116 | 1,397 |
The last three columns count saved answers: 3,869 in total. Each question was asked on all three services, producing 174 question/service combinations.
What we asked and collected
We wrote the questions; they are not customer prompt data. Collection took place October 8–9, 2026 (UTC).
What we compared
Source counts include subdomains and count each answer once per company. Repeat checks keep the first saved response per exact text within each question and service.
What we read
We reviewed recommendation passages in 111 answers across four selected question/service groups, checking full answers when needed. Suggestions for a particular need count as recommendations. These cases do not establish a category-wide rate.
Why the answer counts vary
Collection checked whether brands’ shares of mentions had stabilized. Of 174 combinations, 113 met that rule, 34 reached the collection limit and 27 stopped under a rule for cases without a detected brand signal.
The tool’s 95% convergence setting concerns that mention-stability check. It does not make each report finding 95% certain or validate the separate recommendation review.
How we checked the evidence
We checked saved text and source links against raw archives and triaged 1,137 insight headlines. We rejected a ranking claim after a generic term was counted as monday CRM, and checked product/parent-company naming splits.
We collected forced-search ChatGPT responses through an API workflow, Perplexity Sonar search, and AI Overviews through SerpAPI. These workflows do not represent every consumer experience. Claude was not measured. Different answer texts do not prove independent source retrievals. Sales, content effects, market share and trends over days were not measured. [M1]
11 / Gist Research
The evidence behind the report.
Each reference connects a finding to the saved questions, responses and source links in the accompanying package.
| Ref | Question / evidence location |
|---|---|
| E1 | Best small-business CRM · Perplexity · B03 |
| E2 | B2B sales with several decision-makers · Perplexity · X02-1 |
| E3 | Minimum small-team setup · AI Overviews · X05-1 |
| E4 | Cited domain × literal name in prose · 30 CRM questions |
| E5 | Referral-driven service businesses · ChatGPT · X02-2 |
| E9 | Daily tasks to test in a CRM trial · ChatGPT · X06-2 |
| E10 | When to use multiple pipelines · AI Overviews · X10-3 |
| E11 | Domain presence across the same 30 CRM questions |
| E15 | Multiple pipelines: source and shortlist variation · Perplexity · X10-3 |
| E16 | Exact citation-set repetition · Perplexity · 30 CRM questions |
| E17 | Worth-evaluating passage versus scored labels · ChatGPT · X10-3 |
| M1 | Definitions, convergence scope, alias checks and repeat sensitivity |
The reference labels identify the saved question, engine and counting rule behind each finding. Read the complete 58-question appendix. Saved answer numbers are zero-based indices within a question and engine.
Go deeper
The full category. A wider view.
For five narrower examples across ChatGPT and Perplexity, read 5 Things a CRM Citation Count Cannot Tell You, with selected answers and an exact source-set check.
This feature follows the gap between citations and recommendations. The larger CRM study adds company rankings, question-level comparisons and a separately reviewed recommendation snapshot across all 58 questions.
Explore the larger report