1. A citation does not identify the overall recommendation
We asked which content-management system, or CMS, was best for a small-business website. In the first saved Perplexity answer, a HubSpot CMS comparison article appears in the source list. The answer favors WordPress overall. It also includes HubSpot Content Hub as an option for marketing and CRM integration.
The accurate description preserves both facts: HubSpot receives a citation and a conditional product recommendation; WordPress gets the overall preference. Calling this a straightforward win or loss for HubSpot would discard part of the answer.
It also does not show that HubSpot’s article caused the WordPress recommendation. A source list establishes that the link was present, not the model’s path from source to advice. Evidence C1
2. A company recommendation may be for a different product
That CMS answer discusses website publishing. For a separate question about the best email-marketing software for a small business, the first saved ChatGPT answer recommends MailerLite, Brevo and Mailchimp for different needs, with other email products included. These belong in the broader business-software story, but they are not automatically recommendations to buy a sales CRM.
Companies with several products make this especially easy to miscount. A company name can be correct while the inferred product category is wrong.
Keep the question’s task and the recommended product role beside the company label. Then decide whether the record belongs in a CRM comparison, an email comparison or another view. Evidence C1 and C2
3. A recommendation can appear without an owned citation
The first selected ChatGPT answer about setting up a lead-nurture email campaign lists HubSpot as an option combining email nurturing with CRM tools. Its cited URLs point to Mailchimp resources and an FTC guide; none is a HubSpot-owned URL.
That answer gives HubSpot a product role without giving it a source link. The reverse relationship also occurs in the CMS example: a source link does not establish the overall preference.
A useful report therefore keeps recommendation and citation columns separate. Each describes a different relationship between the business and the answer. Evidence C3
4. More saved answers can repeat the same source set
For the source-repetition check, we grouped the 1,116 Perplexity answers by their 58 original questions. Within each question, we compared the sets of cited URLs. For 49 of the 58 questions, every saved answer carried the same exact source set. Answer counts varied by question; 49 is a count of questions, not answers.
The prose could differ while the source links stayed the same. Repeating one question many times therefore did not necessarily broaden its cited-source list.
That recurrence is useful to record, but it does not turn repeated links into independent endorsements. Nor does it prove caching or reveal how the provider retrieved its sources.
When evaluating breadth, inspect both the answer count and the number of distinct source sets within each question. Preserve engine boundaries and the URL-normalization rule; otherwise, a query parameter can silently change what counts as a different set. Evidence C4
5. A useful answer may advise delaying a CRM purchase
We asked whether a solo consultant with few active clients should use a CRM, a contact manager or existing email and calendar tools. The first saved Perplexity answer recommends starting with existing email and calendar tools plus a structured contact list. It makes adopting a CRM conditional on a concrete follow-up problem.
Several CRM-related sites appear in its citation list. That does not make the answer a recommendation to buy those products now.
The important distinction is between advice about the category and advice to choose a vendor. An answer explaining when a purchase becomes useful can be responsive even when it offers no named CRM recommendation. This is one saved case, not evidence of how often solo consultants need CRM software. Evidence C5
Read the relationship, not just the link
These cases establish that source visibility, company visibility and product advice can diverge within the same answer. Citation counts remain useful when their meaning stays intact. They become misleading when a report quietly converts them into purchase preference, independent corroboration or evidence of revenue.
Methods and evidence
Gist’s CRM research asked how AI helps a small business choose customer-management and related software, and whether the companies it cites are also the products it suggests. The archive covers 58 authored questions, including choosing a website content-management system, setting up email marketing and deciding whether a solo consultant needs a CRM.
This article uses ChatGPT and Perplexity API answers saved on October 8–9, 2026. Four examples each examine one first saved answer to a particular question. A separate source-repetition check uses all 1,116 Perplexity answers across the 58 questions. Those are repeated responses, not 1,116 distinct questions or buyers.
Download the selected evidence for exact questions, excerpts, saved source URLs and collection timestamps. Citation here means a URL in the saved response’s citation list; it does not prove that every linked page supplied a claim.
Method note: examples use the first saved nonempty answer in each cited ChatGPT or Perplexity question–engine cell. They describe model advice, not verified product claims or whole-archive recommendation rates. AI Overviews and all-engine aggregate rankings are excluded. The repetition calculation uses the full Perplexity archive, grouped by the 58 authored questions; exact URL sets ignore order and duplicate URLs while preserving query strings.
- C1: B04, Perplexity, epoch 0, October 8. Small-business CMS choice.
- C2: B07, ChatGPT, epoch 0, October 8. Small-business email software.
- C3: B14, ChatGPT, epoch 0, October 8. First email nurture campaign.
- C4: Perplexity-only source-set recalculation, 1,116 answers across 58 questions. No AIO records included.
- C5: CRM-X01-2, Perplexity, epoch 0, October 9. Solo-consultant tool choice.
Related: The original CRM citation feature. This new companion uses a narrower two-engine scope; older report totals should not be substituted for its denominators.
For a practical reporting format, use our weekly AI visibility report guide. Keep the recommendation role and cited URL separate in each record.