Ramp and Brex achieved the same inclusion result in our ChatGPT recommendation panel: 20 appearances in 20 answers. They did not receive the same narrative position. Ramp appeared before Brex in every answer.
The source record makes that gap more interesting. ChatGPT visibly cited Ramp-owned pages 39 times and Brex-owned pages 41 times. Both brands had abundant first-party evidence. Equal inclusion and near-equal owned citation recurrence still produced a 20-to-zero order result.
That finding supports a narrow conclusion: in this prompt panel, Ramp occupied the general default position while Brex was repeatedly framed as a high-growth or global specialist. It does not tell us whether live search, information learned during model training, brand familiarity, answer templates, or another mechanism caused the order.
What we actually asked
The prompts did not name Ramp or Brex. They described four finance-team situations, and we ran each prompt five times in a fresh Temporary Chat.
Two prompts show the range of the panel:
Which finance platforms should a VC-backed US startup with global travel consider for corporate cards and spend management? Explain tradeoffs.
Compare good spend-management options for a 1,000-person US company that wants procurement, bill pay, and expense automation.
The first prompt introduced a buyer profile commonly associated with Brex: venture backing, technology, and global travel. The second moved toward a larger organization and a broader finance-operations stack. Ramp still appeared first in all ten answers.
The other two scenarios tested a 200-person US technology company needing controls, reimbursements, and accounting integrations, and a US finance team replacing manual expense reports while improving card controls.
| Buyer situation | Distinguishing requirements | Runs | Inclusion | First target brand |
|---|---|---|---|---|
| 200-person US technology company | Controls, reimbursements, accounting integrations | 5 | Ramp 5; Brex 5 | Ramp 5 |
| VC-backed US startup | Global travel and spend management | 5 | Ramp 5; Brex 5 | Ramp 5 |
| 1,000-person US company | Procurement, bill pay, expense automation | 5 | Ramp 5; Brex 5 | Ramp 5 |
| US finance team | Replace manual reports, improve card controls | 5 | Ramp 5; Brex 5 | Ramp 5 |
| Total | Four unbranded buyer situations | 20 | Ramp 20; Brex 20 | Ramp 20 |
The complete wording of all four prompts is preserved in the study record. Showing two here gives readers the experimental contrast without turning the article into a prompt appendix.
The answers did more than change the order
First appearance can be a formatting choice, so we recoded the language surrounding each brand rather than treating position as a complete ranking.
Fourteen of the 20 answers gave Ramp an explicit label such as “default,” “best overall,” or “first choice” near its initial description. Nineteen connected Ramp with a broad finance-operations combination such as cards, expenses, accounting automation, accounts payable, or procurement.
Brex appeared in every answer and received positive treatment. Its assigned role was more specific. Fifteen answers connected its initial description to a startup, technology, fast-growth, or venture-backed profile. Fourteen connected it to global, international, distributed, or multi-entity needs.
This was not an absence-versus-presence result. It was a default-versus-specialist result.
The global-travel prompt makes that distinction especially clear. Brex remained part of every shortlist, and the answers emphasized its international and multi-entity relevance. But the prompt did not move Brex ahead of Ramp in any of its five runs. Other companies, including Navan and Airwallex, also gained prominence when travel and foreign-currency operations entered the question.
What sources did ChatGPT visibly cite?
Every valid answer used web search and displayed source links. We preserved those destination URLs and counted each visible citation occurrence. A repeated link counts repeatedly because recurrence across answers is part of the observable source footprint.
| Visible source category | Citation occurrences | What appeared |
|---|---|---|
| Ramp-owned | 39 | ramp.com 24; support.ramp.com 15 |
| Brex-owned | 41 | brex.com 39; developer.brex.com 2 |
| Other vendor-owned | 118 | Navan, BILL, Paylocity, Airbase, Airwallex, Payhawk, Coupa, and others |
| Independent review or analyst | 3 | Gartner 2; G2 1 |
| Total | 201 | 198 were controlled by a financial-product or software provider |
The panel’s visible evidence environment was overwhelmingly commercial. It was not limited to Ramp and Brex, but it was dominated by companies describing their own products. That is itself an AI-visibility finding.
It also changes how we interpret the result. The earlier version of this analysis compared brex.com with ramp.com and missed Ramp’s separate support domain. Once company-owned domains are normalized, target-owned citation recurrence is nearly equal: 41 for Brex and 39 for Ramp. Citation volume did not explain Ramp’s perfect first-position record.
Visible citations are not a complete source log. They show the pages linked in an answer, not everything retrieved, everything encoded in model parameters, or the internal weight assigned to each signal. The consumer interface did not reveal whether the ordering came from live search, associations learned before the query, or a combination of both.
Does evidence outside the brands support the answer framing?
Independent evidence does not explain the model’s mechanism, but it can test whether the roles assigned in the answers also exist outside vendor messaging.
The strongest quantitative corroboration we found comes from The SaaS CFO’s 2025 finance and accounting technology survey. The survey included 637 participants, mostly in the United States and across SaaS company sizes. In its corporate cards and spend-management chart, Ramp received 18% of responses and Brex 8%; Ramp held the top vendor position, while Brex ranked behind American Express. This is an audience-specific, self-selected survey—not general market share—but it provides external evidence that Ramp had broader reported adoption in a population resembling several of our prompts.
Review platforms show less separation in product satisfaction. G2’s Ramp–Brex comparison displayed both at 4.8 out of 5 when we checked, with 2,398 Ramp reviews and 1,559 Brex reviews. Review counts and ratings can change, and reviewers are not representative of all customers. Still, the larger Ramp review footprint provides more public, customer-generated language around its category role.
Capterra’s review-based comparison independently reproduces much of the role split seen in our panel. It describes Brex as a fit for global or multi-entity operations and Ramp as especially strong in spend-workflow automation and policy controls. That does not prove ChatGPT used Capterra’s framing—the panel visibly cited Capterra zero times—but it shows that the default-versus-global-specialist narrative exists beyond the brands’ own sites.
The external evidence therefore corroborates the pattern without establishing its cause. Ramp had broader adoption in one relevant survey and a larger review footprint on G2. Brex matched or slightly exceeded Ramp on several G2 reviewer-rated dimensions and received a distinct global/multi-entity role on Capterra. The panel flattened those nuances into a consistent order.
Owned content mattered, but it did not explain the ranking
Official pages remain the best source for current capability claims. Ramp’s product overview covers cards, expenses, bill payments, procurement, travel, and accounting automation. Brex’s spend-management page covers cards, reimbursements, bill pay, business accounts, travel, budgets, policies, and global operations. Both brands make a broad platform case.
Brex’s global product material gives the specialist association substantial owned support. Ramp’s integration directory makes its finance-stack breadth easy to retrieve. These pages help explain why both companies were present and described accurately. They cannot, by themselves, explain why Ramp came first every time.
Corporate context also changed shortly before our test. Capital One completed its acquisition of Brex on April 7, 2026. We found no basis for claiming that the acquisition affected answer order.
What we know, and what remains unknown
The evidence supports three firm observations:
- Inclusion tied: Both companies appeared in all 20 answers.
- Order did not: Ramp appeared first in all 20 and received explicit default language in 14.
- Sources concentrated: Only 3 of 201 visible citation occurrences came from independent review or analyst properties.
The strongest supported interpretation is that Ramp had become the panel’s general category default, while Brex retained a strong but more conditional high-growth/global role. External survey and review evidence is consistent with that interpretation.
We do not know how that pattern formed. ChatGPT may have drawn on associations learned from previously crawled material, selected live search results, reused a familiar recommendation structure, or combined all three. A search-enabled answer can display citations without those citations fully explaining its ordering. The interface does not expose enough provenance to separate these mechanisms.
A causal study would need to manipulate the evidence environment: hold prompts constant, change or remove specific source classes, compare search-on with search-off conditions where available, and repeat the panel across models and dates. This study did not do that.
What Brex should take from the result
Brex does not need a generic “publish more” prescription. It had slightly more target-owned citation occurrences than Ramp and appeared in every answer. More first-party volume is not the demonstrated gap.
- Broaden external proof: Build independently reviewable evidence for ordinary US finance-team workflows, not only startup, technology, and global use cases. The measurement target is a change in assigned role, not merely another citation.
- Test default language: Track how often unbranded answers call Brex the default, best overall, or first choice across mainstream mid-market scenarios. Inclusion is already saturated in this panel.
- Preserve differentiation: Keep the global and multi-entity story. It was consistently legible and commercially meaningful even when Brex appeared second.
- Measure source mix: Monitor whether independent customer, analyst, and practitioner sources begin appearing beside first-party pages. Three independent citations across the entire panel is a weak external evidence layer for the category, not just for Brex.
What Ramp should take from the result
Ramp’s position creates a different responsibility. Broad default status can spread outdated or overgeneralized claims quickly.
- Protect accuracy: Keep country coverage, plan gates, integration depth, implementation requirements, and eligibility language easy to verify.
- Watch role drift: Measure whether “best overall” language persists when prompts become more global, regulated, procurement-heavy, or ERP-specific.
- Earn external depth: Do not treat first position as proof that owned content is sufficient. Independent evidence was scarce across the panel, and answer order is not a product-performance measure.
The marketing lesson
A mention-share dashboard would call this comparison a tie. A citation dashboard might also call it almost even. Both would miss the most stable behavior in the dataset.
AI-visibility measurement should keep four outcomes separate: inclusion, first position, assigned buyer role, and visible source mix. For this panel, the decisive signal was not whether Ramp and Brex were discoverable. It was which company the answer treated as the default before introducing specialized alternatives.
That is also the right unit for action. A brand that already appears everywhere should not optimize for another mention. It should decide whether the role attached to its name matches the buyers it wants—and then test whether stronger external evidence changes that role.
Sources and methodology
AgenQuest ran four unbranded buyer prompts five times each, producing 20 valid answers on August 27, 2026. Testing used the signed-in ChatGPT consumer web interface from the United States, a fresh Temporary Chat for each run, web search, and the visible Medium response-depth setting. Both brands received identical treatment because neither appeared in the prompts.
We coded brand inclusion, first target-brand appearance, explicit default/best language, recurring buyer-role language, and visible destination URLs. We classified source ownership at the domain level. The result is directional: repeated model calls are not a representative sample of all users, and visible citations do not reveal complete model reasoning.
Current product claims were checked against Ramp and Brex documentation. Comparative context came from The SaaS CFO survey, G2, and Capterra. Review and survey evidence has selection effects and was used as corroboration, not as proof of product quality or model causation.
About this analysis
AgenQuest Research independently conducted this analysis using publicly available information and AI responses observed on the dates stated. Ramp and Brex did not sponsor, review, or participate. AI outputs are non-deterministic and may change. The findings describe the tested prompts and period, not every possible user experience or either company’s overall quality.