AI citations in finance software

Maxtor Labs ran 10 frozen, non-branded US-English buyer prompts on 5 and 6 August 2026.

We used the ChatGPT, Gemini and Perplexity model families available through the research access layer on those dates.

Each prompt ran once with web search enabled. The remaining study allowance supported three Claude-family responses.

Of 33 submitted responses, 30 contained substantive answers. Those 30 form the cohort’s eligible-answer denominator.

Twenty-two of the 30 eligible answers exposed citations.

The prompt set covered finance jobs including AP automation, spend controls, payments, procurement and travel. The analysis then asked three practical questions:

1. Which sources did the answers cite?
2. Which page types could be identified from the visible title or path?
3. How often was each fixed-cohort brand named, and how often did a named answer also cite its owned domain?

Those questions produce a useful content diagnostic. They do not reveal the full retrieval process or explain why any individual answer was generated.

Vendor-owned pages appeared in all 22 answers that exposed citations. They accounted for 201 of 304 unique cited URLs and 346 of 505 citation occurrences.

Third-party sources supplied 89 of the 304 unique cited URLs.

They comprised 55 editorial, list or comparison URLs and 34 review, analyst or directory URLs.

Community and video sources supplied 14 unique URLs. No URL in this run was classified in the official, government or compliance source group.

In the 22 answers with citations:

• vendor pages: 22/22;
• editorial sources: 18/22;
• review sources: 13/22;
• community or video: 11/22;
• official sources: 0/22.

Across 505 citation occurrences, vendor pages accounted for 346, editorial sources 95, review sources 47, community sources 17 and official sources 0.

The mix suggests that a FinTech marketer should review both sides of the buyer’s evidence path. Your own pages need to answer the question.

Independent pages need to describe the product accurately enough that a buyer can verify the answer elsewhere.

A page type could be resolved for 185 unique cited URLs. The title or path for another 119 Gemini grounding redirects exposed the source domain but not a dependable destination type.

Six commercially useful page types accounted for 172 of the 185 resolvable URLs:

• comparison, list or alternatives pages: 138;
• security or compliance pages: 16;
• category or use-case pages: 11;
• integration or documentation pages: 6;
• customer proof or ROI pages: 1;
• dedicated implementation pages: 0.

The six rows are a selected view, not a complete taxonomy. The remaining 13 resolvable URLs used other labels in the source evidence.

The 119 opaque redirects remain in the 304-URL source totals but are excluded from the 185-URL page-type denominator.

Comparison pages were common in this citation set. Standalone implementation and proof pages were scarce or absent under the selected classification.

Use those patterns to choose pages worth reviewing. They do not imply that publishing a particular page format will improve a future answer.

If an AI answer compares vendors for a finance job, check whether your site helps the buyer weigh the options and their trade-offs.

The fixed cohort contained eight brand-domain pairs: Ramp–ramp.com, Brex–brex.com, BILL–bill.com and Tipalti–tipalti.com.

The other four were Coupa–coupa.com, Airwallex–airwallex.com, Navan–navan.com and Rho–rho.co.

Across 30 eligible answers, the observed results were:

A named answer visibly mentioned the company. A named-plus-owned-citation answer both named the company and cited its canonical domain somewhere in the same response.

The gap between the two columns is useful because a brand can appear without its own site being used as visible evidence.

An answer can also cite a brand’s domain without recommending the company for the buyer’s use case.

The prompt mix matters as much as the count. A company’s result partly reflects how often it fit this fixed set of AP, spend, payments, procurement and travel questions. Rho’s 0 named answers and 0 named-plus-owned citations describe this run only.

Choose a finance job that matters to your business. Trace the pages a buyer needs to assess fit, compare products and check practical requirements.

Let your buyer questions set the priority. The study points to pages worth inspecting; it does not prescribe a universal order.

Show who each workflow fits

A buyer comparing AP automation, spend controls, procurement, travel or international payments needs to know whether the product fits the job.

A focused page should explain the target company, supported workflows, approval behaviour, reporting, geographic coverage and material limits.

Do not force every finance job into one generic platform page.

Make comparison criteria explicit

A useful comparison names the criteria, explains where each option fits and shows real trade-offs.

For finance software, those criteria may include legal-entity support, payment rails, ERP depth, approval flexibility, geographic coverage, migration work and implementation ownership.

Keep product details and dates current. Include cases where another product may be the better choice. A thin competitor page that repeats positioning claims will not answer the buyer’s decision.

Explain what an integration actually does

For each priority ERP or accounting connection, explain supported objects, sync direction, update timing, approval behaviour, permissions, error handling and setup steps.

If support varies by edition, region or paid dependency, state that next to the capability.

Link the overview to technical documentation and implementation guidance. A logo does not tell a finance team whether the connection will work in its environment.

Make security evidence checkable

Finance buyers need more than a row of badges. Explain access controls, audit logs, encryption boundaries, payment-security responsibilities and the route for reporting an incident.

Link to a maintained trust centre where qualified buyers can verify reports and certifications.

State the date and scope of each certification. A SOC 2 report does not automatically cover every product, region or workflow.

Describe the work of implementation

Explain implementation phases, responsibilities, required data, configuration, integration work, training and dependencies.

Replace unsupported “live in days” promises with an evidenced range or example. State the conditions behind it.

Let buyers inspect the result

A strong case study states the starting process, company context, implementation period, calculation method and measured result.

Separate customer-reported figures from independently verified findings. An ROI tool should show its formula and let the buyer change the assumptions.

Review credible category pages, profiles and independent comparisons alongside your own content.

Keep a manageable set of accurate references that your team can maintain.

Give relevant publishers a current fact pack with product scope, supported segments, integration links, security resources, implementation material and customer proof.

Correct inaccurate profiles where the publisher supports updates. Keep pricing models, former brand names and category labels current.

Do not ask a publisher or partner to repeat a claim your own site cannot support.

An independent comparison that points to a vague or outdated landing page gives the buyer less confidence, even if the page appears as a citation.

The prompt themes also map to visible US search demand. DataForSEO Google Ads estimates preserved in the study reported:

• “accounts payable automation software”: 3,600 estimated monthly searches;
• “procurement software”: 1,600;
• “travel and expense management software”: 1,600;
• “business spend management software”: 210;
• “accounts payable automation ROI”: 40.

The evidence file records United States as the location and English as the language. These are close keyword equivalents for the prompt themes, not the original prompts.

Five narrower keyword equivalents returned no search-volume field:

• global business payments software
• finance software ERP integration
• finance software security compliance
• finance software implementation
• finance operations software alternatives

A missing field does not mean nobody asks those questions.

Keyword tools often compress or omit specialist and long-tail demand.

Use integration, implementation and security pages because they remove buying friction.

Measure their value through qualified visits, assisted conversions and sales questions rather than judging each page by search volume alone.

The study used one run per submitted cell, an uneven model sample and a research access layer rather than every consumer interface.

Product claims inside the answers were not independently fact-checked.

We used the five source groups shown above. Page subtypes were assigned only where the visible title or path supported a dependable classification.

Results can change with the prompt, model, interface, market, date and account state.

The findings show source and page patterns worth testing.

They do not prove why a brand was named, why a page was cited or which change would improve a future answer.

The six public CSV files remain unchanged at the existing article path:

• Fixed eight-brand cohort summary
• Sanitised source summary
• US keyword-demand evidence
• Source-mix visual data
• Cohort visual data
• Page-strategy visual data

Test the questions that shape your buyer’s shortlist. Save the visible sources and inspect the pages behind any missing, partial or unsupported answer.