What Fintel Connect found
Fintel Connect, a financial affiliate network, put 800 questions to ChatGPT, Claude and Gemini in August 2026 and studied the 2,400 answers. The questions covered credit cards, savings accounts and personal loans in the United States. None named a brand.
| Finding | Why it matters to a lender |
|---|---|
| About half the answers used no live web source | Your current pages cannot influence these answers |
| In those answers, a brand was named about a quarter of the time, usually a large bank | Answers from memory favour the best-known incumbents |
| When the model searched, it named a brand nine times in ten | Search is where a smaller lender gets into the answer |
| Decision questions pulled a live source 90% of the time, awareness questions 9% | Pages aimed at people choosing get read; explainers mostly do not |
| Comparison pages sat under 72% of answers that recommended a product | Third-party coverage carries most recommendations |
| A brand was named first 25% of the time when its own page and a comparison page were both read, 11% when neither was | The two sources work together |
| The first brand named got about 90 words of description, the third about 68 | Position decides how much the reader hears about you |
Fintel tested one model per platform, once each, over three weeks. Its findings on page format and question wording held across its 2025 and 2026 studies. Its split by platform did not: ChatGPT moved from mostly comparison sources to mostly brand pages in a year.
What it means in Canada
The study is American, so its lists of banks and comparison sites do not transfer. How the models behave should. Canadian application data shows which lenders that affects.
Across Smarter Loans personal loan applications from August 2025 to July 2026, fair credit was the largest self-reported band in every province with enough applications to report. The exceptions were Saskatchewan and New Brunswick, where low credit led. Fewer than one in six people who applied reported good or great credit in any of those provinces.
Across the same Smarter Loans applications from August 2025 to July 2026, about seven in ten asked for $1,500 or less.
In Fintel's data, a model answering from memory names the largest banks. A lender built for fair credit and smaller amounts is unlikely to come up that way. It has to be found when the model searches, through pages it can read at the moment someone asks.
Our take
Most of what the report prescribes for pages matches what we see models read. That means a rate with a date beside it, and fees and eligibility in plain text rather than a PDF. It also means comparison pages that rank options rather than list them. Fintel found three quarters of comparison-site citations were ranked or comparison pages.
Where we read it differently is the role of comparison sites. The report treats them as a channel a bank buys into, but they are also a second source the model reads beside your own page. Fintel notes that conflicting information can reduce the chance a brand is included. A rate on your own site that differs from your profile elsewhere is that conflict.
The report also frames model memory as something that shifts over years. That holds for answers built without search. For answers built from search, a page published this week can be read this week. The report's own decision-stage figures show it.
Platform-by-platform advice goes out of date within a year. The findings worth planning around are the ones that held across both years: page format, dated figures and question wording.
What this means for lenders
- Put your rate, fees, amounts, terms and who qualifies on one plain-text product page, with the date you last checked them beside the rate. Fintel found product and rates pages made up close to six in ten citations of brands' own sites.
- Say who the product is not for, as well as who it suits. Models match a person to a product, and the exclusions make the match precise.
- Name the provinces you lend in on the page. Canadian questions often include a province, and a page that does not say where you lend gives the model nothing to match.
- State your APR range as a number. "Competitive rates" gives a model nothing to quote.
- If you lend in Quebec, publish the French version of the product page.
- Keep your profiles on comparison sites current, so your own page and the comparison page say the same thing.
- Write for people choosing: eligibility, current rates and how you compare. Explainers are rarely read when a model searches.
Lenders in the Smarter Loans network can see how their profile reads in the lenders directory. More of our data is at Smarter Loans Research.
Frequently asked questions
Do AI assistants use comparison sites for financial recommendations?
Mostly. Fintel Connect found about seven in ten AI answers that recommended a financial product read at least one comparison or affiliate page. The rest relied on brand pages, forums or training.
How can a lender appear in ChatGPT answers?
Through pages the model reads when it searches. Fintel found models search mainly on decision questions, and read plain-text product pages with dated rates and comparison pages that rank options.
Why do AI answers name the big banks?
When a model answers without searching, it names brands it learned in training, mostly the largest. Fintel found about half of 2,400 answers used no live source.
Does Fintel Connect's study apply to Canada?
Its findings on how models search and what they read should carry over. Its lists of leading banks and comparison sites are American and do not transfer directly.
About this study
Fintel Connect's 2026 AI Search Guide for Financial Marketers put 800 non-branded consumer questions, once each, to ChatGPT, Claude and Gemini through their APIs between August 5 and 24, 2026. It covered credit cards, business credit cards, high-yield savings and personal loans in the United States. It recorded 7,809 citations. The report is published at Fintel Connect. Smarter Loans figures are drawn from personal loan applications between August 1, 2025 and July 31, 2026. Credit band is self-reported at application.
Written by:
The Smarter Loans Editorial Team produces in-depth, original content to help Canadians navigate borrowing, credit, and personal finance with confidence.
