Case study: how Lone Star Credit Union went from 21% to 72% of ChatGPT's North Texas car loan answers
One new page took Lone Star Credit Union from 16 of 75 to 72 of 100 ChatGPT answers about North Texas car loans. What it published, and how sure we are.
- Category: Auto loans · North Texas
- Platform: MetriFi GEO
- Baseline Visibility: 21% (16 of 75 answers)
- Achieved Visibility: 72% (72 of 100 answers)
- Time to Impact: 3 days
- Outcome: WIN
Ask ChatGPT who offers car loans in Rockwall County, Texas, and it answers with a list of local lenders. Before October 2026, Lone Star Credit Union was on that list in 1 of 25 answers, even though Rockwall County residents can join the credit union and finance a car with it.
Lone Star published one new page on October 7, 2026. MetriFi tracked three North Texas car loan questions before and after. Before the page, ChatGPT named Lone Star in 16 of 75 answers (21%). From the publish date through November 5, it named Lone Star in 72 of 100 answers (72%). In Rockwall County alone, Lone Star went from 1 answer in 25 to 25 answers in 31.
This case study shows how one financial institution improved its visibility in ChatGPT answers, using MetriFi GEO experiment 87. It is a separate, later experiment from Lone Star’s East Texas auto refinance result, described in Part 1 of MetriFi’s Finopotamus series. It stands out among MetriFi’s recent experiments for three reasons: the change was large, it showed up on every question tracked, and the results provide strong statistical evidence that the increase was not simply random variation.
What MetriFi measured
MetriFi sent three questions to ChatGPT and recorded whether each answer named Lone Star Credit Union:
- Where can I get a car loan in Greenville, Texas?
- Who offers car loans in Rockwall County, Texas?
- Which local lenders finance cars in Hopkins County, Texas?
Visibility is the share of answers that name the credit union. The baseline was 75 answers, 25 per question. The measurement window ran from the publish date, October 7, 2026, to November 5, 2026, and collected 100 answers. All answers came from ChatGPT (OpenAI).
Lone Star has branches in Dallas, Greenville, Kaufman and Terrell. Rockwall County and Hopkins County are in its membership area, but it has no branch in either one. That gap is where MetriFi saw the opening.
What ChatGPT was doing before the page
For all three questions, ChatGPT answered with a named list of local lenders: American National Bank of Texas, Alliance Bank, Credit Union of Texas, City National Bank of Sulphur Springs and others. Each entry was a lender name, the products it offers, and a place: a branch town, a county, often a street address.
The sources ChatGPT cited for those lists were mostly each lender’s own product pages and branch pages, not “best auto loans” roundups. In Greenville, where Lone Star has a branch page, ChatGPT named Lone Star in 15 of 25 answers. In Rockwall and Hopkins counties, Lone Star was named in 1 of 50 answers across the two, and no Lone Star page tied its auto loans to those counties. MetriFi’s interpretation was that ChatGPT had little local information about Lone Star to draw on there.
The opening was real: lenders were being named. MetriFi’s hypothesis was that a page connecting Lone Star’s auto loans to those counties would supply the missing fact.
What Lone Star published
Lone Star published a single page, Financing a Car in North Texas, under its auto loans section. It is about 1,800 words and is written in the same shape as the lender entries ChatGPT was already producing:
- Each branch town with its street address and the auto products offered there: new and used auto loans, refinancing and preapproval.
- Each of its six core membership counties (Delta, Hopkins, Hunt, Kaufman, Rains and Rockwall), with the eligibility wording from Lone Star’s membership rules.
- How to apply or get preapproved, online, by phone or at a branch.
- A short FAQ.
Just as deliberate is what the page leaves out. It prints no rates; it links to Lone Star’s rates page instead, so the numbers stay current and carry their disclosures. It names no competing lenders. It makes no “best” claims. Lone Star also linked to the page from its main auto loans page and from each of its four branch pages.
The results
| Question asked of ChatGPT | Before | After |
|---|---|---|
| Where can I get a car loan in Greenville, Texas? | 15 of 25 (60%) | 34 of 38 (89%) |
| Who offers car loans in Rockwall County, Texas? | 1 of 25 (4%) | 25 of 31 (81%) |
| Which local lenders finance cars in Hopkins County, Texas? | 0 of 25 (0%) | 13 of 31 (42%) |
| All three questions | 16 of 75 (21%) | 72 of 100 (72%) |
The answers point to where the gain came from. ChatGPT cited the new page itself in 62 of the 100 answers after launch. In Rockwall County, it described Lone Star as a lender whose membership area includes the county. In Hopkins County, it listed Lone Star as a lender that serves county residents, with Greenville as the nearest branch. Those are the facts the page added.
The timing lines up too. For the first three days after publishing, nothing moved in Rockwall or Hopkins. ChatGPT first cited the new page on October 9 and 10. From October 10 on, Lone Star was named in 25 of 26 Rockwall answers.
The gain was a place on the list, not always the top of it. In Greenville, Lone Star was now the first lender named in 22 of the 34 answers that mentioned it, up from 4 of 15 before. In Rockwall, Credit Union of Texas still led almost every answer. In Hopkins, Lone Star was usually added at the end of the list, often under a “nearby” heading. Being on the list is the step that has to come first.
How sure MetriFi is about this result
MetriFi evaluated the results using statistical tests to determine whether the observed increases were likely to reflect random variation. Across all three questions, the increase from 16 of 75 answers to 72 of 100 was statistically significant under conventional testing assumptions. Each question also exceeded the standard 95 percent confidence threshold individually, with the strongest evidence in Rockwall County and Hopkins County and a narrower result in Greenville. Because the experiment involved repeated ChatGPT responses rather than fully independent observations, these calculations should be interpreted alongside the study’s methodological limitations.
The result has limits:
- This is ChatGPT only. Other AI assistants were not measured.
- This is a before and after comparison, not a split test. AI answers cannot be split between two versions of a page at the same time, and ChatGPT itself changes over time. ChatGPT cited the new page in 62 of the 100 answers after launch, which is strong evidence that the page contributed to the gain, though it does not rule out other influences.
- The baseline answers were collected in one batch, the day before the page went live. A baseline spread over several weeks would be a steadier read of where Lone Star stood.
- These are repeated answers to the same three questions, so they are not fully independent of one another.
A new page, not a rewritten product page
Lone Star did not rewrite its auto loans product page to win these answers. It published a separate page whose only job is to say where Lone Star finances cars and who can get a loan there, and linked to it from the product page. The product page still does the selling. The new page answers the location question.
That matches what MetriFi sees across its experiments. In the pooled results published in Part 2 of the Finopotamus series, new pages and improved pages raised visibility about equally often, while edits to product pages rarely moved it. What mattered more was what the page was: a page built to explain one thing for one market.
The contrast: a page that did not move
Lone Star had earlier tried the same location-hub approach for membership questions, such as what you need to join a credit union in Greenville. Across its four questions, it did not lift visibility overall. The reason is useful: for most membership questions, ChatGPT gave a generic checklist (bring an ID, make a small deposit) and named no credit union at all. There was no list for Lone Star to join.
Car loan questions are different. ChatGPT answers them by naming lenders, so a page that hands it the right facts has a list to join. In Lone Star’s membership test, a better page did not change answers that named no credit union. Before building a page, it is worth checking whether the answers to your question name any institutions at all.
What this suggests for other credit unions and banks
- Compare your membership area or market with your branch map. Counties where people can join or bank with you, but where you have no branch, are where AI assistants are most likely to leave you out.
- Write the facts the answer is made of. Lender name, product, place and eligibility, in plain text that a crawler can read, not in images or scripts.
- Keep it to one product. Lone Star’s page is about auto loans only, which keeps every section on topic for the question being asked.
- Link to it from the pages AI already reads, such as your product page and branch pages.
- Keep rates on the rates page and link to it, rather than repeating numbers that change.
It is one credit union, one product and one AI assistant, so treat it as a well-measured example rather than a rule. It is consistent with what MetriFi has seen across more than 90 GEO experiments, summarized in What 84 AI visibility experiments taught us about credit union websites: pages that clearly answer a specific, local question tend to get cited.
Results were measured in MetriFi GEO using Lone Star Credit Union’s tracked campaign data and are published with the credit union’s permission. MetriFi GEO measures how often financial institutions are named in AI-generated answers and runs experiments to test what changes that. To see how your credit union shows up today, talk with us.