The starting point
A regulated fintech of about 120 people, with real press coverage and a solid product. Ask a model about them and you got hedging: "appears to be", "may offer", "limited information is available". None of the five models we tested named them unprompted when asked about their category.
Two things were missing. There was no reference page a model could treat as neutral fact. And their review profile was thin: a 3.7 on Trustpilot from about 1,100 reviews, out of hundreds of thousands of customers, most of them from people with a problem.
For a financial company, both gaps hurt more than usual. People don't trust their money to a company the machine isn't sure exists.
What we did
The notability assessment
Before writing a word, we catalogued every independent piece written about them and checked it against Wikipedia's standards. The question was simple: had independent, reliable publications written about them at length, more than once?
They cleared the bar, just. Several national and trade pieces covered the company directly, not only its funding. If they hadn't qualified, we would have told them so and stopped there.
The page
We drafted a neutral, sourced entry where every claim pointed to independent coverage, declared the conflict of interest, and submitted it through Articles for Creation. The client reviewed the draft for accuracy, not tone. We also aligned the structured Wikidata record behind it: founding date, leadership, identifiers.
The reviews
At the same time, we wired a review request into onboarding, at the moment customers finished setup and saw their first result. One ask, no incentive, sent to everyone. Every review got a public reply, including the critical ones.
What changed
- Models naming them went from none of five to four of five. The hedging language disappeared from answers about who they are.
- Trustpilot rose from 3.7 to 4.6 in five months, on a far larger and more recent base of reviews.
The two pieces of work reinforced each other. The page gave models a fact source to state who the company is; the reviews gave them a recent verdict on whether it's any good.
By the numbers
| Measure | Day one | Month 3 | Month 5 |
|---|---|---|---|
| Models naming them (out of 5) | 0 | 2 | 4 |
| Answers using hedging language | 83% | 41% | 9% |
| Trustpilot rating | 3.7 | 4.3 | 4.6 |
| Trustpilot reviews, total | 1,140 | 9,800 | 21,600 |
| Reviews in the last 30 days | 40 | 5,200 | 6,400 |
| Reviews answered publicly | 0% | 100% | 100% |
The Wikipedia project
- 14 independent sources made the cut for the article, out of 52 mentions we catalogued. The rest were press releases, funding coverage or passing mentions.
- Week 6: draft submitted through Articles for Creation, with the conflict of interest declared.
- Week 11: the article was accepted and published.
- Week 12: the Wikidata record was completed, with founding date, leadership, headquarters and identifiers.
Within a month of the page going live, three of the five models were using its wording to describe the company.
The review programme
One email, sent the day a customer completed onboarding and saw their first payment go through. 22% of customers who received it left a review. No incentive, no pre-screening survey. Every review got a reply within two working days.
In their words
"Investors and enterprise prospects check the same two things: who you are, and whether people like you. We finally had good answers to both."
Marketing Director, Regulated fintech
What made it work
Patience with the encyclopedia's process, and the discipline to ask every customer rather than only the happy ones. The rating rose because the honest majority finally got asked, not because anyone was filtered out.