An instant answer still has a date on it.

An AI-generated credit answer can arrive in seconds while resting on evidence that is months old. Speed of delivery, age of evidence and current accuracy are three separate things. A fluent answer reads as present tense, which makes its sources and its timestamps more important, not less.

In August a UK bureau made an authenticated credit score available inside a general AI assistant: live score, history and the factors behind it, in conversation. This was a consumer product, not a commercial-credit tool. The product is not the point. Where credit answers are heading is — into assistants, phrased as sentences, arriving instantly.

A better interface can hide older evidence

Take as read that data goes stale. Grand has argued that a credit limit has a vintage and the customer underneath it keeps changing, and that argument is settled.

The newer problem is presentation. A credit report looks like a document. It carries a run date, a filings section with dates attached, and blank space where a source has nothing to say. Its limits are visible on the page.

A sentence in a chat window has none of that. It reads as a statement about now. The confidence of the phrasing comes from the language model; the age of the evidence comes from the data, and the two are no longer displayed together. An answer can be perfectly accurate about a situation that ended in March.

None of this is an argument against AI in credit work. Grand's own position is that AI's job is to explain what changed, not to replace the judgement. This is the part that argument leaves open: an explanation can be fluent, well reasoned and still built on evidence nobody has checked the date of.

The three dates behind every credit answer

Every credit answer has three dates, and they are rarely the same.

• The event date. When the thing happened: the director resigned, the judgment was entered, the year end closed.
• The source date. When it was recorded or published. A private company's annual accounts are due nine months after its accounting reference date, so a filed set can describe a period that finished well over a year ago. A judgment reaches the register some time after it is entered.
• The answer date. When you were told. This is the one an instant answer makes obvious, and it is the least informative of the three.

The gap between the event and its publication is a visibility gap. The gap between publication and its appearance in an answer is a delivery gap. Neither can be judged from the speed at which the final sentence appears.

Keep market evidence and company evidence apart

One caution, because the two get blurred. Industry series run late: the Finance and Leasing Association's most recent monthly statistics still cover May, published on 17 July, which is the sort of gap the weekly digest keeps track of.

That says something about how well anyone can read the market in September. It says nothing about whether a particular customer is in trouble today. A company-level answer is old when the evidence about that company is old, not when a national dataset is late. Reading one as the other produces false comfort in both directions.

How to tell whether an AI credit answer is current

Ask an answer the three questions a credit manager already asks of a file. What changed? What evidence supports it? When was that evidence true? An answer that shows its sources and dates may be suitable for a decision. One that cannot is useful for orientation, but it is not decision-grade evidence. Closing the gap between the second date and the third is what continuous monitoring is for.

Fluency is not freshness. The better the writing gets, the more deliberately you have to check.

An instant answer to an old question is still an old answer.

See how Grand helps at heygrand.com.