Corporate credit risk analysis has always been a document-heavy, judgment-intensive discipline. A single mid-market borrower can arrive with three years of audited financials, management accounts, a covenant compliance certificate, an internal rating model output, and a credit memo template that runs to fifteen pages. Analysts spend the majority of their time extracting figures, reconciling definitions across statements, and reformatting commentary — work that is necessary but consumes the hours that should go into actual risk judgment. In most credit teams, the bottleneck is not analytical capability; it is the mechanical preparation that precedes it.
The friction compounds at portfolio level. Quarterly reviews, watchlist updates, and annual renewals arrive in waves, each requiring consistent treatment across dozens of obligors. Inconsistency creeps in: one analyst weights leverage differently from another, covenant headroom is calculated on slightly different bases, and the reasoning behind a rating downgrade is buried in prose that the next reviewer must reconstruct from scratch. Regulators and internal audit increasingly ask for traceable, comparable rationale — and most teams cannot produce it quickly.
ChatGPT, used with disciplined prompting, addresses the preparation layer directly. It can read a financial statements package, extract and normalize the key figures, compute ratios against a defined methodology, flag covenant proximity, and draft structured commentary in your house style. It does not replace the credit officer’s judgment on character, industry dynamics, or collateral quality — but it removes the hours of extraction and formatting so that judgment gets the time it deserves. The key is structure: the prompts below are built to force the model to read your standards first, ask clarifying questions, and produce a plan before generating anything.
Where AI fits and where it does not
Be precise about scope. ChatGPT is reliable for extraction, normalization, ratio computation, drafting, and consistency checks against a written policy. It is not reliable as a source of external facts (it does not know a borrower’s payment behavior), and it should never be the final word on a rating decision. Treat every output as a first draft prepared by a competent junior analyst who has read the file but has no authority to approve anything.
The two prompts below cover the two highest-value workflows: a full credit assessment of a single obligor, and a portfolio-level consistency review across a batch of borrowers. Both follow the same discipline — read the files, confirm understanding, plan, then execute.
First, read these files completely before responding:
[credit_policy.md] — our internal credit policy: rating scale definitions, ratio thresholds, covenant standards, escalation triggers
[financials_borrower.md] — three years of audited financial statements plus latest management accounts, converted to markdown tables
[credit_memo_template.md] — our standard memo structure with section headings and required content per section
Here is a reference for what I want to achieve:
[Upload a previously approved credit memo for a comparable obligor as markdown]
Here’s what makes this reference work:
It opens with a one-paragraph recommendation and rating, then supports it with normalized financials, ratio trends over three years, covenant headroom analysis, and a risks-and-mitigants section that names specific triggers rather than generic industry commentary. Tone is factual, no adjectives without a number behind them. Every ratio is defined once and used consistently.
Here’s what I need for my version / SUCCESS BRIEF:
Type of output + length: Structured credit memo draft, 1,200-1,500 words, following our template headings exactly
Recipient’s reaction: The credit officer should think “the numbers are already reconciled and the risks are named specifically — I can focus on judgment calls”
Does NOT sound like: Generic bank boilerplate, marketing language, or hedged commentary without figures
Success means: Every ratio in the memo is traceable to a line in the financials file, and no section of the template is left empty
My context file contains my standards, constraints, audience. Read it fully before starting.
DO NOT start executing yet. Ask clarifying questions first.
Give me your execution plan (5 steps max) before you begin.
Notice what the first prompt does not ask for: it does not ask ChatGPT to assign a rating. It asks for a draft that makes the rating decision easier. Keeping the model on the preparation side of the line is what makes the output usable in a regulated workflow — every figure must be traceable, and the credit officer remains the decision-maker of record.
Portfolio-level consistency
The second workflow is where AI delivers disproportionate value relative to effort. Reviewing twenty borrowers one at a time invites drift; reviewing them against a single written standard, applied by the same reader in the same session, exposes inconsistencies that manual review misses. This is also the workflow that answers the audit question: show us that your watchlist treatment is consistent across obligors.
First, read these files completely before responding:
[rating_methodology.md] — our rating model: factor definitions, weights, thresholds, and override rules
[portfolio_data.md] — current ratings, key ratios, covenant status, and last review date for each obligor in the portfolio
[watchlist_criteria.md] — the written criteria for watchlist placement and escalation
Here is a reference for what I want to achieve:
[Upload a prior internal audit finding or consistency review as markdown]
Here’s what makes this reference work:
It lists each inconsistency as a discrete finding with the obligor name, the metric in question, the treatment applied, the treatment the methodology requires, and the recommended correction. It separates genuine methodology breaches from acceptable overrides and states the rationale for each override.
Here’s what I need for my version / SUCCESS BRIEF:
Type of output + length: Findings table plus narrative summary, 800-1,000 words
Recipient’s reaction: The risk committee should see immediately which obligors need re-review and why, with no ambiguity about the standard applied
Does NOT sound like: A data dump, or findings without a stated required treatment
Success means: Every obligor in the portfolio is either confirmed consistent or appears in the findings table with a specific correction
My context file contains my standards, constraints, audience. Read it fully before starting.
DO NOT start executing yet. Ask clarifying questions first.
Give me your execution plan (5 steps max) before you begin.
A practical tip before you start: build the markdown versions of your policy, methodology, and templates once, and reuse them across every prompt. The quality of the output tracks the quality of those reference files far more than the wording of the request. Start with a single obligor you know well, compare the draft against the approved memo, and refine your reference files until the gap is small. Then move to the portfolio review — and bring the findings table to your next risk committee meeting.
Published on 25 October 2026 on growwithgpt.com