# Results-Season AI Claims Decoder: a free prompt pack and scorecard

From the Novelty shelf at ReadSigil (https://readsigil.com/novelty.html)
Free. Available until Sunday 1 November 2026, 11:59 PM IST.

## What this is

Listed banks, NBFCs and fintechs in India are reporting their
July–September quarter results, and investor presentations and
earnings-call transcripts are full of lines about AI, GenAI and
"digital transformation". This pack helps you read those lines with a
cool head: separate a measured outcome from a press-release adjective.

It has six prompts and a ten-question scorecard. It is not about any
particular company. Use it on any public document you choose.

## Ground rules

1. Use only public documents (published investor presentations, press
   releases, official transcripts). Do not paste anything confidential,
   including your own employer's unpublished material.
2. AI models misread tables and invent numbers. Check every figure the
   model quotes against the source document, page by page.
3. This is a reading aid, not investment advice, a rating or a
   recommendation about any security.

---

## 1. Pull out every AI claim, verbatim

Below is text from a public results document. List every sentence that
mentions AI, machine learning, GenAI, automation, analytics or "digital".
Quote each one exactly, with its page or slide number. Do not summarise
or paraphrase. If there are none, say so.

[paste text]

## 2. Classify each claim

For each quoted claim, label it as one of:
(a) measured outcome with a number and a baseline,
(b) measured outcome with a number but no baseline or time period,
(c) activity or deployment with no outcome ("we launched", "we are piloting"),
(d) intention or aspiration ("we will", "we aim to"),
(e) adjective only ("AI-first", "cutting-edge").
Give a one-line reason for each label. Do not upgrade a claim.

## 3. What would make this claim checkable?

For each claim labelled (b) to (e), write the single follow-up question
an analyst could ask on the call to make it checkable. Examples of good
questions name a metric, a baseline, a time period and a scope (which
product, which customer segment).

## 4. Quarter-on-quarter consistency

Here are the AI-related claims from this quarter and the previous
quarter's document from the same company. Which earlier claims or
targets are not mentioned this time? Which numbers changed definition?
Which pilots from last quarter are now described as live, and which have
gone quiet? Quote both versions side by side.

[paste both sets of claims]

## 5. Cost side of the story

Does the document say anything about technology spend, cost-to-income,
headcount changes or vendor costs alongside the AI claims? Quote what is
there. Then list what is missing that you would need to judge whether
the AI work is paying for itself. Do not estimate missing numbers.

## 6. Risk and governance mentions

List any mentions of model risk, data privacy, fairness, customer
complaints, outages, or regulatory guidance in connection with AI or
digital channels. If the document makes big AI claims but says nothing
on risk or governance, say that plainly as an observation, not a verdict.

---

## The ten-question scorecard

Score one point for each "yes". Use it to compare documents, not to
grade companies.

1. Is at least one AI claim backed by a number?
2. Does that number have a baseline (before vs after)?
3. Is the time period stated?
4. Is the scope stated (which product, segment or process)?
5. Is a revenue, cost or risk outcome named, not just usage?
6. Are last quarter's AI targets followed up?
7. Is technology or AI spend disclosed in any form?
8. Is there any mention of model risk or governance?
9. Is a customer outcome (complaints, turnaround time, approval rate) named?
10. Could an outsider verify any claim from public data?

0–3: mostly narrative. 4–6: some evidence. 7–10: unusually checkable.

---

## Why it expires

These prompts are most useful while results documents are fresh and
calls are still happening. After 1 November the pack comes down; the
scorecard idea will still be in your head, which is the point.