---
title: "Step card 4: Ground it"
lang: en
author: "Jeff Bishop, BITS"
description: "A step card for the AHG 2026 Accessibility Agents workshop."
---

# Step card 4: Ground it

Block 4 of 7, 1:15 to 2:10.

## Where we are

Step 4 of 7: your agent, grounded in real evidence.

## What you will have at the end

The same agent, giving cited answers from real evidence, and a before and after you can show a colleague.

## Steps

1. Open Copilot Chat: Control+Alt+I on Windows, or Command+Control+I on a Mac.
2. Type /ground-my-agent and press Enter.
3. Use the same file as this morning.
4. Your agent reads that file's checker report from sample-course/evidence: GLOW's audit for the documents, axe-core for the web page.
5. Read the new answer, then Copilot's before and after.
6. Optional: open the syllabus in Word, choose Review, then Check Accessibility, and paste the results into Copilot Chat with your agent. Same skill, different checker.
7. Optional: the course announcement is online at letitglow.app/ahg/site. Run Accessibility Insights FastPass on it and compare with its checker report.

## You are on track if

Your agent's new answer cites at least one WCAG 2.2 criterion with a link, and says what a person still needs to check.

## If you are stuck

Read the before and after in examples/maria-alternate-format-planner.md, section 4, and use a ready-made agent with /ground-my-agent. It works the same way.

Raise a hand at any time. A helper comes to you.

## Your win

Your agent stopped guessing. That is what grounded means, and it is the trusted in the session title.

## On Monday

Before you ask any AI about a document, run a checker and give it the results: Word's, PowerPoint's or Acrobat's own checker, Accessibility Insights, or axe.