Screenshots age faster than the words around them. A menu changes, a button moves, and the article still looks correct to the writer who last edited it. This audit finds the misleading images before an agent follows one in front of a customer. It is written for the knowledge owner, the agent who uses the article, and the reviewer who approves the update.
Why screenshots drift
A screenshot is a dated claim about a screen that someone else controls. The product team can rename a field, move a tab, or add a confirmation step without touching the knowledge base. The surrounding paragraph may stay true while the image quietly becomes wrong. Agents trust pictures because pictures feel like evidence, so a stale image can defeat a correct sentence.
The drift is rarely dramatic. A gray button becomes blue, a label changes from Save to Apply, or a menu item moves one level deeper. Those small differences matter during a live contact because the agent is matching the screen in front of them to the screen in the article. When the match fails, the agent stops reading and starts guessing, and the customer waits through the search.
Screenshots also carry hidden assumptions about language, screen size, and permissions. An administrator sees a different menu than a frontline agent. A translated interface may use different labels even when the workflow is identical. An audit that checks only whether the image loads will miss the more common failure, which is an image that renders perfectly but describes the wrong role or locale.
Build a screenshot inventory
Start by listing every article that contains an image of an internal tool. Record the article slug, the image path, the date the image was added, and the person who added it. Do not rely on the image file name alone, because names are often copied and reused. The inventory is the map for the rest of the audit.
Separate customer-facing images from internal-tool images. A screenshot of a help center page ages when the help center changes, while a screenshot of an order system ages when the order system changes. The two review owners are usually different people, so mixing them in one list makes follow-up slow. Keep the internal-tool list with the support operations owner and the customer-facing list with the content owner.
Add the last review date and the next review trigger to each row. A trigger can be a product release, a permission change, a translation update, or a tool migration. The trigger is more useful than a fixed annual date because it points to the event that actually invalidates the image. A dated inventory also shows which articles have never been checked.
Check the labels in the image
Open the live tool in the same role and locale that the article targets. Compare the field labels, button text, and menu path with the words shown in the screenshot. Read the labels aloud rather than glancing at the layout, because the eye forgives small differences that a reader will notice. Mark every mismatch, even when the workflow still works.
Pay attention to confirmation dialogs and error states. These screens change often because teams add warnings after incidents. A screenshot that shows the old confirmation may teach an agent to skip a step that now requires a reason code. That kind of omission is more dangerous than a moved button because it can create a compliance problem.
Check what the customer sees as well. If the article tells an agent to ask the customer to click a link, the customer-facing screenshot should match the current page. Review the mobile view separately, since menus collapse and labels shorten on small screens. An image that is correct on a desktop can be misleading on a phone.
Test the steps against the live screen
Follow the article from the first step to the last while the live tool is open. Do not click through from memory, and do not assume that the screenshot order matches the current flow. Count the steps the article describes and compare that count with the steps the tool actually requires. A missing step is easier to find by doing the work than by reading the page.
Use a test account or a sandbox whenever one exists. If the tool has no safe test path, walk through the steps with a supervisor who can stop before any real change is saved. Record the exact screen where the article diverges, and capture a fresh image of that screen if policy allows. The evidence should be stored with the article record, not in a personal folder.
Ask one agent who did not write the article to follow it without help. That agent will hesitate at the same places a new hire would hesitate, and the hesitation often points to a missing or outdated image. Note the question they ask and the step where they ask it. Those notes become the edit list for the next section.
Replace, annotate, or retire
Each screenshot needs one decision. Replace means a new image is captured and the old one is removed. Annotate means the image is still useful but needs a clear note about the current behavior. Retire means the image should be deleted because the words are enough or the screen no longer exists. Write the decision beside the inventory row so the choice can be reviewed later.
When replacing an image, capture the full context the agent needs, including the page title and the relevant field labels. Crop out customer data and internal identifiers before the image enters the knowledge base. Compress the file so the page still loads quickly on a slow connection. Add the capture date to the image record even if the date is not visible to readers.
When annotating, put the note next to the image rather than in a distant paragraph. The reader should see the correction at the moment they look at the picture. If the difference is large enough to need a paragraph, the image probably needs replacement instead. Never keep two screenshots of the same screen without a clear label for which one is current.
Set the next review trigger
Attach the review to the event that changes the screen. A product release note, a permission change, or a tool migration should create a task for the knowledge owner. That task should point to the inventory row and the specific image, not to the whole library. Specific tasks get done, while general reminders get postponed.
Review the inventory after the first month and count how many images changed. That count shows whether the triggers are catching the right events. If images keep drifting between reviews, add the missing trigger or shorten the cycle. The record should show the date, the reviewer, and the decision for each image.
Close the audit with a short note to the agents who use the article. Tell them what changed and which screenshots are now current. A quiet update that nobody notices leaves the old habit in place. The knowledge base update rule is a useful companion when the same question keeps returning.
