Capture Screenshots of Multiple URLs With an AI Agent

Updated 8 November 2026 · Standard formula, shown below

An AI agent can visit a list of web addresses, wait for each page to load, and save a screenshot for review or documentation. You’ll need a browser-based agent that supports navigation and screenshots, plus a clear URL list and a folder where you can store the results.

What You’ll Need

An AI agent with browser control, including the ability to open URLs and capture screenshots. Check that it can save image files or return them for download.
A prepared list of complete URLs, including https:// where appropriate. Remove duplicates and confirm that each address is intended for capture.
A destination folder or other approved storage location with enough space for the images.
Permission to access the pages and capture their contents. Do not ask an agent to bypass logins, access controls, or other restrictions.
Optional capture settings: viewport width and height, full-page versus viewport screenshots, file format, and a naming convention.

Step-by-Step

Decide what each screenshot should show. Choose whether you need the visible viewport or the entire scrollable page. A viewport screenshot is quicker and more consistent; a full-page image includes content below the fold but may be very tall and can render some dynamic pages differently. Set a standard viewport size if you plan to compare pages.
Prepare and check the URLs. Put one address per line in a text file or spreadsheet. Include the full scheme and hostname, and check for typos, duplicates, and unnecessary tracking parameters. If a URL needs a particular query string to display the right page, preserve it exactly. Avoid putting passwords, access tokens, or private information in the list.
Set up the agent’s browser session. Open the agent’s browser tool and choose the destination folder or download location if the tool offers that setting. Set the viewport, screenshot type, and image format. PNG is useful for crisp text; JPEG can produce smaller files. If the agent uses a fresh session for each run, note that pages requiring sign-in may not be available.
Give the agent a constrained task. Ask it to process only the supplied URLs, one at a time. Specify the wait condition, screenshot type, naming pattern, and what to do if a page fails. For example: “Visit each URL in the list in order. Wait until the page’s main content is visible and network activity has settled, up to 15 seconds. Capture a full-page PNG named 001-example-com.png. Do not click links, submit forms, or enter credentials. Record the URL and outcome for each item.” Adjust the timeout to suit the sites; a fixed short pause may not be enough for slow pages.
Run a small test first. Start with one or two representative pages. Confirm that the agent opened the correct address, waited for the intended content, captured the right portion of the page, and saved files with usable names. This catches configuration problems before a longer batch runs.
Run the complete list in a controlled batch. Ask the agent to continue through the remaining URLs sequentially or in small batches. Sequential processing is generally easier to inspect and is less likely to overload sites. If the tool supports parallel tabs, use them cautiously: too many concurrent visits can cause timeouts, rate limits, or inconsistent rendering.
Have the agent report results. Request a compact log with each URL, screenshot filename, and status such as captured, timed out, redirected, or blocked. Ask it to report failures rather than silently skipping them. If screenshots are returned as attachments, save them to the agreed folder and preserve the mapping between each file and its URL.
Verify the files yourself. Open a sample from the beginning, middle, and end of the list, and check the dimensions, content, and filenames. Confirm that the expected page—not a consent screen, error page, or blank load—appears. Review every image when completeness or accuracy matters. Re-run only the failed or incorrect URLs after fixing the cause.
Keep the output organized. Use a separate folder for each capture run and retain the URL list and result log alongside the screenshots. If you repeat the task, include a date or run identifier in the folder name. Handle captured personal or confidential information according to your organization’s storage and retention rules.

Common Mistakes

Using vague instructions. “Screenshot these sites” does not say which URLs, how long to wait, or whether to capture the whole page. Define those choices before starting.
Assuming a page has loaded because navigation finished. A page can display its shell before images, fonts, or data appear. Wait for a visible landmark or a suitable page-ready condition, with a timeout.
Using ambiguous filenames. Names such as screenshot.png are easily overwritten. Use an index and a sanitized hostname or short page label, and check that the agent does not reuse names.
Ignoring redirects and access screens. A URL may redirect to a different page, sign-in screen, consent prompt, or error. Capture and log the actual outcome instead of treating every image as a successful page capture.
Running too many pages at once. Excessive parallel activity can slow the browser and trigger site defenses. Use small batches and respect site access rules.
Expecting identical images every time. Ads, timestamps, rotating content, personalization, animations, and responsive layouts can change between visits. Record the viewport and run conditions if comparisons matter.

Tips

Use a numbered filename pattern, such as 001-example-com.png, and keep the original URL list in the same order.
Tell the agent to wait for a specific visible element when you know what marks a page as ready. Use a reasonable maximum wait so one stalled page does not stop the whole run.
For consistent comparisons, keep the viewport, browser zoom, locale, and session state the same. Note that different sessions may show different personalization or consent banners.
Ask for a failure summary and a retry list. Retrying only failures makes the run easier to audit and avoids unnecessary repeat visits.
For long pages, consider whether full-page capture is readable. Splitting a page into sections may be more useful than one extremely tall image if the tool supports it.
Keep the agent’s actions limited to navigation and capture unless other actions are explicitly needed and authorized. Avoid asking it to interact with forms or change account settings.

Troubleshooting

No screenshots were saved: Check whether the agent can create files or only display images in its conversation. Look for a download or export option, confirm the destination is writable, and ask it to return the images as attachments if supported.

The images show blank pages or loading indicators: Increase the wait limit, or ask the agent to wait for a visible page heading or main-content element. Retry a single URL first. If it still fails, the page may rely on scripts or resources the browser session cannot load.

The agent captured the wrong page: Compare the final address with the requested URL and inspect any redirect. Provide the exact URL and ask the agent to report the final address before capture. Do not treat a redirect as an error automatically; verify that it leads to the intended content.

A site asks for sign-in or blocks access: Stop and use only an authorized, approved session. Do not instruct the agent to evade access controls or solve challenges on your behalf. If access is legitimate, follow the site’s normal sign-in process yourself and check whether the agent is allowed to use that session.

Files overwrite one another or are hard to match: Add a unique index to every filename and request a URL-to-filename log. If a hostname contains unusual characters, use a simple sanitized label rather than relying on the complete URL as a filename.

Some pages time out in a large run: Reduce the batch size, process pages sequentially, and retry the affected URLs individually with a longer wait. A timeout can also indicate a temporary network issue or a page that never reaches the requested ready state.

Frequently Asked Questions

Can an AI agent capture screenshots of many URLs automatically?
Yes, if its browser tools support navigation and image capture. It should process a supplied list and report which URLs succeeded or failed.

Should I use full-page or viewport screenshots?
Use viewport captures for a consistent view of what appears on screen. Use full-page captures when below-the-fold content matters, and check that the resulting image remains readable.

How long should the agent wait before taking each screenshot?
There is no universal delay. Prefer waiting for a visible content element, with a maximum timeout such as 15 seconds that you can adjust for slower pages.

Can it capture pages that require a login?
Only through an authorized browser session and where the site and your organization permit it. Do not use the agent to bypass authentication or access restrictions.

What is the best image format?
PNG is a good default when text clarity matters. JPEG may reduce file size, but can soften small text and interface details.

How can I make screenshots comparable across runs?
Keep the viewport, browser zoom, page state, and capture method consistent. Also note that dynamic content, personalization, and page updates can still cause differences.

Method: the standard formula for this calculation, rounded as each result states. See how the calculators work or report an error.