| You get | A proposed link review sheet with evidence and unresolved checks |
|---|---|
| Stays with you | Permission to inspect destinations and approval before sending |
| You need | Public links you own or are permitted to inspect, and an intended action for each |
| Rough cost | Start manually with a few links and existing AI access; browser or model usage may add charges |
| Skip it if | You have one familiar link that you can open and check yourself |
1What it looks like handed off
Before sending a resource email, you have a sheet recording each requested URL, the observed destination, what the reader should be able to do and what was actually checked. A page that opens is separate from a form that is visible. A visible form is separate from a message that was delivered.
In our own October20 website checks, four links used a section name that did not exist on the destination page. Earlier automated checks had passed without establishing that the inquiry form was visible. We corrected the links and checked the actual blank form on desktop and a narrow viewport. No inquiry was submitted. This is our site-maintenance example, not a client result or a claim that AI checked every link.
2The job, end to end
- List each public link and the action its reader expects, such as opening a worksheet or finding a blank contact form.
- Ask for a separate observation and evidence field for each check. Keep untested fields blank.
- Open the destination through a permitted browser. Check redirects and the actual section, rather than the address alone.
- Wait for scrolling or navigation to settle. Check that the intended action is visible and usable at the screen sizes you support.
- Review downloads as files, not just links. Confirm their names, formats and expected contents without running unfamiliar files.
- Approve the corrected link, retain failed observations and recheck the changed path before sending.
Try the fictional practice sheet
These three cases are made up and use documentation-only example domains. Do not browse them as a live test. Read the supplied observations, prepare your decisions, then compare with the separately authored key. This tests whether your review preserves evidence gaps; it does not measure browser or model performance.
Using only the supplied fictional observations, prepare one review row per case. Distinguish page access, intended action and completed delivery. Preserve unknowns. Do not infer a section exists from an HTTP status or URL. Do not browse, sign in, submit forms, send messages or approve the links. Leave approval blank for me.
For your real list, replace the fictional observations with dated evidence from permitted checks. Do not paste signed links, password resets, customer records or authentication tokens into a public sheet or AI prompt.
3What stays with you
You choose the intended destination and decide whether the agent may inspect it. Respect the destination's access rules. Stop at sign-in, a permission barrier or an automation restriction; use an authorized manual check when appropriate and record the limit. The review sheet is not permission to submit a form or send a message.
4What happens when it is wrong
Keep the failing URL and observation. Correct your own link or ask the page owner about the destination. A section name can change while the page keeps opening normally. If the page scrolls after loading, check its settled state before deciding the target failed. Recheck the actual reader action after a correction.
5What it costs
Try three public links before automating a whole library. Count preparation, browser checks, corrections and final review, along with any tool charges. Compare that total with checking the same links manually. No time saving is established by this practice exercise.
6When not to bother
For a single familiar link, opening it yourself is usually enough. Avoid an automated audit when site rules prohibit it or when the only available evidence is a successful page response. If you need proof that a real submission reaches someone, arrange a separately authorized delivery test; this review does not provide it.
Updated : added related guides. Cost estimates and source-check dates retain their stated dates.
7Want this running on your own computer?
I'll set it up, and keep it running if you like. On a Mac mini, a Windows PC or Omarchy, with the agent you use or want: OpenClaw, Hermes, Claude Code, Codex or another. Approval settings depend on the agent, tools and configuration. Tell me about the job and I'll quote it. If a $20 subscription does it, I'll tell you. You keep the computer, the accounts and the setup.
Tyler Mayberry, who runs Cirlet (formerly Animas)
For independent insurance agencies nationwide, Cirlet offers AI and automation help. The first 30-minute conversation is free. Paid work and ongoing support are scoped and priced before they start.