How do I check what my site looked like before a change?

The HTML as it was served at that moment, which is the part that is reliably preserved.

Many of the referenced assets, captured separately and at possibly different times.

Not everything: scripts frequently did not run, third-party content frequently was not captured, and anything requiring interaction is absent.

Captures are irregular: a busy site may have several a day and a quiet one a few a year.

Which means a capture answers what was served, and approximately when, rather than what a page looked like.

Recovering the text of a page that was deleted without a backup, which is the most common reason anyone looks.

Establishing when a change happened, by finding the last capture with the old state and the first with the new.

Listing URLs that existed before a migration, which is how a redirect map is reconstructed after the fact.

Confirming whether a page carried a particular declaration — a canonical tag, a noindex, a structured data block — at a point in time.

Seeing whether a site was substantially different under a previous owner, which matters when a domain was bought rather than registered.

Rendering is approximate: missing stylesheets and unexecuted scripts make the capture look worse than the original did.

Dates are the capture date, not the publication date — a page captured in March may have been published years earlier.

Absence of a capture proves nothing: the archive simply did not visit, which is common for deep pages.

A page excluded from crawling at the time may have no captures at all despite having existed.

So an archive is evidence of what was served when it was visited, and silence in it is not evidence of anything.

Backups. An archive is a third party's incidental record, not a recovery mechanism you control.

Version control for content, which is what a content system's revision history is for.

A redirect map made at migration time, which is far cheaper than reconstructing one afterwards.

Analytics history, which an archive holds nothing about.

Which is why the honest advice after using an archive once is to stop needing it.

Use the archive's documented public interfaces rather than scraping its pages.

Expect rate limits, and collapse listings rather than requesting every capture of a busy site.

Never present an archived capture as the current state of a site, which is the misreading that causes real confusion.

Remember the archive is a third party with its own policies, including removal requests, so a capture that existed may stop existing.

And never use it to reach content a site currently protects, which is the one use it should not be put to.

List the captures for the exact URL first, then for the host with a wildcard, since a page that moved is often captured under its old address only.

Collapse the listing to one capture per day on a busy site, otherwise the list is thousands of near-identical rows.

To date a change, narrow the window by halves: look at a capture in the middle of the range, then in the middle of whichever half still contains the change.

Read the HTML source of a capture rather than its rendering when the question is about a tag, since the source is the reliable part.