Can AI Rebuild a Lost Website from the Wayback Machine?

Yes, AI can rebuild a lost website from the Wayback Machine if the old details are checked against what's true today, because the archive usually keeps the website's pages. An AI coding agent reads a small business site's old pages in the Internet Archive, builds a fresh site and publishes it. In September 2026, IvI SEO rebuilt five small business websites from Wayback Machine captures with Codex, one prompt each, and every first build was online within 16 minutes.

Lose your website and it's natural to think you're back to a blank page. Often, you're not. Every site I brought back was a small UK business site: a cattery, a dry cleaner, a carpet cleaner, a restaurant and lounge, and a restaurant and bar.

What the Wayback Machine is

The Wayback Machine is the Internet Archive's collection of past versions of web pages: you type in the old web address and pick a date it was captured.

What came back from the archive

• Between the five sites, page text, services, phone numbers, emails, opening hours, menus and FAQs came back.
• A dry cleaner's price-list PDF came back too, and its 47 starting prices were turned into a web page.
• Original photos came back on three of the five.

What the AI made fresh

Codex, the AI coding agent I used, gave each site a new design that kept the old site's content and character. On the last two, my prompt also asked it to look at similar sites for ideas, and it did.

Where it went live

Codex published every rebuild itself, onto Cloudflare, the web host the new sites run on.

If the old host or developer still has a backup, start there, because that's the complete copy. The archive is for when nobody has one.

How long does it take AI to rebuild a website from the Wayback Machine?

It takes AI minutes to rebuild a website from the Wayback Machine when the site is small. The AI builds the site and puts it online itself. Pointing the business's own domain at the new site is a short job for a person, and the AI stops to ask for it. IvI SEO's nine-page restaurant rebuild had a preview online in under 14 minutes and was live on its own domain about 20 minutes after the prompt.

I timed each one from the Codex logs, from my prompt to Codex's first reply saying the site was deployed or live. Domain set-up and anything I asked for afterwards aren't counted. A preview is a temporary web address for checking the site before the real domain points at it.

Here's what the times leave out. On the restaurant and bar, Codex asked for the nameservers to be changed at the domain registrar (nameservers tell the internet where a website lives), and I did that about 14 minutes in. On the carpet cleaner, I signed in to Cloudflare once so Codex could publish, and the HTTPS certificate took a while to switch on after the domain moved. On the carpet cleaner, dry cleaner and cattery, I asked for forced HTTPS and one version of the address afterwards, because my prompt hadn't. The dry cleaner's domain was connected later. The biggest addition was the two redesigns I asked for later, which come with the brief further down.

Codex did the graft. I did the bits that needed my logins, plus a few follow-ups. And a page count isn't a timetable: the nine-page site beat the four-page cattery, and the one-page carpet cleaner took longer than the two-page dry cleaner. That's five small sites in September 2026. A big site or an online shop is another job, so no promises there.

What can't AI get back from the Wayback Machine?

AI can't get back from the Wayback Machine any photos or pages the archive never saved, or anything that ran behind the scenes, like forms, logins and databases. The Wayback Machine only saves the public pages it reached, so photos it missed or that were stored elsewhere aren't there, and neither is anything that ran on the server. Clearly labelled stock photos can fill the gaps only until the owner's own photos are back, and call and email buttons can take the place of forms. Two of five small business sites IvI SEO brought back from the archive had no original photos there, and one site's pages came from captures taken between 2021 and 2025.

• Photos: The restaurant and lounge had no photos in the archive, so stock photos labelled as illustrative stood in, and the restaurant and bar's photos didn't come back because its old WordPress image addresses hadn't survived. Real photos make a business look real, which is why I'd swap stock for photos of the actual work on a Google Business Profile too.
• Pages from different years: The restaurant and bar's homepage came from an April 2025 capture, its menu from August 2022, and its private hire, bar, catering and contact pages from 2021, so details from different years sat side by side and each one needed checking (more on that further down).
• Forms and bookings: A form's page can be archived, but whatever sent the message isn't. The rebuilds ended up with call and email buttons instead (the restaurant and lounge's first build had a form, swapped for call-only), so no form pretends to send a booking that never arrives.
• Anything behind the scenes: The database, logins and payments behind an old site are never in the archive, and WordPress gets its own question further down.

The archive is a photo of the shop window, taken from the street. It never got near the till or the back office.

What should you tell AI when it rebuilds a website from the Wayback Machine?

Tell AI the old domain and a Wayback Machine link from when the site worked so it can rebuild the website, and say what to keep and what has changed. What to keep usually means the text, photos and contact details, and what has changed might be a new phone number or the address the business trades from now. For a fresh design, ask the AI to look at similar sites for ideas and to steer clear of the generic look many AI-built sites have. Say where to publish the new site, and ask for HTTPS on every page and one version of the address, either with www or without. The AI then works through the archived pages itself, instead of someone copying each page by hand.

• The domain: The old web address tells the AI what to rebuild and where the new site will live.
• A Wayback Machine link: A capture from when the site worked is the AI's starting copy, so check it shows the real site, not an error page.
• What to keep: Asking the AI to keep the old text, photos and contact details makes it clear that a fresh look doesn't mean fresh words.
• What's changed since: Today's phone number, address or opening hours belong in the brief if you know them, so the AI isn't left trusting the old site.
• A fresh look: Asking the AI to research similar sites and avoid AI slop helps, but also tell it how the new site should differ from sites it has already made.
• Where to publish: Saying where the new site goes live, with forced HTTPS and one version of the address, saves asking for them afterwards.

The brief grew over the day. The first three sites got a short one: the domain, the Wayback link, "I want it to be very similar, but with a fresh look", the current footer address and "push it to cloudflare once done". By the last site, the brief IvI SEO typed into Codex had grown into this, with placeholders for the domain and link:

Can we create a new website: [your domain] pls look on the waybackmachine: [Wayback Machine link] I want it to be very similar, but with a fresh look [...] please take any details you can from the old website. including photos, contact details and location of HQ. I want you to then push it to cloudflare once done [...] Make sure its forced https and is the www. ver. Please do some research on similar websites so ours can be even better and brand new looking and not AI SLOP

Rough? Very. It still worked. A brief can ramble; it just has to be clear.

Even so, "not AI SLOP" wasn't enough on its own. Both builds with that line came back looking like the other sites I'd made that day. For the cattery, I wrote "looks very AI sloppy, can you make it much nicer, looks like the other sites we've built together today", and a redesign was live about 12 minutes later. For the restaurant and bar, it was "can you reinvent this page fully? this looks like the other sites we've made today, you are better" (yes, a pep talk for an AI), and the redesign was live about 23 minutes after I asked. What got each one a design with its own character was saying exactly what was wrong: it looked like the others.

Rebuild your own site, or one the owner has said yes to, because the old words and photos still belong to them. Adding "stop and ask me when you're unsure" wouldn't hurt either; mine didn't have it.

What should you check before an AI-rebuilt website goes live?

Check that every detail from the Wayback Machine is still true today before an AI-rebuilt website goes live, especially phone numbers, addresses, opening hours and prices. The archive shows what was true on the day each page was captured, which can be years ago, and it keeps the old site's own mistakes. Check the phone number, address and opening hours against the Google Business Profile and with the owner, then make the website and the profile agree. Date or remove old prices and offers, remove claims nobody can confirm, and check the new site's preview on a phone and a computer. On IvI SEO's rebuilds, the AI's checks kept a cattery's archived Wednesday hours of 8:30am to midnight off the site, but nothing in the archive showed that a restaurant's phone number had changed.

• Phone numbers, addresses and opening hours: Where a rebuilt site's phone numbers, addresses or opening hours disagree with the Google Business Profile, the owner's answer goes on both, because the profile can be out of date too. Wrong numbers and hours are among the first Google Business Profile mistakes to fix.
• Prices and offers: Old prices and offers from the archive need dating or removing, because a years-old price reads like today's. The restaurants' old prices were dated or left out, and the dry cleaner's were labelled to confirm.
• Unverified claims: Claims about insurance, star ratings or licences stay off a rebuilt site until the owner confirms them, which is why the rebuilds left them out.
• The preview on a phone and a computer: The preview deserves a proper look on a phone and a computer before the domain points at it, while mistakes still cost nothing.
• Old page addresses: Old page addresses are worth keeping where you can, so bookmarks and old links still land on the right page; the restaurant and bar rebuild kept its old menu page address.
• Sitemap and analytics: Every rebuild got a sitemap, the list of pages that helps search engines find them all, and analytics to count visits was added afterwards.

Now the honest bit. Before the first deploy, Codex caught a fair bit. The cattery's archived Wednesday hours read 8:30am to midnight. Unless the cats were having a lock-in, that was wrong, and Codex swapped it for a line asking people to call instead. It also left out the cattery's 2022 prices and a licence-rating claim. The restaurant and bar's archived hours contradicted each other, so its rebuild asked visitors to call.

Two things got past it and went live. The restaurant and lounge went live on its own domain with the archive's phone number, which was out of date. Asked whether everything was correct, Codex answered: "It matches the archived site, but I can't confirm everything is current." The current number replaced the old one about 5 minutes after launch. And the restaurant and bar's archived menu had lost its section breaks, so the first build showed meat dishes under vegetarian mains. That slip was live on the site's own domain for about half an hour, until Codex found and fixed it during the redesign. That was the same pass where it checked the address, phone number and hours against its Google listing, without being asked.

Here's the thing: the AI can check against the archive and public listings, but only someone who knows today's details can say they're right. So that person reads every page before the domain is pointed at it. I treat the website as the master copy of a business's details, so they have to be today's, not the archive's.

Can you restore a WordPress site from the Wayback Machine?

No, you can't restore a WordPress site from the Wayback Machine, because the archive keeps the pages visitors saw, not the database or code behind them. What you can do instead is rebuild the pages from the archive, either as a new site or by putting the text back into a fresh WordPress site. Copying a whole WordPress site needs its files and its database, and the Wayback Machine holds neither. Rebuilt text can go into WordPress without an AI plugin, through the WordPress REST API, the built-in way other software adds content to a WordPress site.

Copying a whole WordPress site is a different job from rebuilding one. It takes the files (theme, plugins and uploaded images) and the database (pages, posts, settings and users) together, so it needs the old site itself.

One of the five I rebuilt was a WordPress restaurant site. Its pages came back, its WordPress image addresses didn't, and it was rebuilt as nine static pages: finished pages sent as they are, with no WordPress behind them.

Want the rebuilt text in WordPress instead? Uploading content to WordPress with AI covers how an AI gets content in without a plugin, and getting a WordPress REST API key for AI covers the application password the AI needs to get in.

On a static site, changes go through the AI or a developer, while WordPress lets the owner edit pages themselves. All five of mine were rebuilt as static sites.

Want a hand bringing a lost website back?

You can get a hand bringing a lost website back, or rebuild it yourself with AI from the Wayback Machine. IvI SEO's local SEO service covers getting the website, the Google Business Profile and other listings to agree, and can include a new website when one fits the needs and agreed scope of the work. Whichever route, check every detail from the archive against what's true today before the new site goes live.

Doing it yourself? The brief and the checks above are all you need, and my prompt shows a brief doesn't have to be pretty. (Fair warning: you might end up typing "looks very AI sloppy" too.)

If you'd rather have help, I'll tell you honestly whether a rebuild is worth it for your business, as part of the wider local SEO work.

Either have a nose at the rest of what I do, or grab the brief and have a go.

Your call.