What an automated SEO tool should let you approve and undo

An automated SEO tool writes to your live website and, in some cases, spends your money. That's the point of it: the work gets done without someone copying titles into a CMS at eleven at night. It is also the risk. A bad change goes live on a page that earns money, and a bad purchase can't be taken back.

Most product demos show what the tool can do. Few show what you can stop it doing, or what happens when you want something back the way it was. Those are the questions that decide whether you can trust it with a site that matters.

This guide is a checklist for those questions, section by section, with what a good answer looks like. At the end, one page change and one link purchase are traced through a change log so you can see what "undoable" should mean in practice.

Approval modes

The default should be that nothing goes live until a person approves it. You should see the proposed change, the reason for it and the page it affects, and be able to approve, edit or reject it.

Ask: What is the default when I connect a site? Can I edit a proposal before approving it, or only accept or reject?

Autopilot per site and per change type

Autopilot (the tool making changes without asking) is useful for low-risk, repetitive work. It shouldn't be all or nothing. You might trust the tool with meta descriptions on a blog but want to approve every H1 on product pages.

Ask: Can I turn autopilot on for one site and not another? For one type of change and not another? Is it off until I turn it on?

Checks before every write

Time passes between a change being proposed and being written. In that time, someone on your team may have edited the page, or someone may have switched autopilot off. A careful tool checks again at the moment of writing, not just when it drew up the proposal.

Ask: Immediately before writing, does the tool confirm the page is still as it was when the change was proposed? Does it confirm autopilot is still on for autopilot changes? What happens if either check fails?

A change log with before and after

Every change should be recorded with the page, the field, the exact text before, the exact text after, who or what approved it and when. "Updated title" isn't a log entry. The old title, in full, is.

Ask: Can I see the exact previous value of every field the tool has ever changed? Can I export the log?

Exact rollback that won't overwrite later edits

Rollback should put back exactly what was there before, character for character. Just as important is what it won't do: if someone has edited the page since the tool's change, a blind rollback would wipe out their work. A good tool notices and refuses, and tells you why, so a person can sort it out.

Ask: Does rollback restore the previous value exactly? What happens if the page has changed since? Will it ever overwrite an edit made by a person?

Verification that a change is live

A CMS can accept a change and not publish it: a caching layer, a draft state, a theme that ignores the SEO field, a site that needs a separate publish step. The tool should read the live page after writing and confirm the change is actually there.

Ask: After a change, does the tool re-read the live page? What does it do if the change isn't showing?

Daily limits and cooldowns

Changing many pages at once makes it hard to tell which change did what, and a mistake in the tool's reasoning spreads further before anyone notices. Changing the same page repeatedly makes results impossible to measure, because each change is judged before the last one has settled.

Ask: Is there a cap on autopilot changes per site per day? Is there a minimum gap before the same page is changed again? Can I see both in the settings?

Spending caps

If the tool buys links or anything else, spending needs hard limits that no single recommendation can exceed: a monthly budget, a maximum price per item, and a threshold below which nothing is bought without a person. Money not spent this month shouldn't pile up into a bigger month later.

A purchase is the one thing a tool can't truly undo. Once a seller has placed a link and been paid, there is no rollback. So for spending, the approval and the caps are the undo button.

Ask: Which spending limits can't be exceeded, even on autopilot? Does unspent budget roll over? If two people approve the same purchase, is it placed twice? Can I see the price, page, anchor text and reason for every order?

A pause switch

You need a way to stop everything, quickly, without cancelling your account or disconnecting the site: during a site migration, a legal review, a product launch or a bad week. Pausing should stop queued changes too, not only new proposals.

Ask: Where is the pause switch, and what does it stop? If I pause mid-queue, will changes that were already approved still be written?

An exceptions inbox

Automation should know when to stop and ask. If a change needs a fact the tool doesn't have (a price, a delivery time, a policy), it should ask you rather than guess. If it fails its own checks, it should hand the job to a person rather than try until something passes.

Ask: Where do questions and failures go? Does the tool ever fill a gap with an invented fact?

For reference, Ship/Scale works this way: approve mode by default, autopilot opt-in per site and per change type, at most 3 autopilot changes per site per day and 14 days before a page is changed again. It re-checks the page and the autopilot setting before each write, logs every change with what it replaced, restores it exactly on rollback (refusing if the page has changed since) and re-reads the live page to verify. What it can edit in the first place is covered in the guide to what SEO automation can edit.

A worked example: one change and one purchase through the log

An HR software company, Brightledger, uses an automated tool on its site. All names, times and figures are illustrative.

A title change

The tool proposes a new title for /payroll, whose main search is "payroll software for small businesses". The current title is "Payroll | Brightledger" (22 characters). The proposed title is "Payroll software for small businesses | Brightledger" (52 characters, inside a 60-character limit).

The rollback refusal is the log doing its job. Restoring the whole page to its 4 March state would have deleted Tom's new paragraph. Instead, a person decides: put the old title back by hand and keep Tom's edit, or keep both changes. Either way, nobody's work disappears without someone choosing it.

The cooldown matters here too. Even if the title works, the tool can't propose another change to /payroll until 18 March, 14 days after the title went live, so the day 7 and day 14 measurements describe this change alone.

A link purchase

The same month, the tool recommends a link to /payroll from an HR trade publication that links to two of the three competitor pages ranking above it. The month's link budget is $3,000, with a maximum of $600 per link, and $1,850 has been spent so far.

There is no rollback row here, because there can't be. That's why the log records everything before the money moves: the price against the cap, the budget left afterwards, the anchor text and the reason. And it's why the second approval matters. Without a check that the order had already gone through, Sam's click would have bought the same link twice.

Ship/Scale handles link orders in the same way: each is logged with price, page, anchor and reason, and placed once even if approved twice.

Where this checklist goes wrong

• A setting on a sales page isn't a setting in the product. Ask to see each control in a trial account, and test rollback on a page that doesn't matter before you rely on it.
• Rollback can't undo effects. Restoring an old title doesn't restore last month's rankings or clicks. Rollback limits the damage; it doesn't reverse time.
• Logs only help if someone reads them. Put a short weekly look at the log on someone's list, especially in the first months of autopilot.
• Caps can be set too loose. A $50,000 monthly budget with no per-link maximum technically counts as a spending cap. Set limits you'd be comfortable seeing fully spent on a bad month.
• Your own systems can interfere. Caching, staging workflows and CMS plugins can all get between a tool's write and what visitors see. Live verification catches most of this, but only on the pages it checks.

The short version

Before trusting an SEO tool with a live site, check that nothing goes live without approval unless you've switched autopilot on for that site and change type, and that every write is re-checked beforehand and verified afterwards. Insist on a log with the exact before and after, and a rollback that refuses rather than overwrites later edits. Look for daily limits, cooldowns, hard spending caps, a pause switch and a place where the tool asks instead of guessing. For purchases, remember there is no undo: the approval and the caps are the control.