You do, every time, not the agent. Nothing it writes or switches on applies until you have seen the exact change, in full, and pressed apply yourself, and even then it goes back to re-read the live page to confirm the change actually landed.
This is the question that sits underneath everything the Reflexa agent does on your site: not just whether it can change a page, but whose decision it is when it does. The short answer is that it is always yours, walked through end to end below, one change at a time.
No. Every kind of change goes through the same proposal first, whatever it is. A switch shows you what it will turn on. A piece of writing shows the full text sitting in place on the page, not a summary of it. A code change, on a connected repository, shows you the exact diff. You read it as it will actually appear, and nothing is sent to your site until you click.
| Kind of change | What you see before you click |
|---|---|
| A switch (crawler access, Organization data, sitemap, /llms.txt) | Exactly what it turns on, with one click to apply it |
| Page text (title, description, opening, FAQ) | The exact words, shown as they will sit on the live page |
| Anything else on the page (stylesheet, snippet, redirect, new page) | The exact change, before it is applied anywhere |
| A connected repository | The commit's diff, before it is pushed to your default branch |
Why it matters: a vendor who says software "fixes your site automatically" without showing you what that means in practice is asking for blind trust. Reflexa's rule is the opposite: see it, then click it.
Yes. Even the four things the agent can apply itself, AI
crawler access, Organization data, the sitemap and /llms.txt,
go through a proposal and a click, not a single toggle that fires on
its own. It is easy to assume a "switch" means the software
decides and acts, but here the switch is only the label for the kind
of change; the approval step in front of it does not change.
Do this: for the full list of what each switch does and how it is reversed, see Can an AI agent actually change my website?
The change goes live, and then the agent checks its own work instead of taking its own report on trust. A page edit lands as a WordPress revision with the previous value kept. A code change lands as a commit named for what you approved. Either way, the agent goes back and reads the live page again afterward to confirm the change actually landed as shown, rather than marking the job "applied" and moving on.
Why it matters: "applied" and "actually there" are not always the same thing, a caching layer or a theme override can quietly swallow a change, and checking after the fact is the only way to know the difference.
Whoever is logged into the Reflexa dashboard for that one site. The agent is paired to a single site by that site's own token and has no login or decision-making of its own, so approval is never the agent's choice to begin with, it is a permission the dashboard account for that site already holds. The token also keeps sites separate: it is refused the moment it is used against a different host, so a second site on the same account cannot be touched by mistake.
Yes, every kind of change has its own undo. A switch turns off and the site is exactly as it was before. A page edit is a WordPress revision, so the previous value is still there to restore. A commit is undone with another commit. Nothing is written into your theme or your content directly, which is part of why an undo is always available rather than a best effort.
None of this requires buying anything, and each step takes a few minutes:
You do, every time. Each proposal, whether it is a switch, a piece of writing or a code change, is shown to you exactly as it will appear before anything is sent, and it only applies once you press apply yourself.
No. The four switches, AI crawler access, Organization data, the sitemap and /llms.txt, each need one click from you too. Each is shown as a proposal first and applied once you click, the same two-step process as a piece of writing or a commit.
The agent re-reads the live page after you click apply, rather than trusting its own report that the change went through, and tells you if what it finds does not match what it proposed.
Anyone with access to the Reflexa dashboard login for that one site, since the agent is paired to a single site by that site's own token and has no separate login of its own.
Yes. A switch turns off and the site returns to exactly how it was. A page edit lands as a WordPress revision with the previous value kept. A commit is undone with another commit.
The free check reads your site the way ChatGPT, Claude and Perplexity read it, and shows you exactly what it would ask you to approve. 3 minutes, evidence included.
The product: Reflexa Agent · how it works · på svenska