Reflexa installs an agent on your website. It reads your pages the way Google and the AI engines read them, applies the fixes it is allowed to apply, writes the words for a page and publishes them the moment you approve, and counts every AI visit page by page. This is exactly what it does, what it knows about itself, and the eight things it will never do — enforced in code, not promised in a policy.
Most tools that look at your website do it from the outside and hand you a list. Reflexa does the outside part too — and then goes in. The agent is a small piece of software installed on the site itself, paired to that one site by that site's own token, and from there it can change things.
That is the whole difference. A finding you have to act on yourself is a chore. A finding the agent can carry out is one click.
You ask in your own words. It answers from what it knows about the site and comes back with a plan where every item is typed: a switch it flips itself, words it writes for your approval, a thing only a person can do. Attach a screenshot and it reads that too. Nothing is applied until you press it.
It reads your pages the way Google and the AI engines read them. 23 structural checks, in two families: what a search engine needs before it can index and rank a page, and what an answer engine needs before it can quote one. No model is involved and nothing is installed — this part runs from outside and costs nothing. The 23 checks, one by one →
It switches on what keeps AI out. Four things it applies
itself, from inside, each with one click from you: crawler access for AI
retrieval fetchers in robots.txt, Organization data in the page
head, the sitemap, and an /llms.txt built from your own page
titles. Every one is reversible — switch it off and the site is exactly as it
was. Nothing is written into your theme or your content.
It writes the words, and publishes them once you approve. A page title, the search description, an answer-first opening paragraph, an FAQ block. The agent writes the exact text; you read it in the dashboard, exactly as it will appear; nothing is sent until you press apply. The change lands as a WordPress revision with the previous value kept, so it can be put back exactly.
On a coded site, its hands are your repository. Connect the site's GitHub repository once — one App install, one repository — and every change you approve becomes a commit on your default branch, named for what you approved: the robots.txt rules, Organization data in every page, sitemap.xml, /llms.txt, a page's title, description, opening or FAQ, a redirect. You read the commit; undo is another commit. Nothing is committed until you approve it.
It fixes a broken link, once you approve it. Ask it to fix a URL that answers 404 or points at an old page and it names the redirect exactly — from which URL to which page on your site — and checks that the target answers. Approve, and the 301 is live from the next request, on WordPress and on a coded site alike; undo takes it off again.
It counts every AI visit, page by page. Which of your pages an engine opened to answer a real person right then, which it only crawled to learn, and who clicked through from an answer — per page, per engine, per day. What that counter actually sees →
It answers what you ask. In your own words, about your own site. It answers from what it actually knows — the checks, the AI visits, Google Search Console — and returns a plan where every item is marked: this one it switches on itself, this one it writes and you approve, this one only a person can do.
Before it answers anything, the agent reads its own state: whether it is on your site, whether its heartbeat is current, which plan you are on, which counters are connected. If you ask for something out of reach it says so and says why, instead of promising it. An agent that is not installed will tell you it can write the words but you will be the one pasting them.
The same goes for what it cannot see. With no traffic counter connected it will not tell you that AI visited your site zero times — it will tell you it cannot see. An absent counter is not a zero, and the difference matters when you are making decisions from it.
You are letting software into your website, so these are the things it cannot do, whatever you or anyone asks it. They are enforced in code — the plugin refuses the instruction and reports the refusal — not written in a policy page.
On a coded site — Next.js, Astro, a static build, anything behind a CDN — one file in your middleware, your CDN worker or your page template, and it is in. On WordPress the Reflexa plugin installs in three steps and needs no developer. The structural checks run from outside either way. How to put it on your site →
No. The four switches each need one click from you, and every word it writes is shown to you exactly as it will appear before anything is sent. Nothing is applied that you have not approved.
Yes, all of it. A switch turns off and the site is exactly as before. A page edit lands as a WordPress revision with the previous value kept, so it is restored exactly. Nothing is ever written into your theme or your content.
The switches stop applying and the site keeps whatever was last set — you can turn each one off first if you want it reverted. Your checks and everything the agent remembers stay in your Reflexa account.
Most sites are not. The structural checks run from outside on any site, and for the counting and the fixes a coded site needs one file in its middleware, CDN worker or page template. WordPress has a plugin that does the same for you.
No. It reads public pages, the site's own published page list, and the requests AI engines make. It never touches checkout, cart or account pages, never reads customer records, and never emails anyone.