PreferiumAutonomous AI-SEO. The engine runs the enabled work itself.
Switch the workflows on, and the engine runs the enabled, supported ones on its own — every day, without anyone opening the panel. Not drafts waiting for approval: deployed changes, with title, meta, H1 and schema re-checked in a real browser afterwards.
Sample data — Northwind Cycles is a fictional bike shop. The numbers show the product, not measured client data.
Mention rate per engine — the share of tracked prompts that name the client in the answer. Citation share, the number that matters, is on the scorecard.
Sample data — Northwind Cycles is a fictional bike shop. The numbers show the product, not measured client data.
The limit you see at 08:02 is one of the rules in Instructions: “max 12 changes per week”.
One engine. Five chapters to read.
Each chapter has its own page with the modules, the panel view and the numbers. The chapter bar above takes you between them.
You connect and configure. The engine does the work in between.
One DNS record. No plugin, no server access.
The engine sits in front of the client site, not inside it. That is why nothing is needed from the client's developers — and why a client leaves as easily as it arrived.
Your clients are never locked in.
We take a snapshot of the original DNS record when a site connects, and keep showing it — so you always know exactly what to restore.
The panel your clients log in to.
The scorecard, the checks, the log of every change and the button that rolls it back live in one panel — under your brand, on your domain. The same one you saw at the top of the page.
Talk to usEvery change is versioned. A rollback hits the edge directly, so it is live on the next request — no republish, no queue.
You hear it from the panel first.
Alerts by email, mirrored to Slack, one type at a time. Each can be switched off individually, so you hear about what matters and nothing more — and the reports carry your brand, not ours.
Plus eleven more: ranking drops, citation milestones, portfolio summaries, delivery status and onboarding guidance. Everything we have sent sits in a searchable delivery log.
The engine inside your own systems.
The panel is not the only entry point. What is readable in the panel is readable over the API — and the AI assistants you already use can be connected straight to a workspace.
In short. 17 questions, in 3 groups.
What the engine does, what your clients see, and what applies to the agreement. None of the answers require you to book a call first.
Do my clients need to install a plugin?
No. Each client connects with a single DNS record (a CNAME). The optimization happens at the edge, in front of the site, so it works on WordPress, Webflow, Framer, Kinsta, Next.js or a fully custom stack — no plugin, no CMS access.
What does my client actually see?
Only your brand. Preferium runs white-label — your logo, your colors, your URL. Your clients never see Preferium; they see the agency they hired.
How does billing work between me and my clients?
You connect your own Stripe and define your own packages and prices. Your clients order those packages — subscriptions and credit top-ups — inside your branded panel, and every client payment runs on your Stripe. Separately, Preferium bills your agency wholesale for the domains and credit units each active package delivers. You set the client price and keep the margin.
Can clients order without my help?
Yes. Your clients can order the packages you have defined right inside your branded panel, paid on your Stripe. Checkout always happens inside the panel, never on a page that charges without your setup behind it.
How long until a client site is live?
Usually minutes. One DNS change connects the client; the platform then detects the site, fills in the brand details and runs the first audit. From there the engine can optimize every page at the control level you set.
When does partner registration open?
Preferium for agencies is in private preview. Join the waitlist and you will be invited as a founding partner as registration opens.
What does “autonomous” mean here, concretely?
That the engine finds, fixes, deploys, verifies and rolls back — on its own, every day. On Full autopilot you never have to open the panel: everything is there when you want to look, not because you must, and every change can be undone with one click.
Does it publish without anyone reviewing first?
On Full autopilot, yes — that is what separates it from the category, where drafts wait for approval. Invisible-only auto-publishes title, meta and schema and holds visible text for review. Review everything ships nothing without your approval. You set the level per client.
Is it safe to let it change a live client site?
Yes. Nothing is written to the client’s CMS — every change is an edge delivery, reversible per element, so the worst case is always one click from undone. On top of that, the four elements most likely to hurt if they go wrong — title, meta description, H1 and schema — are re-opened on the live page in a real browser after they ship and rolled back automatically if the page regressed. The remaining element types deploy without that automatic check today; they are still reversible per element.
Isn’t rewriting HTML at the edge a form of cloaking?
No. Cloaking means showing crawlers different content than humans see, and that does not happen here — the content is the same for everyone, and it is checked rather than asserted: after a deploy a parity guard re-fetches the live page twice, once as a browser and once as GPTBot, compares the prose the engine never touches, and alerts an operator if the two ever diverge. The one place delivery differs is a JavaScript-only single-page app, where the origin serves an empty shell to anything that cannot run JS. There the page a human sees after the JavaScript has run is pre-rendered for those crawlers — same content, reached a different way. That is dynamic rendering, a pattern Google documents and supports.
Which AI engines do you track?
Six answer surfaces, on every plan: ChatGPT, Claude, Perplexity and Gemini as separate engines, plus Google AI Overviews and Google AI Mode tracked separately per keyword — they are two different surfaces with different citations. For each, the tracked prompts are run weekly and the answer is read: who is cited, share of voice against competitors, sentiment, and new or lost mentions. Grok and Copilot are reading surfaces, not separate measurement sources yet. The clean, semantic HTML served at the edge makes the content readable there too, without reporting citations from them as if they were measured.
Do you write articles for our clients?
No — and that is deliberate. The market is full of “AI visibility” services that only produce more content. Preferium is the layer underneath: it continuously fixes the technical and structural SEO of the whole site — titles, schema, canonicals, hreflang, health issues, answer blocks — and maintains authority signals over time. It makes the content your clients already have crawlable, parseable and citable. A perfect article on a broken foundation is still invisible to AI.
Are the numbers on this page real?
No — and it says so next to each of them. The numbers here show what the panel looks like, for the fictional Northwind Cycles, and carry a visible sample-data label. Real numbers live in each client’s own panel, from GA4, Search Console and the weekly citation measurement.
Is client data isolated and GDPR-ready?
Yes. Each client is its own account with strict data isolation — clients never see each other. Client data is stored in the EU (Frankfurt or Stockholm), and a Data Processing Agreement is offered. Preferium holds no security certifications, and says so plainly rather than implying otherwise.
What happens to the optimizations if a client cancels?
What happens to deployed changes at termination is set out in the agency terms. What you always keep is the record: every deployed change can be exported as CSV at any time, so nothing is locked in and everything can be re-implemented anywhere. Ask for a technical disconnect instead, and the original site is served again.
Does it work for non-English sites?
Yes. The engine detects each page’s language and optimizes in that language, with correct hreflang handling — a Norwegian page gets Norwegian titles and answer blocks, a German page gets German. Citation tracking runs in the client’s market and language too.
Why not Wix or Squarespace?
Preferium connects through DNS, and closed builders do not allow the routing the edge needs — the only workaround there is a JavaScript snippet, which AI crawlers do not execute. Only installs that actually reach them are sold. Shopify and similar stores pin the domain to their own servers for SSL; a native connector is on the roadmap. Any platform where the client controls DNS works today: WordPress, Webflow, Framer, Kinsta, Next.js, or fully custom.
Want to see it on a client site?
Pick one of your client sites. We run it through the health checks, show you what the engine would fix first, and walk you through the panel your clients would see under your brand.