Autonomous AI SEO.
Your rules.

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.

Your brand in the panel on your own domain, reports and alerts from your own address, and a vendor-neutral help centre. The DNS target ssl.preferium.com is publicly visible and includes the Preferium name; legal documents identify Preferium where required. The launch offering includes sign-in and authentication invitation emails under your brand, plus vendor-free delivered HTML.

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: a monthly platform subscription, a monthly seat per connected client domain, and credit packs when you need them. You set the client price and keep the margin.

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.

Once the client’s DNS record is in place, the platform verifies it, issues the certificate, confirms the origin answers, detects the site and runs the first audit on its own. From there the engine works at the control level you set.

There is no date yet. Agencies on the waitlist are invited first, as founding partners, before registration opens to everyone.

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have to open the panel: everything is there when you want to look, not because you must, and every edge change can be withdrawn with one click.

On Full autopilot, enabled changes within your rules deploy without a queue. You control visible rewrites per site. Invisible-only holds that text for review, and Review everything needs approval for every change. Menus, footers and sidebars are never touched.

It is built to fail toward the original page. Nothing is written to the client’s CMS: every change is an edge delivery that can be withdrawn per element and stops on the next request — what search engines already cached updates on their next crawl. After every deploy a real browser re-opens the live page and re-checks title, meta description, H1 and structured data; a change that made things worse is rolled back automatically and you are told. Drift detection every 30 minutes repairs changes something overwrote. Everything else stays reversible with one click. If anything in the pipeline fails, or Preferium’s own data is unreachable, the original page is served. Not covered: an outage of Cloudflare itself, or a DNS change at the client’s provider.

No. Cloaking means showing crawlers different content than humans see. The engine serves the same HTML to both, and that is checked rather than asserted: the cloaking guard fetches each optimized page twice, once as a person and once as GPTBot, and flags any serious difference to you. 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.

Four assistants with prompt tracking, six answer surfaces in total. Scheduled prompts are sent to the live ChatGPT, Claude, Perplexity and Gemini with web search, and Google AI Overviews and Google AI Mode are tracked separately per keyword — two different surfaces with different citations. The cadence follows the intensity you set, up to weekly, and any prompt can also run on demand; your plan sets how many of the four assistants are included. 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.

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, 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.

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 scheduled citation measurement.

Each client is its own workspace with data isolation at the database level — clients never see each other. Client workspace data is stored in the EU (Stockholm). Edge delivery runs on Cloudflare's global network, and AI measurement providers are listed in the subprocessor register. A Data Processing Agreement is offered. Preferium holds no security certifications, and says so plainly rather than implying otherwise.

Three cases. A client who cancels keeps being served for 14 days — an undo window — and billing stops at once. A client who has earned it can move to frozen delivery: the published pages stay live, unchanged and unbilled, with no active optimization. A client who disconnects stops being served, the certificate is released and the site data is deleted; the client restores the original DNS record from the snapshot taken at connect. In every case the deployed changes can be exported as CSV, and everything written for the client is the client’s. The agency white-label terms govern.

Yes. The engine detects each page’s language and optimizes in that language — a Norwegian page gets Norwegian titles and answer blocks, a German page gets German — and eight hreflang checks find broken links between language versions. Citation tracking runs in the client’s market and language too.

Preferium’s current DNS connection model does not support Wix or Squarespace. Other tools may offer integrations for those platforms; we do not claim that a JavaScript snippet is their only option. Shopify and similar stores also need a supported connector. On other platforms, onboarding checks the routing and origin before accepting the site; DNS access alone is not a compatibility guarantee.