Move repeatable SEO generation to a rate built for scale.

$5 of credit at signup, no card, valid ten days — roughly 2,400 pages, which is enough to run a real slice of your own catalogue and put the output beside what you generate today. If you would rather not wire anything up first, send 20 to 50 anonymised production prompts — titles, metas, briefs, whole article bodies — and we will run them through OmevAI Lite and Pro and hand back the raw outputs, so your editors can grade them blind against your current model. Those prompts are Customer Content like any other request and help train our models, so send anonymised ones with no personal data in them. Either way it costs an afternoon, not a quarter.

One configuration change, on one job. The best first candidate is usually metadata or anchor text: high volume, tightly specified and easy for an editor to grade quickly. Leave briefs and flagship pages exactly where they run now, and every call you do not route keeps hitting the provider you use today.

Reverting the route is a config value. Reverting a published page is not. So generate with both providers for a week and publish nothing until you have compared them. Keep your current provider wired as the fallback for the one job you moved, and your existing retry and quality checks still stand in front of the CMS, because only the destination changed. What you are risking is a week of engineering time, not your index.

We train on every request, and there is no opt-out on self-serve. That mandatory training loop is part of why the rates look the way they do, so check it first rather than last. Self-serve content must contain no personal data, raw content is kept no more than 24 months and is then deleted or irreversibly anonymised, and the Terms, Privacy Policy, DPA, Acceptable Use Policy and subprocessor list are all published for your reviewer to read rather than take on trust. If a client contract bars you from sending their content to a training-enabled endpoint, that conversation comes before the pricing one — and if it cannot be resolved, we are the wrong supplier and better known now.

Less than you may be used to, and we would rather say it on the page than in a procurement call. No SOC 2, no ISO 27001, no contractual SLA. We publish no uptime, latency or throughput figure and will not invent one. Enterprise sets limits by contract; on Pay as you go there is nothing to buy priority with. Run a real batch on the free credit and decide from what you measure, not from what we claim.

We publish no benchmark score and claim no parity with anyone: a score on someone else's prompts tells you nothing about yours. The claim is narrower than that. On well-specified, repetitive work built from a template you already validated, the two numbers that decide the real cost are how often the first pass comes back usable and how much of it survives editing, and both are measurable on your own prompts in a day. Set a pass mark per job, and leave anything that misses it on the model you run today. Cheaper generation does not change the editorial questions either — topic selection, internal linking and whether the page answers the query are still yours, and where the law requires you to disclose that content is AI-generated, that duty stays with the publisher.

On self-serve, there is no contract, seat or monthly plan. Top up from $10, move one low-risk job and keep the current provider connected while you validate it. Paid credit is valid 30 days from activation, and there is nothing to cancel: stop routing traffic and spend stops. Enterprise is a signed arrangement with custom limits and a manual approval step. The migration case grows with volume, so start with the workflow that repeats most and let the measured cost per approved page decide how much more to move.