ArticleGrade reads the live page Google and AI answer engines actually see, scores it with two independent models, and tells you whether it's good enough to get surfaced — then fixes what isn't. Any site, any CMS.
Being live is not the same as being found. The bar for getting surfaced at all — by Google's ranking systems and by AI answer engines — went up, and it's the same bar for every website owner, on any CMS.
The quality signals ArticleGrade grades — E-E-A-T, factual depth, clean non-slop writing, structure, schema and author data — are exactly what both Google's ranking systems and AI answer engines like Google AI Overviews, ChatGPT, Perplexity and Copilot reward when they decide what to surface and cite. So "will my page be seen at all — by Google and by AI?" is the question this pipeline answers, page by page.
Four moves take a page from a link you paste to one that clears the bar you set: fetch it, score it, gate it, and loop the fixes until it passes.
We pull the rendered URL exactly as Google and AI crawlers index it — not your draft or your CMS preview. Schema, author block, meta and links are extracted from what actually ships, and recovered even when a site blocks ordinary bots.
live fetchrendered HTMLschema + authorThe page is graded 0–100 across six dimensions and 26 gates. One engine evaluates quality and facts; a second evaluates SEO and structure — two independent reads, not one model marking its own work.
26 gates6 dimensionsdual-LLMYou get a pass/fail verdict against the bar you set, the score, and a list of phrase-level issues — each with the exact text, a High / Med / Low severity, and a one-line fix. Delivered as JSON via API or in the dashboard.
pass / failphrase-level fixesAPI or UIWhere you opt in, ArticleGrade applies the fixes — tightening claims, adding sourcing, removing AI artifacts — then re-fetches and re-scores. The remediate-and-re-audit loop repeats until the page passes your threshold, so nothing ships below the line.
apply fixesre-scoreloop to passThe headline score rolls up the buyer-legible signals below — the same ones search and AI both reward, built from the dimensions Google's own quality guidance describes.
Experience, expertise, authority and trust — the framework Google's raters use, its core ranking reflects, and AI answer engines scrutinize before they cite.
The tells of unedited generation — hedging, filler, repetition, template leakage and a flat machine register — the slop both Google and AI answers skip over.
Whether claims are specific, sourced and verifiable — or thin, vague and unsupported. The difference between a page an answer engine cites and one it ignores.
Title, meta, heading hierarchy, internal links, schema and formatting — the on-page fundamentals that frame everything else and let both crawlers parse the page fast.
We split the job across two independent models, each doing what it's best at — and we screen with the cheap one before spending on the expensive one.
Judges whether the content is actually good: E-E-A-T, factual depth, sourcing, and the artifacts of unedited AI. This is the verdict that drives the headline score and the remediation loop.
Judges the technical layer — title, meta, headings, links, formatting and CMS artifacts — fast and cheaply. It runs on every page as a screen, and never fact-checks (so it can't "correct" from stale training data).
A two-stage architecture keeps quality high and cost low — a near-free structural screen on every page, the full dual-LLM deep audit only where it counts.
Grading a live URL means fetching arbitrary pages — so the fetch layer is hardened against the things that go wrong when you do that at scale, on any CMS.
Private, loopback and cloud-metadata addresses are blocked, with DNS pinned at connect-time to close rebinding — every redirect hop re-validated.
When a WAF blocks a normal fetch, we recover the page's schema and author from its public structured data instead of guessing.
We audit the page and store the result for your dashboard. We don't train on your content, and the model providers are configured not to retain it.
Anyone can ask a model "is this good?". The defensible part is everything around the question: a live-fetch harness that survives the open web, a 26-gate rubric tuned on hundreds of real audits, two-stage screening that makes it economical, and a remediate-and-re-audit loop your pipeline integrates via POST /api/v1/audit — reading back {"verdict":"pass","score":82} to gate anything under your bar before it goes live.
Paste any URL and watch the real pipeline score it — free, no signup, in about twenty seconds. Find out what Google and AI actually see.