Programmatic SEO · Ship gate
The exact bar we score every data page against before it ships.
We publish programmatic data pages at scale and outrank much larger data companies on their own keywords. No backlink campaigns, no PR. Three parts: a scorecard (is it good), hard gates (is it disqualified), and the ship bar (does it go live).
Every rule below exists because a page failed without it. Nothing here is theory.
Seven dimensions, scored 0 to 5. Pass line is 4+. Each comes with the lesson that put it here.
The searcher's answer, usually the headline number, is stated as a sentence in the H1 and hero ("There are 202,536...").
LearnedOur first site audit found 1,583 pages with multiple H1s and 1,428 titles too long to display. You cannot rank on basics you got wrong at scale.
Every shared figure is copied byte-for-byte from a single registry, with its exact scope label. Never re-derive or round a number that already exists.
Learned"268,069 companies," "161,817 practices," and "202,536 dentists" can all be true for one vertical. Without scope labels they read as three contradictions, and Google notices contradictions before buyers do.
No paragraph could be pasted onto another vertical unchanged. The H1 is not a formula and not a near-duplicate of any page you already ship.
LearnedWhen we audited our indexing, 42 of the 50 non-indexed sample pages sat inside a single 158-page near-duplicate cluster. Google didn't penalize us. It just declined to index the copies.
The page answers the exact query it targets, and the query reaches your actual buyer.
Learned"[vertical] leads" and "[vertical] email list" look like the same keyword. The first reaches contractors buying consumer leads, the second reaches our ICP. We almost built 995 pages for the wrong buyer.
Every sentence carries new information. No hedging, no filler openers, and the hero number is never restated.
LearnedAbout 30 pages failed QA on one single tell: the hero number repeated in the proof tiles, FAQ, and accordions. We paused a paid generation run mid-flight and fixed the recipe, not the pages.
One H1, clean heading order with no skipped levels, the required beats present (proof band, tiles, table, "who buys this," FAQ, close), valid schema.
Blunt operator voice with an actual opinion and a genuinely disqualifying "when not to buy" beat. Never emoji, never a pun.
LearnedThree full redesign passes came back with the verdict "only one page is great, the rest are AI slop." The fix wasn't a fourth pass. It was a written voice bar every page gets scored against.
Voice and anti-fluff are the dimensions with teeth (see the ship bar). A 5 on structure cannot rescue a 2 on data.
Any one of these kills the page. Auto-fail, no exceptions.
GATES FAIL CLOSED: a check that could not run is also a fail.
A reviewer caught one of our pages claiming 480,000 in the hero and 200,000 further down. One page, two answers, zero trust. Numbers must render byte-exact against the registry.
We audited 29 "top picks" against live third-party sources before shipping: 2 had fabricated prices (one listed at $249/mo was really $449) and 1 was the content vendor promoting its own product. Every claim gets a source label or is framed as opinion.
Cross-sell by name, never by number. A foreign vertical's count in your body text is the strongest signal the page is a copy-paste of its neighbor.
Banned H1 skeletons auto-fail, and no two pages may share an H1 shape or tail. Templates scale; template headlines get you clustered and skipped.
More than one H1, a title outside 10 to 65 characters or not aligned to the slug, a missing or non-self-referencing canonical, or any noindex.
Under the length floor, or "contact sales" as the only answer. If the page carries nothing a static list couldn't, it has no reason to rank.
Every internal link must resolve to a real live page. A hub pointing at a spoke that doesn't exist fails closed.
A page ships only when all of these are true.
Every hard gate passes. Green across roughly 25 automated checks per page. Passed equals total. Not 24 of 25.
An independent reviewer clears the floor. A second reviewer who cannot edit the page scores it against a written gold standard: voice ≥ 3, storytelling ≥ 3, anti-slop ≥ 4. A reviewer error is a fail, never a silent pass.
The hero number obeys the one-mention rule and every proof number is a genuinely different figure.
One last lesson: we once rebuilt 23 pages that were already live and passing. Result: zero net-new pages and 5 regressions. The bar protects what's shipped as much as it filters what's next.
The moat isn't the content. It's refusing to publish a page that doesn't earn its spot.
Teja Varma · GTM engineering, Orbital · came here from LinkedIn? This is the rubric.