We run AI-powered SEO the way it should be run — technical foundations first, then an in-house content engine that publishes weekly, with humans editing the parts that matter. The traffic is real. The leads convert. AI engines cite our answer capsules, not our competitors'.
Same four pillars across every site we run. Mix shifts by market — the structure doesn't.
Every site we run earns a Day 1 audit: speed, indexation, schema, internal links, broken anchor map.
Direct-answer capsules, FAQ density, schema. ChatGPT, Perplexity, Claude, and Gemini cite the capsule verbatim.
Service × vertical = predictable, scalable spokes. Tier-2 pages with stack, math, FAQ, and related blog posts.
Calculators and audits, not gated PDFs. Each tool indexes for a specific buyer-intent query and routes to a discovery call.
The same pipeline runs across every site we ship. Brand voice, internal-link graph, and stream overlays change per project — the gates stay identical.
Apify Google Search → 3 queries → competitor filter → Claude Haiku synthesis (key facts, statistics, common questions, content angles).
Claude Sonnet generates the H1, sections, subsections, FAQ block, CTA. Mandates: 9-11 H2s, comparison table, case study.
Score every internal page in the site's page index against keyword + outline. Pick top N (default 10) by relevance, with pillar boost.
Fresh Apify search for "{keyword} statistics research". Filter blocked competitor domains. Prioritise trusted-source matches. Dedupe by domain.
Industry-aware Unsplash queries. Photographer dedup. Claude visual-similarity verifier. 5 photos per post — 1 hero + 4 sections.
Claude Opus drafts ~2400 words to outline. Stream voice overlay (operator-essay, industry-watch, counter-narrative, service-spoke). Banned-phrases gate.
Claude Haiku generates SEO title (≤60), meta description (≤160). Validator checks primary keyword presence in title, seoTitle, meta, answerCapsule.
Build-time gate: word count ≥1100, Flesch ≥40, grade ≤14, ≥3 numeric tokens, ≥5 internal + ≥4 external editorial links, alt text on every img.
The bottleneck is editorial taste, not generation cost. Max Plan auth means generation is effectively free; the only meter that matters is human review time. Here's the math we run on every programmatic build.
Five lanes feed each other: schema-correct on-site → community drumbeat → directory presence → LLM citations across the four AI engines that matter. Each citation is its own measurable signal we track per query.
No "10 tips for X" filler. If a draft doesn't fit one of these seven shapes, it gets killed before publish. North star: capture intent, become the citable reference, or convert in 7 minutes.
Take a system we run internally, give it a name, document it in public. Low search volume, high inbound links because the named concept becomes citable.
One project, one system, one outcome — with public numbers. Real revenue > vanity metrics. If we can't share numbers, the post dies.
"We analysed N → here's what's broken." Original data, link-bait shape. Anchors a quarterly link-building push.
Pick a fight with conventional wisdom that's mathematically wrong. Sabri / Hormozi shape. Build the tribe.
Programmatic Tier-2 spokes — "{service} for {vertical}". Stack, math, use cases, FAQ. Operator bookmark target.
"Vapi vs Retell", "Make vs n8n vs Zapier" — high-intent buyer queries competitors don't bother writing. Honest framing only.
Each tool is its own indexed page that doubles as a lead magnet. The page IS the funnel.
Not a list of "blog ideas". An assembly line for AEO-era discovery, run the same way for every project, with the mix calibrated by market.
Each tool indexes for a high-intent query and routes to a Closed Loop Score. No email gate at entry — soft gate at the result page where the value lives.
4 services × 8 verticals = 32 cluster spokes. Each spoke is a Tier-2 page targeting "{service} for {vertical}" with stack, math, FAQ, related blog posts.
Every spoke and blog post emits FAQPage schema and a 40-60 word direct-answer capsule near the top. AI engines pull the capsule verbatim.
"Vapi vs Retell", "Make vs n8n vs Zapier", "Voice agent failure patterns" — high-intent comparison queries. Honest framing only.
Reddit drumbeat (M/W/F, 6 subs, 80/20 value-to-promo), citation tracking across ChatGPT, Perplexity, Claude, Gemini. Real attribution on real queries.
Every tool is its own indexed page that doubles as a lead magnet. The page IS the funnel — no email gate at entry, soft gate at the result where the value lives.
Quantifies your inbound-to-revenue funnel against peer benchmarks. Output: a $/month leakage figure tied to specific stages.
A diagnostic against the 89% of mid-market agencies that fail an honest audit. Operator-side checklist, no agency required.
Maps your inbound → qualification → booking → follow-up loop and shows the exact stage where it leaks. Closest to the brand promise.
Sends a phantom inbound through your real telephony stack and grades the response. Visceral. Build last because telephony has the most edge cases.
Every line below is a real commit landed on prod. The kind of in-production iteration that proves the engine is alive, not a deck.
Removed reveal-animation gate that left the section invisible until React hydrated. Now renders immediately under the Score-loop CTA on every post.
Inline TOC paragraphs auto-rewritten to canonical <nav.toc><ol>; every <table> tagged class="bp-comparison" so the bordered, zebra-row, lime-header styling engages. Idempotent.
Build script now strips the source <figure> when promoting it to hero. No more "two of the same image" at the top of every post.
Three layers: 10-result candidate pool, photographer-username dedup, Claude vision-style pass that flags "same scene, different ID" duplicates.
Write prompt now hard-rejects "X vs Y" / "trigger → outcome" / 3+ column data rendered as <ul>. Post-write stderr verifier catches misses.
Auto, 3 AM Problem, onboarding videos, AI bundling — all regenerated through the v2 image pipeline. Old slugs redirect permanently to new ones.
Four voice streams (operator-essay, industry-watch, counter-narrative, service-spoke) with per-stream depth overrides + voice overlays.
4 services × 8 verticals. Each spoke: stack, math, use cases, FAQ, related blog posts. OG images auto-generated at build.
We don't pitch decks for SEO. The engine that publishes our blog, the topic cluster that wires every post to a vertical landing page, the validators that kill drafts before they ship — all of it runs against this domain in production.
We diagnose first, build second. If your loop isn't ready to be closed, we'll tell you that — and the call is still free.