The SEO/GEO Workstream
Most SEO audits echo a tool's flags, mix up metrics, and declare things absent from a sample. I built an Audit Engine that fetches every page live, asks five AI engines the same buyer questions, scores nine weighted dimensions into an SEO score and a GEO score, and ships a 30-60-90 roadmap with named owners. Fifteen client audits have run through it. This is what it does and how I built it.
The SEO/GEO Workstream
Generic AI content fails because the model knows nothing about the company. I built an Article Factory that reads an eight-file context layer before it writes, passes every keyword through a five-check gate, drafts against a 30-point scorecard, verifies with code, matches a lead magnet, and hands off to any publishing system. It runs for about $12 to $15 a month per client. This is what it does and how I built it.
The SEO/GEO Workstream
Sites break quietly, and SEO wins fade when they were a project rather than a system. I built an Optimization Loop: an automated health check every two days, a weekly audit that ends in three to five action items, a weekly AI citation run across five engines with every answer stored, and a full re-audit every six months on the same method. This is what it does and how I built it.
A 59-market home-services company was buying every lead with Google Ads. In one quarter we replaced a $20,155-a-month ad habit with organic search, and monthly leads went from 212 to 427, measured in the client's own CRM.
In February 2026 a marketing-talent marketplace had no way to tell whether ChatGPT, Claude, Gemini or Perplexity had sent it a single customer. By August the channel had billed $305,120, 2.6% of company revenue, and every dollar traced to a signed deal.
When a marketplace cut paid spend 55%, every channel fell 35 to 43% except leads from ChatGPT, Claude, Gemini and Perplexity, which fell 5%. A regression across 26 weeks shows why: $100k of ad spend buys about 162 form fills and a statistically zero number of AI-search leads.
Nobody at the company knew whether AI assistants recommended them. Six months later a weekly panel across ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews showed the brand cited 74 times against Upwork's 53 and Toptal's 34, across 183 citations among six tracked competitors.
A marketplace's organic search looked healthy: 1,031 leads and $366,294 in signed revenue a year. A first-party audit found 2,650 pages telling Google to ignore them, 87% of clicks coming from people who already knew the brand, and a conversion count that was 98% pageviews.
The standard objection to SEO is that it brings browsers, not buyers. A 59-market home-services company's own CRM said the opposite: organic visitors asked to be contacted 2.4 times as often as direct ones, converted to bookings at 24.8%, and were worth $542 each against $465.
At baseline, three of four AI assistants had never mentioned the company. Twelve weeks later it was named in 46% of AI answers across 25 buying questions and 59 US cities, and cited as a source 60% of the time when assistants showed sources. The method, including the miss we published.
A pre-revenue medical device company had zero content, zero ranking keywords and a $5,000-a-month paid budget it refused to increase. We built a compliant content engine, with a 16-term banned-word list, 14 quality gates and a 25-point scorecard, that put nine articles through medical review with six approved and none rewritten.
A Chicago balloon retailer's Shopify store showed up in Google 234,066 times in 90 days and got 2,319 clicks. Half its pages had no title at all. Pages without titles earned 0.38% of clicks; pages with titles earned 1.28%. Same store, same quarter, 3.4 times the rate.
A QR-code SaaS with 12,349 referring domains and 1.9 million impressions in 50 days had a blog hub that returned a 404 while 179 posts sat in its sitemap, and a flagship money page earning 1,904 clicks from 486,174 impressions. An enterprise audit, in plain terms.
A 25-year B2B advisory firm had a 74/100 SEO health score, 1,168 ranking keywords and about 4,500 monthly organic visits. In six category questions across AI assistants and Google AI Overviews it was cited zero times, with a 4% AI share of voice. How I found that with no analytics access at all.
Agencies charge $800 to $2,000 per article. I built a multi-tenant factory that produces SEO and AI-search articles for about $12 to $15 a month per client, queues up to 1,000 a week, and gates every piece with a 30-point scorecard plus a link check that overrides the AI. Then I built the gate that can block our own work.
A DTC dress-shirt brand's Search Console showed 17,077 clicks from 1.39 million impressions in 50 days. A brand-versus-non-brand split showed 84% of those clicks were people typing the brand name, and its category term ranked at position 5.6 with a 0.37% click rate. Same trap I'd found at a marketplace weeks earlier.
A 12-location therapy practice needed per-market SEO and AI visibility in a vertical where a wrong claim is a compliance problem. It became the reference implementation for my Audit Engine's growth-plan layer, and the engagement where every defect the system could produce was caught and turned into a permanent check.
Pitching SEO and AI-search work to eight accounts meant eight discovery cycles. I ran one live outside-in scan across all eight in a day, for about 2,100 rank-data credits, and deployed nine client-ready proposals the same afternoon. Three of the eight ranked for zero keywords.
SEO delivery is normally a shared template and a shared team. Every client of mine runs in a dedicated environment with its own tooling and context layer. I built a hosted, ten-section systems tour to show a private equity firm's Head of AI how it works, with every figure dated and sourced and the unbuilt parts labeled. He probed the isolation model and didn't break it.
A Detroit-metro home-services brand consolidating two longstanding trade businesses under a new name became a full tenant of my Article Factory: its own isolated project, a deployed dashboard, and a complete eight-file context layer including who the business is not for. The build took one command and about fifteen minutes.
Nothing matches that yet.
