# Rejected by His Own Gate: How Mike Khaytman's Codex pipeline runs SEO content for six personal sites

Mike Khaytman, an SEO lead with 13 years in the field, runs the SEO content for his personal sites through a small pipeline of scheduled scripts that call Codex. It researches keywords and trends, writes a sourced brief, drafts, runs every draft through a quality gate and then publishes and deploys. Since July 2026 it has published about 70 articles across six sites. Four lower-stakes sites publish automatically and two queue drafts for his review. The gate is the part he trusts: when he ran an older backlog of articles written from one generic prompt through it, it rejected all 68, while articles the pipeline wrote with sources gathered first passed.

**Try this step:** <strong>Split the work into five stages.</strong> Research pulls trend signals and keyword data, scores candidates and dedupes against what each site already covers. Brief turns a scored opportunity into a sourced brief. Draft writes from the brief. Gate checks for unsourced claims, derivativeness, staleness and guardrail violations. Publish commits the post and deploys the site.


| | |
|---|---|
| **Who** | Mike Khaytman, SEO Lead at Bonsai Media Group, Fort Lauderdale (mikekhaytman.com). LinkedIn in/mike-khaytman and GitHub mkhaytman87, both linked from his site, verified 2026-11-20 |
| **Problem** | Closing the gap between finding an SEO opportunity and publishing on it across several small sites, without publishing unsourced or derivative articles |
| **Stack** | Codex (codex exec) for briefs, drafts and gate scoring, DataForSEO API plus free trend sources, Python and SQLite, scheduled jobs on his own machine, Astro sites on Cloudflare Pages; a separate personal agent, Kai, on OpenClaw |
| **Confidence** | Medium (first-person, his own counts; no traffic, ranking or revenue results for the sites) |
| **Last verified** | 2026-11-20 |

## The problem
He writes that the hard part of content SEO "is usually the gap between finding an opportunity and acting on it. Research gets done, the list sits in a tab, and nothing gets published." He wanted the easy-to-postpone parts done every day (API calls, data compilation, cross-referencing, first drafts) and to keep the judgment calls: what each site is about, what the gate rejects and which sites get human review.

## The blueprint (as he describes it)
1. **Split the work into five stages.** Research pulls trend signals and keyword data, scores candidates and dedupes against what each site already covers. Brief turns a scored opportunity into a sourced brief. Draft writes from the brief. Gate checks for unsourced claims, derivativeness, staleness and guardrail violations. Publish commits the post and deploys the site.
2. **Separate research from publishing in time.** A morning job does research only and publishes nothing. A later job writes, gates and publishes, capped by each site's weekly cadence. A Sunday job reconciles everything and emails him a digest.
3. **Decide autonomy per site.** Four lower-stakes sites publish automatically. The two higher-stakes sites queue drafts for his review.
4. **Keep the machinery small.** There is no LangChain or similar framework. A small set of scripts calls the model with instructions, validates the output and publishes it. Codex runs through codex exec, and Python and SQLite hold the pipeline's memory of what has been researched and published.
5. **Filter trend feeds by niche.** Unfiltered, the top "AI marketing" candidates were unrelated breaking news. The scorer now applies a niche allowlist and needs repeated observations before it trusts a trend.
6. **Verify state, not output.** For a full day the deploy log and the push command disagreed about whether pushes had worked, because the check read command output. The pipeline now verifies against the remote.
7. **Treat infrastructure bugs as SEO bugs.** One site had no 404 route, so Cloudflare Pages answered every unknown URL with the homepage and a 200 status. The pipeline work surfaced it, and it is now a standard check.

## Stated results
- **Output:** about 70 articles across six sites since July 2026.
- **The backlog test:** an older setup had written a backlog from "a single generic prompt." The gate rejected all 68 for unsourced claims, derivativeness and staleness. The backlog became "a list of topics to rewrite properly instead of content to publish."
- **Setup time:** "a first version takes a weekend. Making it trustworthy takes longer".
- **Traffic, rankings or revenue for the six sites:** None stated.

## Caveats
- The six sites are not named, so their output and traffic can't be checked.
- The pipeline is not connected to Search Console yet, so in his words it "doesn't learn from what ranks." AI citation tracking is not built.
- He says a human SEO could do any single step better. The pipeline's advantage is consistency.
- The 97.78% first-pass accuracy figure in the same post (14 errors in 633 claims, "reported as an upper bound") comes from a content engine he designed at his agency for client work, not from this pipeline.
- His homepage invites hiring conversations for AI enablement, SEO or AI search work, and says he is open to advisory work and select projects.
- Handles: LinkedIn in/mike-khaytman and GitHub mkhaytman87 (both linked from mikekhaytman.com). No verified X, Bluesky, Threads, Instagram or TikTok profile was found.

## FAQ
### What does Mike Khaytman use to write the articles?
Codex, called through codex exec, for briefs, drafts and gate scoring. DataForSEO and free trend sources feed the research stage, and Python scripts with SQLite run the pipeline on scheduled jobs.

### Does every article publish without review?
No. Four lower-stakes sites publish automatically. The two higher-stakes sites queue drafts for his review.

### What does the gate check?
Unsourced claims, derivativeness, staleness and guardrail violations. It rejected all 68 articles in an older backlog written from one generic prompt.

### Has the pipeline grown traffic?
None stated. It is not connected to Search Console yet.

## Sources
- https://mikekhaytman.com/blog/articles/ai-agent-seo-workflow (primary: "How I built an AI agent that runs my SEO workflow", Feb 28, 2026, updated October 2026)
- https://mikekhaytman.com/ (about and experience)

---
*Want agents like these working your distribution? [Get It Built](/get-it-built). Get new teardowns weekly: [subscribe](/subscribe) (double opt-in).*