Six front doors,one department.

00 · The thesisNot six copies. Six experiments.

At first glance the six sites look like duplicates of one another — six domains, six landing pages, all promising an automated marketing department for a small business. They are not duplicates. They are six deliberate explorations of the same underlying idea, each testing a different way to make that idea land.

The idea itself is single and specific. A working marketing team has roughly seven roles — content, SEO, social, email, revenue, analytics, and creative. Hiring all seven is out of reach for almost any small business. So the premise is to run those seven roles as software on a schedule, working from the client's real data rather than headcount, with a human approval gate on anything that ships (email and money especially), and to offer the whole thing as a Managed service at $499/mo, flat (founding rate).

That premise never changes across the six. What changes is the front door: how the idea is framed, what proof it shows, which buyer psychology it speaks to. One site leads with a live audit. One draws the department as an engineering blueprint. One streams the work happening in real time. One frames it as the team you couldn't afford to hire. One writes it up as a long-form feature. One lays three designs side by side as a decision aid. Same room, six doors — an intentional positioning experiment to learn which entry point converts which kind of buyer.

The distinction that matters most, and the one this essay keeps returning to, is that five of the six share a common kit, while one of them is functionally distinct — it has a real, running backend that actually does the work the others describe.

01 · The shared foundationFive sites, one kit.

Being honest about this is important, because it is where a casual reviewer would be misled. Five of the six builds — Deptless, Deptzy, Auto Marketing Dept, the Automarketing long-form, and the AM Dept pitch — share the same underlying kit. That kit is the actual product surface a prospect can touch:

• the /marketingapp/ control panel — the department console
• a guided onboarding flow that takes a business from cold to configured
• a cost calculator that anchors the price against hiring or an agency
• a manual documenting how the department runs
• the full pitch → compare → finish funnel
• the CANTLOSE launch coupon

What differs between these five is framing, design language, and emphasis — not raw capability. Point the app at the same business through any of the five and you reach the same console with the same features. The blueprint site and the org-chart site are not two different products; they are two different arguments for the same product, dressed in two different visual systems. Treating them as five distinct capabilities would be a mistake. They are five distinct pitches over one shared engine of tooling.

The five are a positioning experiment. The differences are real and deliberate — but they live in the storytelling and the design, not in what the software can do.

02 · The one that's differentThe engine actually runs.

Auto Marketing Engine (automarketingengine.com) is the exception, and it is a real one. Where the other five describe an automated department, the engine operates one. It is backed by a live service — autoengine.service — that responds to a real API. Post a URL to /api/analyze and it fetches the live page, reads it, and returns a genuine audit: HTTP status, title and meta, headings, schema, social tags, sitemap and llms.txt presence, image alt coverage, internal and external link counts — then scores five disciplines (SEO, Content, AI Search, Social, Technical) and an overall number, each finding traced to a real signal on the page with a specific recommended fix.

That live audit is the seed. From it the engine stands up a workspace: a prioritized fix list, a 30/60/90 roadmap, a content calendar, deliverables with an approval step, a recurring engine that re-runs at midnight Central, a shareable per-site report, and an ROI calculator driven by the real audit data rather than by guesses. Because the backend is genuine, the engine can also do something none of the others can: score every site in a portfolio at once. It was load-tested across 245 owned domains and all 245 passed — a real stress test, not a mockup, and the strongest single piece of evidence that the machine works.

This is the honest dividing line: the engine is the only one of the six where the software actually performs the analysis it advertises. The other five are excellent front doors to a shared toolkit; the engine is the toolkit with a working brain attached.

03 · Deep diveAll six, one at a time.

For each build below: its approach, what it implements, its completeness and effectiveness, and — candidly — what it lacks. Nothing here is invented; each reflects what the live site actually is.

04 · Shared gapsWhat none of them do — yet.

Two honest limits apply across the set, and clients should hear them plainly.

The five non-engine sites all lack the intelligence layer. No live audit, no real-time site analysis, no network portfolio, no shareable per-site report. Those capabilities are flagship-only by design — they live on the engine and nowhere else. The five are front doors to the shared tooling; they don't carry the brain.

All six share the Managed-model limits. None of them publishes fully autonomously to a client's own accounts. None writes finished, bespoke, publish-ready copy end to end without a human. The pattern throughout is the same and it is deliberate: the software sees, scores, and scaffolds — briefs, drafts, fixes, calendars, recommendations — and a human reviews and ships the last mile. That is exactly why the $499/mo founding price is described as Managed rather than self-driving, and it is what keeps the offer honest.

The gate is not a missing feature. It is the product. "Automated department, human approval" is the promise — not "no humans."

05 · How they differA taxonomy.

The six vary along four axes: the buyer psychology they target, the proof mechanism they lean on, their design language, and where each is strongest.

Read down the "proof mechanism" column and the experiment is clear: a real audit versus a schematic versus a live feed versus an org chart versus an essay versus a comparison. Six ways to make one true thing believable — and a controlled way to learn which kind of evidence a given buyer actually trusts.

06 · The verdictCompleteness & effectiveness, ranked.

Judged on how completely each delivers the promised department and how effectively it converts, the order is not close at the top and closer in the middle.

1. The Engine (automarketingengine.com) — most complete, most effective, and the only one with a working brain. The real audit, the portfolio, the shareable report, and the ROI calculator built on real data make it the true flagship. Everything else is positioning; this is product.
2. Deptzy — Watch It Work Live — the strongest of the five kit-sites, because its proof (a genuine activity feed) is the closest the non-engine group gets to showing real work.
3. Deptless — the Blueprint — the best explanation and the strongest design identity; the clearest answer to "what is this?"
4. Auto Marketing Dept — the Team — the most relatable and the cleanest cost argument for a small-business buyer.
5. Automarketing — the Long-Form — the deepest and best for trust and search, but slow and low on demo value.
6. AM Dept — the Pitch — useful as a decision aid and as a record of the experiment, but the least differentiated now that it's genericized.

The middle four are genuinely close and, importantly, they are ranked as positioning, not as capability — under the hood they are the same kit. Their order would shift with the buyer in the room. The point of the verdict is not to retire five sites; it is to be clear-eyed that one of them is the engine and five of them are arguments for it.

07 · A test procedureProving it works — not just that it runs.

The 245/245 QA run proved the engine runs on every owned domain: it reaches the site, scores it, and returns a workspace without failing. That is function, not effectiveness. Whether it actually works — accurate audits, relevant deliverables, measurable lift — is a separate and harder question, and it deserves a real experiment rather than an assertion. What follows is a pre-registered procedure to gather that evidence. It has not been run; it is the design.

1 · The sample

Draw a stratified random sample of roughly 54 sites from the two networks' combined ~240-site frame — the engine's own network worksheet already scores all 240, so the frame is real and enumerable. Stratify on two axes. First by cluster / niche: Austin, tech/AI, real estate & Hot Springs, sports, TV/media, coworking, marketing, and the WholeVoyage travel/villa cluster. Second by traffic tier — high / mid / low — read from the Alltime AWStats hits already recorded in the alldomains worksheet, so no site's weight is guessed. Within each stratum, split the sites roughly in half into a treatment group (department actively run) and a control group (audited only, left untouched), so any change can be attributed to the intervention rather than to the calendar.

n ≈ 54 · each stratum split ~50/50 into treatment and control · every named site verified present in the engine's live network worksheet, 2026-07-04.

2 · The procedure, phased

• Baseline (week 0). For every sample site, record the engine's overall score plus the five sub-scores and the underlying signals via POST /api/analyze, alongside current traffic from AWStats — all timestamped. Reuse the existing /qa/ and /api/network tooling rather than building anything new.
• Intervention (weeks 1–8, treatment only). Activate the department, let the recurring midnight-CT engine produce deliverables, and — because the honest limit is the human last mile — actually ship a fixed cadence of approved work: for example, two content pieces, three SEO fixes, and two social drafts per site per week. Control sites are audited on the same schedule but receive no intervention.
• Measurement (weeks 0 / 2 / 4 / 8). Re-run the audit for readiness deltas, pull AWStats traffic deltas, and track indexed-page counts and Search Console impressions where access exists.
• Deliverable-quality check. Pull a random sample of generated briefs, SEO fixes, and social drafts and have blind human raters score each on a rubric — relevance, factual accuracy, usability — because output quality, not reach, is the real open question.

3 · Metrics & success criteria

Five measures, fixed in advance: audit accuracy (spot-check a set of findings against the live page), deliverable-relevance rating (the blind rubric score), readiness improvement (treatment vs control), traffic lift (treatment vs control), and reach / error rate. A positive result is stated unambiguously before any data is collected: treatment sites show a meaningful readiness gain and traffic lift over control, and the blind raters judge the deliverables usable. A null or negative result — no lift beyond control, or deliverables rated thin — is reported exactly as found.

4 · Rigor

Pre-register these criteria before baseline so the bar cannot move afterward. Blind the quality raters to which group a deliverable came from. Use the control group and the traffic tiers to isolate the engine's effect and to control for seasonality and pre-existing momentum — a busy site rising in a busy month proves nothing without its untouched twin. The sample is a deliberate fraction of the 240+ sites across both networks: large enough for signal, small enough to run honestly without conscripting the whole network.

This section describes a plan to gather evidence. It reports none. The 245/245 run answered "does it run?"; this answers "does it work?" — and until it is run, that question stays open on purpose.

08 · Play-by-play & resultsThe build, and the evidence so far.

This section is the running record — how the six were built, and what the two evaluation phases have shown to date. It is deliberately incomplete: the effectiveness verdict is still open, because the intervention and its week-2/4/8 measurements have not yet run. What is here is real; what is pending is marked as pending.

A · The build, in order

The six implementations were stood up over a single working session. Early on, four of them were near-clones of the shared kit — same console, same funnel, different paint. Rather than ship four look-alikes, each was deliberately differentiated by approach: the data-first engine, the systems-first blueprint, the hiring-first team you hired, the editorial long-form, and the comparison pitch.

Midway through, a closer look caught that one site — deptzy.com — was a near-exact copy of the blueprint, not a distinct build at all. It was rebuilt from scratch into its own identity: watch it work live, an observability-first ops feed streaming the real engine's own activity rather than restating the blueprint's argument. A positioning pass followed. The negative framing — the "nobody," "no-people" language that sold the idea as an absence — was flipped to the active voice the product actually earns: "staffs itself," "running every day." And the comparison pitch, which had shipped with real client names exposed on the page, was genericized so it reads as options, not as a named proposal. With the set coherent, two evaluation phases were run.

B · Results so far

Functional QA — does it run? The engine was run against all 245 owned domains. The result: 245/245 analyzed, zero hard failures, zero HTTP 500s. The run was no rubber stamp — it surfaced two real defects and fixed them. First, a str.lstrip("www.") call was corrupting every domain that begins with a "w": lstrip strips any leading w/. characters, not the literal www. prefix — so wholetech.com was being stored and looked up as holetech.com, walhus.com as alhus.com. Second, seven redirect-alias reports were returning 404. Both were fixed and the suite re-ran to zero failures; all 245 reports now return 200 with real branded content. See the QA report →

Week-0 baseline — the starting line. With the machine proven to run, the experiment's baseline was frozen. From a frame of 245 live sites, a stratified sample of n=54 — 27 treatment, 27 control — was drawn across eight clusters: Real estate & Hot Springs (12), TV/media (10), Tech/AI (8), WholeVoyage travel/villa/film (6), Coworking (6), Austin local (4), Marketing (4), Sports (4) — traffic-tiered on all-time AWStats hits with a deterministic 50/50 split. The arms start balanced, which is the whole point of a baseline: mean overall readiness T=89.1 vs C=89.3 (a 0.2-point gap); mean all-time hits T=5,692 vs C=4,916; mean last-month hits T=2,481 vs C=1,998; the score distribution runs 90–100 = 28 sites, 80–89 = 26. Frozen 2026-07-04. See the baseline → · raw baseline.json

What this proves — and what it doesn't

Together, these two phases establish that the engine runs reliably at network scale and that the experiment is set up correctly, with two well-matched arms. They do not yet prove effectiveness. That claim requires the weeks 1–8 intervention on the treatment arm and the week-2/4/8 measurements against the untouched control. Those results will be added to this section as they arrive.

The honest status, stated plainly: the machine works, and the experiment is properly built. Whether the department moves the numbers is still being measured — not yet claimed.

09 · Why sixAnd what to do with them.

Six sites for one product is not indecision — it is an A/B test across buyer types. Different buyers trust different evidence. The engineer trusts a schematic; the skeptic trusts a live audit; the owner-operator trusts a cost comparison; the researcher trusts a long read; the doubter trusts watching it run; the deliberate buyer trusts seeing the options laid out. Rather than guess which framing wins, all six run at once, each a clean instrument for measuring which door a given buyer walks through.

The recommendation is straightforward. Treat the Engine as the flagship and the destination — it is the only one that proves itself in software, and every other site should funnel a warmed-up prospect toward it. Keep the five kit-sites as specialized front doors, each pointed at the buyer psychology it serves best: send the systems thinker to the Blueprint, the skeptic to the Live feed, the cost-conscious owner to the Team, the researcher to the Long-Form, and the comparison shopper to the Pitch — then hand them all to the Engine to close. And keep the pricing description exactly as honest as it is: a Managed department at a founding $499/mo, where the software does the seeing and scaffolding and a human ships the last mile. That honesty is not a weakness in the pitch. It is the reason the pitch holds up.