The AI Factory build log

Objections, mined and answered

The Customer Factory now collects the reasons buyers hesitate, from site copy, on-site behavior, public reviews, owned channels and competitors' reviews. The owner approves the list. Retargeting ads then answer each objection directly and are tested against a proven winning ad as the baseline.

Competitors' customers as a research source

New research sources include review sites, browser-extension store reviews, YouTube comments, competitors' FAQ and pricing pages, and competitors' live ad copy, each matched to the right competitor by exact name.

The loop refills itself

Buyer research now runs weekly, angles that spent money without converting at any stage are retired, and the judge scores each ad against the page it will land on. A weekly read-only check confirms every stage of the loop actually ran: the direct answer to the audit three weeks ago.

Outcome: 340+ articles, six sites at three a week

Since April, the Content Factory has delivered more than 340 articles across eight sites, and six sites now publish three a week. Every article published since August has passed every blocking check, and the engine's own test library has grown to more than 100 known-good and known-bad articles. If you want to see the output, many of this site's Insights articles come from it, including this week's SEO content marketing system for small businesses.

Revenue means money collected

The Customer Factory now counts revenue only from invoices that actually collected payment, and cost per result uses only ad-attributed results. A raise or pause is reported only after Meta accepts it, implausible rows from the ad platform are rejected, and stored history can be restored.

One compliance check before anything reaches Meta

Every ad, lead-form text and retargeting ad now goes through the same compliance check before it publishes. When a required line is missing, the system adds it instead of throwing the ad away.

A second look for under-tested ads

Strong ads that never got enough spend to prove themselves are retested in a fresh ad set instead of being killed, and no ad is killed before a minimum spend. Adopting an experiment winner now shapes new creative for the next 30 days.

Where buyers drop off

The dashboard's acquisition view now shows the paid funnel with the stage where buyers drop off, every ad in flight, and which traits go with results, and it exports to a summary you can paste straight into an AI chat.

Fact-checking on big government tables

The Content Factory can now verify figures in very large statistics tables, including dollar figures printed without a dollar sign. Before, it failed safe and blocked good, data-heavy articles it couldn't check.

A new kind of client: a local service business

The Content Factory onboarded its first local service business, a real estate team. Local search works differently from national search, so the engine changed with it.

Local authority sources

For local tax and closing-cost questions, the engine now searches state and county authorities first, so a question like "who pays transfer tax?" is answered from the agency that sets it.

Links into the pages that matter

The engine now reads sites built on Astro as well as Next.js, so links to neighborhood and service pages are checked against pages that exist and offered to the writer as link targets.

Ad-policy compliance, every night

The Customer Factory now runs a rule-based checker for Meta's financial-services policy on copy, on image checks and inside the judge. Each night, any disapproved ad is detected, taken out of rotation and reported in the standup, and any test it was part of is refilled. Pages that send visitors straight to checkout are now judged on checkouts and trials rather than clicks.

A creative gate before the owner's inbox

Most drafts were being rejected by hand, which meant the owner was doing the judging. Each ad idea in the Customer Factory now goes through brief, copy, scoring by a judge that sees real performance evidence, and up to three rewrites. Only drafts worth testing reach the owner, with the reasoning on the card. The judge learns from past approvals and is recalibrated weekly against actual results.

Outcome: about 450 ad creatives

Across the businesses it runs for, the Customer Factory had generated about 450 ad creatives by early October, in ten formats, including a new one built from real product screenshots.

Briefs aimed at one search

Each Content Factory brief now carries the target keyword and its intent, where the keyword has to appear, the format Google currently ranks for that search, and the competing page-one articles, read before writing. Topics where page one holds fewer than three articles are dropped, because a blog post can't win them.

Weekly indexing checks, monthly ranking checks

Every week the engine checks Google Search Console for each article: indexed or not (and why), whether Google chose the right canonical page, sitemap status, clicks, impressions and position. Every month it re-checks rankings, queues refreshes for articles sitting just off page one, and leaves top-three articles alone.

Statistics traced to the original study

When a page repeats a figure from a named study, survey or regulator, the engine looks for the original publication, and blocks the figure if it's only cited to the page repeating it.

Plans built from searches a site can win

The Content Factory now builds each six-month calendar from keyword volume, difficulty, the site's current rankings and competitors' content gaps. It keeps only searches the site can realistically win at its current authority, one article per search. The plan arrives for review, and approving it is the go-ahead.

No two articles compete for the same results

A planned article is dropped if four or more of its top-ten Google results already match an existing post's. Briefs start from what Google shows today: pages in the AI Overview, the featured snippet and the organic results are tried first as sources, through the same checks as anything else.

Failed drafts get repaired, and retries learn

When a draft fails a check, the engine edits only the sentences that failed and re-checks the whole article. If it still fails, the next attempt's brief lists exactly what was rejected last time.

Missions and a daily standup

Each business on the Customer Factory now has a stated mission and a playbook for its stage. A daily standup reports what the system did, what it's doing, what's blocked, and what needs a decision.

An experiment engine

The system runs one-variable experiments with one ad set per variant and reads results statistically. A weekly planner proposes the next tests, formats are chosen by past performance, near-duplicate ads are blocked and worn-out angles are retired.

An owner dashboard that answers five questions

Are we making money? Where are we wasting it? What are customers saying? What should we stop? What should we do next? The dashboard sits behind sign-in, works on a phone, and runs on a fictional demo brand so prospective clients can see it working before they commit.

Money moves to what works

When the daily cap blocks a raise, the Customer Factory now trims the weakest ad sets to fund the winners. Paused ads leave the ranking pool, and budget changes are paced to stay within Meta's limits.

Leads are what predict a customer

Each business can now define which deeper actions count as a lead (finishing a free tool, generating a result, signing up for a guide), so the optimizer judges ads on behavior that predicts a customer rather than on clicks.

Every format is tracked

Every creative records its format, and the digest shows which formats earn their share of spend. A written playbook now governs how new creative experiments are run.

Budget decisions made daily, inside the cap

The Customer Factory moved from recommending changes to making them. It pauses losing ads and raises budgets on winners, always within the daily cap. It waits for a minimum volume of results before deciding anything, and reads live budgets so it never acts on stale numbers.

This is the step most "AI marketing" stops short of, and the one that replaces the monthly check-in with an agency: the decision happens the day the data says so.

Lead-form campaigns

The system can now run Meta instant-form lead campaigns: it creates the form, seeds the ads from existing approved creative, and syncs new leads every hour.

One engine, any brand

The Customer Factory is now brand-agnostic, with a written onboarding playbook and a site template. A new business gets its own configuration and data instead of a copy of the code.

Outcome: live in a day

A second business, in a completely different category from the first, went from configuration to live ads in a single day, with browser and server-side conversion tracking. Onboarding is now a playbook we follow, not a project we reinvent.

Guardrails for regulated categories

Copy rules can now ban whole classes of phrasing for a category, starting with trade-instruction language for financial products.

Meanwhile, in the Content Factory

The Content Factory delivered 23 articles across seven sites, a mix of new posts and the refreshes that started running last week.

Refreshes that keep an article's search history

The Content Factory now runs the refreshes in every calendar. A refresh rewrites an existing article at its existing URL, so the rankings and links it has already earned stay attached, and the update is built cleanly against the site's current code.

Ads go live through a guarded switch

The Customer Factory switched on its first live ads. The go-live step turns on only ads that have been checked and approved, and nothing else.

A brand memory the system has to follow

Positioning, founder voice and customer-truth notes are stored per brand, and written rules ("never do this in an ad") are applied to every generation. Each week, the owner's rejections become proposed new rules, so the system learns taste from decisions rather than from a style guide nobody reads.

Market sensing and a weekly truth memo

Watchers scan live web and social search for the category and save dated digests. A weekly memo says plainly what's converting, what only looks good, and what hasn't been proven yet. Landing pages can also be split-tested 50/50 now, and the daily digest shows a per-ad scoreboard.

The calendar is checked against the live site

Before and after every run, the Content Factory compares its plan with what's actually published and repairs any mismatch in either direction. The plan a client approved and the articles on their site are now provably the same list.

One rejected draft no longer costs a post

Every site now gets up to three attempts per run. Before, a single rejected draft meant that site simply went without a post that day, and sites late in the rotation could be skipped altogether.

On-page SEO is checked at delivery

FAQ sections are emitted once, in the format the site's template reads, so FAQ structured data renders correctly in search. Every internal link has to point to a page that exists, and each article targets four to six of them.

Exact on numbers, fair on paraphrase

The fact-checker now reads a focused excerpt around each figure instead of a whole long document. In very long documents it could miss a true claim and reject it. Figures and quotes stay strict, and a faithful paraphrase of a named source is no longer refused.

Claims and opinions are treated differently

The Content Factory now classifies each sentence as either a verifiable claim (a figure, a quote, an "according to") or interpretation. Claims go to an independent AI judge that has to quote the exact passage that supports them. Analysis stays the writer's own and is never dressed up with a citation it doesn't need.

Primary sources come first

For every niche, the engine searches a curated list of authoritative publishers before anything else. Sources too old to support a current-year article are rejected.

Every article ships with its own hero image

Each post gets a generated hero image that is checked, retried if it fails, and required before the article can go out.

Back on schedule

All seven content calendars resumed on the stricter brief this week. It's the standard we'd want from any generative engine optimization service: if an AI assistant is going to cite you, every number had better hold up.

Sources first, writing second

The Content Factory now collects a "source pack" of real pages before it writes a word. The writer may cite only those stored pages, so every figure points to something the engine actually read, not something a model remembered.

A quality gate that blocks instead of warning

Every draft now passes automated checks before it can be published: citations resolve, each number traces to a source, the source supports the claim, quotes are exact, and internal links point to real pages. Any failure stops the post. None of the checks is advisory.

Checks that can't quietly get worse

A library of known-good and known-bad articles runs on every change to the engine, so a new rule can't loosen an old one without anyone noticing. Publishing paused briefly while the gate was built, which we'll take over publishing something we can't stand behind.

Ad copy competes before images are made

The Customer Factory now writes twice the copy it needs. A direct-response judge scores the variants and only the top half get images, so design effort goes to the strongest ideas. It also reads the live landing page and scores each ad on whether its promise matches what the visitor will see.

Every stage of a paid-ads loop, wired together

The Customer Factory is now one connected system rather than a set of scripts. It researches buyers, writes copy, makes images, sends ads for review, publishes, measures and optimizes. Each business it runs for gets its own configuration and its own data warehouse, so one engine can serve several brands without their data mixing. (If you want the idea in one page: how a closed-loop paid ads system works.)

Spend tied to revenue from day one

Ad results from Meta and payments from Stripe sync into the warehouse every day, and every ad is tied to the sign-ups and payments it produced. "Is this ad making money?" now has a direct answer instead of a guess.

Creative that starts from what buyers actually say

A research step reads what buyers say online and ranks their pain points. Copy is written from those pain points and checked for claims the business can't back up. Images get an AI vision check for readability and brand fit before a person ever sees them.

A human approves every ad, and spending has a hard ceiling

Each ad arrives in Slack as a review card for a one-tap approve or reject. Approved ads publish paused first. A daily spend cap is enforced in code before any change reaches Meta, and a kill switch pauses everything at once.

Meanwhile, in the Content Factory

The Content Factory published on a Monday, Wednesday and Friday schedule across seven sites with no manual work: 15 articles this week.