AI Marketing Content Generator for Agencies — SoloCMO
Content generation · 7 page types

Start it. Walk away. Come back to a finished page.

Eleven passes take an article from live research to a finished page — researched, written, edited five times over, linked, tagged, illustrated and scored. The whole run takes 13 to 17 minutes, and you don’t sit through any of it.

Three hard gates stop anything over the word ceiling or outside its keyword ranges from moving forward — so what’s waiting when you get back is a page, not a draft you have to audit.

Finished articleapp.solocmo.io
Screenshot The finished article in the editor, images placed 16:9 · export 2000 × 1125 (@2x)

Lead with the output, not the interface. Anonymize client name, logo and phone number first.

The problem

The problem with AI writing isn't quality. It's that nothing checks it.

Ask any model for a 1,200-word article and you'll often get 2,600. Ask it to use a keyword eight to twelve times and it'll use it forty. Ask for local specificity and you'll get the city name dropped into generic sentences. The output reads fine, which is exactly the problem — you have to read the whole thing carefully to find out it isn't publishable.

At fifteen clients, that check doesn't happen. So the article ships long, over-optimized and vague, and three months later nobody can work out why the content program isn't moving anything.

The real question

"Couldn't I just do this in ChatGPT?"

Yes — if you supply the brief, the research, the keyword ranges, the internal links, the schema, the QA and the photography yourself, every time, for every client. The model was never the hard part. The process around it is.

A chat window

You are the pipeline

  • No memory of who this client's customer is
  • Nothing counts your keyword frequencies
  • No word ceiling — it will happily hand you 2,600
  • You write the brief before you can ask for anything good
  • Internal links, schema and alt text are separate jobs you'll skip
  • Nothing compares the finished piece back to the brief
  • No live score, no readability check, no keyword tracking as you edit
  • Stock photos, or a generic AI stranger in a generic room

SoloCMO

The pipeline is the product

  • Audience and selling points inherited from the client's strategy work
  • Every term checked against an imported frequency range
  • A word ceiling that blocks, with the overage counted
  • Pass one builds the brief from live cited research
  • Links, schema, meta and images are passes, not afterthoughts
  • A brief-alignment gate and a scored final QA
  • An editor that scores readability and keywords live as you type
  • Images built from the client's own staff, premises and branding
The pipeline

Seven phases. Eleven passes.

Each phase does one job and hands off. Drafting is five sequential editing passes on its own — cadence, AI cleanup, bookends and conversion — because one pass produces a draft and five produce something a person would actually read.

You don't sit through them. Set the variables, start the run, and the pipeline works in the background — pass by pass, or all of them from wherever you are. A full article takes 13 to 17 minutes and tells you when it's done. Every one is saved as a project you can reopen, hand to a second seat, or re-run a single pass on.

01

Research

1 pass

02

Outline

1 pass

03

Drafting

5 passes

04

Internal links

1 pass

05

SEO tags

1 pass

06

Images

1 pass

07

QA + final

1 pass
Why it's different

The gates are the product.

Between passes, the pipeline audits its own work and refuses to move forward when the draft misses spec. Not a warning you can scroll past — a stop, with the exact numbers and what to trim.

You can override it. The point is that overriding becomes a decision you make, instead of something that happens because nobody counted.

Keyword gate failed 13 terms over limit
Word count 2,677 against a 1,200 target — over ceiling, must cut
hot water extractiontarget 5–12×42×
carpettarget 14–36×45×
steam cleantarget 5–11×30×
staintarget 1–6×

A real gate failure. Three routes out: re-run the pass, fix only the flagged issues, or override and proceed anyway.

Gate 01

Keyword frequency

Every term checked against the range you imported from NeuronWriter or POP, with the overage counted per keyword.

Gate 02

Word-count ceiling

A target with an allowed band around it. Over the ceiling is a stop, not a note — because running long is the failure mode every model shares.

Gate 03

Brief alignment

The finished draft compared back against the pass-one brief: must-cover questions, local specificity, conversion presence.

The full pipeline is on every plan. No feature gates, no upsells, no separate optimizer subscription.

Pass one

It researches before it writes, and shows you the sources.

The first pass builds a full content brief from live web research — dozens of cited sources for a single article — and it isn't a keyword list. It's search intent and the actual questions being asked, audience psychology in real customer language, what top-ranking pages already do, and what this article has to do better.

Plus the local angles: humidity, construction age, neighborhood names, the seasonal conditions that make a generic article obviously generic.

  • Search intent with the specific questions to answer
  • Fears, dream outcomes, trust language and objections, quoted from forums and reviews
  • Keyword coverage plan with target frequency and best placement per term
  • Competitor angle summary — what they do, what this has to beat
  • Structure map with a word budget per section
  • Every source cited and listed, with source attribution visible in the draft
  • Or skip it — supply your own research from client interviews or proprietary data, and the pipeline builds on that instead
Research synthesisPass 01
Screenshot Research brief — coverage plan and sources 16:10 · export 1600 × 1000 (@2x)
Fill once

The brief is a form, and half of it is already answered.

Article type, strategy type, word count, target city, keyword frequencies, formatting rules, the phone number and the call to action — set once, and every one of the eleven passes uses them.

The fields that decide whether the article sounds like this business rather than its category come straight from the strategy work: target audience links to an avatar, and the selling points are pulled from the USP. You don't rewrite them per article.

  • Seven article types and six strategy types, including hyperlocal and comparison
  • Keyword frequencies imported from NeuronWriter or POP
  • A negative prompt — topics, claims and angles to exclude from this specific article
  • Formatting rules down to whether bold, italics and blockquotes are allowed
  • Reading level inferred from the avatar
Article variablesFill once
Screenshot Variables form with avatar and USP linked 16:10 · export 1600 × 1000 (@2x)
Pass six

And the photos are their actual staff.

Every other pass has a competitor somewhere. This one doesn't. Upload the owner and team once at onboarding, plus a few shots of the premises, the vehicle and the logo — and from then on article images are generated as scenes featuring those real people, in that real space, with their signage where it actually hangs.

It matters because "send me photos of your team" is where content programs stall. You get three blurry shots from 2019, or nothing, so the article ships with stock — and stock is the fastest way to tell a homeowner a business might not be real.

Headshot

Interior

Vehicle

Logo

GeneratedArticle image
Example The same person, the same room, the same brand 4:3 · export 1200 × 900 (@2x)

Recognizable

The person a customer will meet

Not a lookalike and not a stock model — the actual staff member, close enough that a returning customer recognizes them.

On location

Their room, their truck, their sign

The real interior, the branded vehicle, the product wall exactly where it hangs — the details a prospect uses to decide whether a business is real.

Once, then forever

One upload covers every article

Ask at onboarding. Every article after that gets illustrated without going back to the client for anything, with alt text and composition notes written per image.

What else ships with it

The article isn't the whole deliverable.

Two more passes produce the things that usually get skipped because they're fiddly — and skipping them is why so much published content underperforms the work that went into it.

Internal linksPass 04
Screenshot Links with placement notes 4:3 · export 1200 × 900 (@2x)

Pass 04

Internal links, placed

Anchor text and destination for each link, plus where in the article it goes — including a money-page link inside the first hundred words.

Meta & schemaPass 05
Screenshot Meta tags and JSON-LD 4:3 · export 1200 × 900 (@2x)

Pass 05

Tags and schema

Meta title and description inside their character limits, and a full JSON-LD graph — article, FAQ and local business — ready to paste.

Content diagnosisPass 11
Screenshot Content diagnosis with issues and keyword audit 4:3 · export 1200 × 900 (@2x)

Pass 11

It scores its own work

Graded out of ten across keyword, structure and conversion, with issues split into critical and moderate and a fix for each. It will tell you an article it just wrote scores 7.3 — a tool that always says the work is good isn't checking.

The last mile

Every optimizer tells you to use a term three more times. This one puts it where it belongs.

The QA pass tells you the article scores 7.3 and names the issues. The built-in editor is where you resolve them — with a live optimization score, live readability, and every required keyword tracked against its target range as you type.

Click an under-target keyword and it's inserted at the best place in the article. No exporting to a separate optimizer, no pasting into a readability checker, no hunting for a natural spot.

  • Live optimization score and Flesch readability with a grade level
  • Sentence-level highlighting — very hard, hard, and hedging language flagged in place
  • Select any passage and simplify it, rewrite it in plain English, punch it up, or set a specific grade level
  • Required keywords listed with live counts, color-coded on target, under, or over
  • Click a keyword to place it where it fits, rather than being told to find room
  • Live outline with a word count on every heading, so you can see where the length actually is
Content editorLive scoring
Screenshot Editor with score, readability and keyword panel 16:9 · export 2000 × 1125 (@2x)

Capture with readability highlighting on — the colored sentences are the proof it's actually analyzing.

Proof

One client, one quarter, one number that moved.

Process only matters if it produces something. Here is what the pipeline did for a single local service client, measured from their own Search Console rather than from ours.

[FILL] vertical [FILL] market [FILL] months

Where they started

[FILL: what the site looked like before — how many pages, what was ranking, what the owner had already tried, why content had stalled. The photo request that never got answered belongs here if it applies.]

What ran

[FILL: how many articles, over what period, which page types and strategy types, and whether the strategy pipeline ran first. Name the modules so a reader can connect this back to the system.]

What changed

[FILL: the outcome in the client’s language, not marketing language. What the owner noticed before the analytics did.]

[FILL]

Ranking terms gained

[FILL: window and source, e.g. first-page terms, 90 days, Search Console]

[FILL]

Clicks per month

[FILL: before and after, same date-range length]

[FILL]

Articles published

[FILL: production hours before vs. after, if you have them]

[FILL: one or two sentences from the client, in their own words. A specific, slightly awkward quote beats a polished one — “I stopped getting asked for photos” is worth more than “great results.”]

[FILL] · [FILL: role, business, city]

Every article in this case study came out of the same eleven passes and the same three gates.

The arithmetic

Count what an article actually costs you.

Not the writing. The brief, the research, the outline, the editing rounds, the internal links, the meta and schema, the photo request that never gets answered, and the read-through where you confirm it's on spec. Done properly, that's most of a working day.

The pipeline runs it in 13 to 17 minutes, and none of those minutes are yours. The question was never whether AI can write. It's what happens to that day, multiplied by every article on every client's calendar.

15 min

Unattended, per article

13 to 17 minutes from brief to finished page, running in the background while you do something else.

11

Passes per article

Research, outline, five drafting passes, links, tags, images, QA — each one a job you'd otherwise do yourself or skip.

3

Gates that block

Keyword frequency, word ceiling, brief alignment. The checks nobody runs at fifteen clients.

1

Photo request, ever

Ask at onboarding. Every article after that is illustrated with their staff without asking again.

Run one of your own clients through it and see what comes out the other end.

Questions

Content generation, answered straight.

How long does a full run take?
13 to 17 minutes for all eleven passes, and it runs in the background — you start it and go do something else. Filling the variables takes a few minutes at the front, and reading the finished article takes a few at the back. The eleven passes in between are not your time.
Why not just use ChatGPT or Claude directly?
Because a chat window has no memory of your client, no keyword targets, no images of their staff, and nothing that counts. You can get a good article out of any model if you supply the brief, the research, the frequency ranges, the internal links, the schema and the QA yourself. This is that process, run the same way every time, for every client, without you holding it together.
What happens when a draft fails a gate?
The pipeline stops and tells you exactly what's wrong: which keywords are over their range and by how much, and how far the word count is above the ceiling. You can re-run the pass, fix only the flagged issues, or override and proceed. Blocking is the default; overriding is the exception.
Isn't AI content risky for rankings?
Thin, unresearched, over-optimized content is risky. That's what the gates exist to stop. Every article is built from live cited research, held to a keyword range rather than pushed past it, checked back against the original brief for local specificity, and scored before it leaves.
How much editing will I still do?
You'll still read it. The five drafting passes exist so that what you read is a finished article rather than a first draft, and the fix-issues action resolves the flagged problems without regenerating the whole piece. When you do want to change something, the built-in editor scores readability and keyword coverage live, rewrites any passage to a chosen reading level, and places an under-target keyword for you rather than telling you to find room for it.
What is this not built for?
Local service search intent. It builds service, location, industry and comparison pages that answer what a homeowner types before they call, and it holds them to a keyword plan and a word ceiling. It is not the tool for thought leadership, original reporting, or anything whose value comes from a point of view no research pass can find. If that is the work, write it yourself.
Where do the article images come from?
From the client's own assets. Upload a headshot or two of the owner and team plus some shots of the premises, vehicle and logo, and article images are generated as scenes featuring those actual people in that actual space with their real branding — instead of a stock technician.
Get started

Publishable, not plausible.

Eleven passes, three gates, and a scored article at the end. Included on every plan — plans differ only in how many clients you run and how much you generate for them.