AI-assisted workflows · AI-assisted production · Sydney

AI-assisted production

The interesting question stopped being whether to use these tools. It is which parts of the work they should touch, and that answer is narrower and more specific than either the enthusiasts or the sceptics suggest.

Grab the snowball. Give it a spin, and watch it grow as you scroll. Just like the output.

6stages in a content workflow
2where a person is not optional
Named authoron everything published
Disclosedwhere it matters, honestly

The short answer

What does AI-assisted production actually mean here?

AI-assisted production means using tools for the parts of content work that are structured and low-judgement, and keeping people on the parts that are neither. Research gathering, outlining, first-draft scaffolds, format conversion and checking are assisted. Deciding what to say, saying it, and standing behind it are not.

That division is not a compromise. It follows from what the tools are good at. They are genuinely excellent at speed, coverage and structure, and genuinely poor at knowing which of two true statements matters, which is most of what good writing is.

We are direct about this because the rest of our site argues that human-made work is the advantage as machine-made content saturates every channel. This service exists to make the human part affordable, not to replace it. Because our Snowball SEO platform automates the search heavy lifting other agencies bill by the hour, more of your budget goes into the judgement rather than the mechanics. It sits inside our wider AI-assisted workflows service.

The difference

Which parts of the work should tools touch?

Six stages, and the answer differs at each. Treating the whole pipeline the same way is how businesses end up either refusing useful help or publishing things they should not.

Where tools help in a content workflow and where they do notFour stages in sequence. Research and gathering is heavily assisted. Outlining is assisted with a person deciding. Drafting is a person's job with a scaffold. Editing, fact-checking and the decision to publish are entirely a person's, because that is where accountability lives.1Research and gatherheavily assisted, genuinely faster2Outlineassisted, a person decides3Drafta person writes, with a scaffold4Edit and publisha person, named and accountableThe line moves left over time as tools improve. It has never yet reached the last box, and we will say so plainly if it does.
The line between the second and third box is where the argument is. Our position is that it sits there today, and we would rather state that clearly than imply more automation than we practise.

Where assistance genuinely helps

  • Gathering and summarising source material, which is slow and mechanical
  • Producing an outline to argue with, which beats an empty page
  • Converting one finished piece into another format
  • Checking: consistency, broken claims, things left out

Where it must not be used

  • Deciding what the piece argues, which is the actual work
  • Anything published under a named person's byline unedited
  • Claims of fact that nobody has verified against a source
  • Original experience, which by definition it does not have

On the wider argument about why human work is the moat, see authentic content in the age of AI.

The cost

What drives the price of setting up a workflow?

How many content types, and how much checking the subject demands. A regulated subject needs more verification than a general one, and that is where the cost sits.

One workflow

Single type

One content type mapped end to end, with the assisted and human stages defined and the checks specified.

A production system

Several types

Articles, social, email and product copy, each with its own workflow, sharing one voice configuration and one checking standard.

Workflow and output

Run by us

The workflow, and us producing to it, which is how most clients actually want this rather than as a system to operate themselves.

We quote by content type and by checking depth after a free consult. See our guide to AI-assisted workflows for the reasoning.

What moves the number, in order

  • How many content types, each of which is its own workflow
  • How much verification the subject requires, which regulation multiplies
  • Whether a voice configuration exists already
  • How much factual source material has to be assembled
  • Whether we produce to the workflow or hand it over
  • Disclosure and record-keeping requirements in your industry

The checking is the expensive half

Assisted drafting is fast, which is the part everyone notices. Verifying that what was drafted is true is not fast, and it is the part that gets quietly dropped when a workflow is designed around speed.

A confident, fluent, wrong paragraph is more damaging than a slow one, because it passes review. That is the specific risk these tools introduce, and a workflow that does not address it directly is not a workflow, it is a shortcut.

So we price checking explicitly rather than folding it into the drafting, which makes it visible and harder to remove without a conversation.

The scope

What you receive

A workflow you could hand to somebody else, and the standards that make it safe.

The workflow, mapped

Every stage, what is assisted, what is not, and who is responsible at each point, written down rather than held as practice.

Checking standards

What must be verified before anything is published, and how, because this is the stage that quietly disappears under deadline.

Tool configuration

Set up against your voice and evidence, and handed over as text you own rather than left inside one account.

A disclosure position

What you tell readers and clients about how work is made, decided deliberately rather than avoided until somebody asks.

A named accountable person

For every piece published, because accountability is what a byline means and what these tools cannot supply.

A review point

When to revisit the workflow, since where the line sits will move and a workflow set once will be wrong within a year.

The return

Where does quality actually get lost?

Not at drafting, which is where everybody looks. At the two stages either side of it, both of which are easy to skip and invisible when skipped.

Where quality is lost in an assisted workflowFour narrowing stages. Deciding what to say, where most quality is won or lost before anything is written. Drafting, which is the visible stage and rarely the problem. Verification, which is skipped under deadline more than any other step. Editing by a person who cares, which is the last chance to catch anything.Deciding what to saymost quality is won or lost hereDraftingthe visible stage, rarely the problemVerificationthe stage skipped under deadlineEditing by someone who caresthe last chance
Attention goes to the second band because it is the one that changed. The first and third are where quality is actually decided, and neither has been made faster by any tool.

Speed at the wrong stage is a liability

If a workflow makes drafting five times faster and leaves deciding and checking unchanged, the result is more content of the same quality, published faster, with the same proportion of errors reaching a larger audience.

That is not an improvement, and it is what most organisations have actually built over the last two years without meaning to.

So we spend the saved drafting time on the first and third stages rather than on producing more. It is a less impressive number to report and it is the only version of this that improves anything.

The process

How a workflow gets built

Three to five weeks, including running real work through it before it is signed off.

Week 1

Map what happens now

The current process as actually practised, which is usually different from the documented one and always more improvised.

Week 2

Draw the line

Which stages are assisted and which are not, decided per content type rather than as one blanket rule.

Week 2

Set the checks

What must be verified, by whom, before publication, written specifically enough to be audited.

Week 3

Configure

Tools set up against your voice and evidence, tested on real briefs from your actual backlog.

Week 4

Run real work through it

Several genuine pieces, end to end, because a workflow that has only been diagrammed has not been tested.

Week 5

Hand over with a review date

Documented, with an agreed point to revisit, because where the line sits will move.

The brief

What to ask before you build an assisted workflow

The questions that matter are about accountability and checking, not about which tools are used.

Who is accountable for each published piece?

A named person. If the workflow cannot answer that, it has automated away the thing that makes publishing safe.

What is verified, and by whom?

Specifically. Verification is the stage that quietly disappears under deadline, and a workflow that does not name it will not have it.

What will you not use these tools for?

A supplier who cannot answer has not thought about it. Original opinion and named bylines should be on the list.

What is our disclosure position?

Decide it deliberately rather than when a client asks. Being asked and having no answer is worse than any answer you might have chosen.

We are happy to answer all four before you commit. On why the human part is the advantage rather than the overhead, see the authenticity argument.

Start here

Ready to build it properly?

Tell us what you publish and how it currently gets made. The free consult is a working session, not a sales call, and you leave with three things whether or not you book us.

  • Your current process mapped as it actually happens, which is usually revealing
  • Where we would and would not use tools in it, with reasons
  • A fixed price, with checking costed separately so it cannot quietly vanish
Get your free workflow map

Good questions

AI-assisted production FAQs

What does AI-assisted production mean?

Using tools for the parts of content work that are structured and low-judgement, and keeping people on the parts that are neither. Research, outlining, draft scaffolds, format conversion and checking are assisted. Deciding what to say, saying it, and standing behind it are not.

Is our content written by AI?

No. A person writes, decides and is accountable, and every published piece carries a named author. Tools shorten the distance to a usable draft; they do not produce what we publish.

Does this contradict your view that human content wins?

No, and the distinction is worth stating precisely. Our position is that human judgement and real experience are the advantage. Clearing structured, low-judgement work faster is what creates the time to do the work that differentiates you.

How much does an assisted workflow cost?

We quote by content type and checking depth. What moves it is how many content types are in scope, how much verification the subject requires, whether a voice configuration exists, and whether we produce to the workflow or hand it over.

Where does quality actually get lost?

At deciding what to say, and at verification. Not at drafting, which is where everybody looks. A workflow that speeds up drafting and leaves those two unchanged produces more content of the same quality with the same error rate reaching more people.

Should we disclose that tools were used?

Decide it deliberately rather than when somebody asks. Our own position is that the named author is accountable for the work regardless of the tools involved, and we are willing to say exactly how anything was made.

What about factual errors?

This is the specific risk these tools introduce: fluent, confident and wrong passes review more easily than slow and hesitant. Verification has to be a named stage with a named person, and it is the first thing to disappear under deadline.

Will this workflow still be right next year?

Probably not, and it is written with a review date for that reason. Where the sensible line sits will move, and a workflow set once and never revisited will be wrong in one direction or the other within a year.